SlideShare a Scribd company logo

Multiple Access Techniques for 5G

Sitha Sok, profile picture
Sitha Sok

This document is the table of contents for Volume 14, Number 4 of the journal ZTE Communications. The special topic of this issue is Multiple Access Techniques for 5G. It includes an editorial by Yuan Jinhong, Xiang Jiying, Ding Zhiguo, and Yuan Zhifeng introducing the topic. There are then 6 research papers on various non-orthogonal multiple access schemes and other multiple access technologies that could be used in 5G wireless networks. The issue also includes a review paper and information about new members of the editorial board.

Technology
1 of 72
Downloaded 199 times
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
ISSN 1673-5188 CN 34-1294/ TN CODEN ZCTOAK ZTECOMMUNICATIONSVOLUME14NUMBER4OCTOBER2016 tech.zte.com.cn ZTECOMMUNICATIONS October 2016, Vol. 14 No. 4An International ICT R&D Journal Sponsored by ZTE Corporation SPECIAL TOPIC: Multiple Access Techniques for 5G
ZTE Communications Editorial Board Members (in Alphabetical Order): Chairman ZHAO Houlin: International Telecommunication Union (Switzerland) Vice Chairmen SHI Lirong: ZTE Corporation (China) XU Chengzhong: Wayne State University (USA) CAO Jiannong Hong Kong Polytechnic University (Hong Kong, China) CHEN Chang Wen University at Buffalo, The State University of New York (USA) CHEN Jie ZTE Corporation (China) CHEN Shigang University of Florida (USA) CHEN Yan Northwestern University (USA) Connie Chang􀆼Hasnain University of California, Berkeley (USA) CUI Shuguang University of California, Davis (USA) DONG Yingfei University of Hawaii (USA) GAO Wen Peking University (China) HWANG Jenq􀆼Neng University of Washington (USA) LI Guifang University of Central Florida (USA) LUO Fa􀆼Long Element CXI (USA) MA Jianhua Hosei University (Japan) PAN Yi Georgia State University (USA) REN Fuji The University of Tokushima (Japan) SHI Lirong ZTE Corporation (China) SONG Wenzhan University of Georgia (USA) SUN Huifang Mitsubishi Electric Research Laboratories (USA) SUN Zhili University of Surrey (UK) Victor C. M. Leung The University of British Columbia (Canada) WANG Xiaodong Columbia University (USA) WANG Zhengdao Iowa State University (USA) WU Keli The Chinese University of Hong Kong (Hong Kong, China) XU Chengzhong Wayne State University (USA) YANG Kun University of Essex (UK) YUAN Jinhong University of New South Wales (Australia) ZENG Wenjun Microsoft Research Asia (USA) ZHANG Chengqi University of Technology Sydney (Australia) ZHANG Honggang Zhejiang University (China) ZHANG Yueping Nanyang Technological University (Singapore) ZHAO Houlin International Telecommunication Union (Switzerland) ZHOU Wanlei Deakin University (Australia) ZHUANG Weihua University of Waterloo (Canada)
CONTENTSCONTENTS Submission of a manuscript implies that the submitted work has not been published before (except as part of a thesis or lecture note or report or in the form of an abstract); that it is not under consideration for publication elsewhere; that its publication has been approved by all co- authors as well as by the authorities at the institute where the work has been carried out; that, if and when the manuscript is accepted for publication, the authors hand over the transferable copyrights of the accepted manuscript to ZTE Communications; and that the manuscript or parts thereof will not be published elsewhere in any language without the consent of the copyright holder. Copyrights include, without spatial or timely limitation, the mechanical, electronic and visual reproduction and distribution; electronic storage and retrieval; and all other forms of electronic publication or any other types of publication including all subsidiary rights. Responsibility for content rests on authors of signed articles and not on the editorial board of ZTE Communications or its sponsors. All rights reserved. Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng 01 Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei 11 Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong 03 Special Topic: Multiple Access Techniques for 5G A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo 17 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I 26 ISSN 1673-5188 CN 34-1294/ TN CODEN ZCTOAK tech.zte.com.cn ZTECOMMUNICATIONS October 2016, Vol. 14 No. 4An International ICT R&D Journal Sponsored by ZTE Corporation SPECIAL TOPIC: Multiple Access Techniques for 5G
ZTE COMMUNICATIONS Vol. 14 No. 4 (Issue 53) Quarterly First English Issue Published in 2003 Supervised by: Anhui Science and Technology Department Sponsored by: Anhui Science and Technology Information Research Institute and ZTE Corporation Staff Members: Editor-in-Chief: CHEN Jie Executive Associate Editor-in-Chief: HUANG Xinming Editor-in-Charge: ZHU Li Editors: XU Ye, LU Dan, ZHAO Lu Producer: YU Gang Circulation Executive: WANG Pingping Assistant: WANG Kun Editorial Correspondence: Add: 12F Kaixuan Building, 329 Jinzhai Road, Hefei 230061, P. R. China Tel: +86-551-65533356 Fax: +86-551-65850139 Email: magazine@zte.com.cn Published and Circulated (Home and Abroad) by: Editorial Office of ZTE Communications Printed by: Hefei Tiancai Color Printing Company Publication Date: October 25, 2016 Publication Licenses: Advertising License: 皖合工商广字0058号 Annual Subscription: RMB 80 ISSN 1673-5188 CN 34-1294/ TN CONTENTSCONTENTS Roundup New Members of ZTE Communications Editorial Board 57 Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen 58 Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu 50 Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu 35 Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson 42
Multiple Access Techniques forMultiple Access Techniques for 55GG ▶ YUAN Jinhong YUAN Jinhong received his BE and PhD degrees in electronics engineering from Beijing Institute of Technology in 1991 and 1997. From 1997 to 1999, he was a research fellow at the School of Electrical Engineering, University of Sydney, Australia. In 2000, he joined the School of Electrical Engineering and Tele⁃ communications, University of New South Wales, Australia, and is currently a professor of telecommunications there. Dr. Yuan has authored two books, three book chapters, and more than 200 papers for telecom journals and conferences. He has also au⁃ thored 40 industry reports. He is a co⁃inventor of one patent on MIMO systems and two patents on low⁃density parity⁃check (LDPC) codes. He has co⁃ authored three papers that won Best Paper Awards or Best Poster Awards. Dr. Yuan served as the NSW Chair of the joint Communications/Signal Processions/Ocean Engi⁃ neering Chapter of IEEE during 2011-2014. He is an IEEE fellow and an associate edi⁃ tor for IEEE Transactions on Communications. His research interests include error⁃con⁃ trol coding and information theory, communication theory, and wireless communications. ver the past few decades, wireless communications have advanced tremendously and have become an indispensable part of our lives. Wireless networks have become more and more pervasive in order to guarantee global digital connectivity. Wireless devices have quickly evolved into multimedia smartphones running applications that demand high⁃speed and high⁃quality data connections. The upcoming fifth generation (5G) mobile cellular networks are required to provide significant increase in network throughput, cell⁃edge data rates, massive connectivity, superior spectrum efficiency, high energy efficiency and low latency, compared with the currently deployed long⁃term evolution (LTE) and LTE⁃advanced networks. To meet these demanding challenges of 5G networks, innovative technologies on radio air⁃interface and radio access network (RAN) are of great importance in PHY designs. Recently non⁃orthogonal multiple access (NOMA) has at⁃ tracted increasing research interests from both academic and industrial fields as a potential radio access tech⁃ nique. A few examples include multiuser shared access (MUSA), sparse code multiple access (SCMA), resource spread multiple access (RSMA) and pattern division multiple access (PDMA) proposed by ZTE, Huawei, Qual⁃ comm, DTmobile, etc. In the mean time, multicarrier (MC) technologies that divide frequency spectrum into many narrow subchannels, such as filter bank multicarrier (FBMC) and generalized frequency division multiplexing (GFDM), become attractive and new concepts for dynamic access spectrum management and cognitive radio appli⁃ cations. With these new developments, this special issue is dedicated to multiple access transmission technologies and O October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 01 ▶ XIANG Jiying XIANG Jiying, PhD, is the Chief Scientist of ZTE Corporation. His research is focused on 3G, 4G, 5G, and multi⁃mode wireless infrastructure technologies. He led the development of the first commercial SDR base station in the industry in 2007. He pro⁃ posed the first solution that support COMP on non⁃ideal back⁃ haul (also called Cloud Radio) in 2012. In 2014, he proposed the “pre⁃5G”conception, which includes massive MIMO, D⁃MIMO, MUSA, and UDN. Pre⁃5G allows 5G⁃like user experience on lega⁃ cy 4G handsets. Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng Special Topic ▶ DING Zhiguo DING Zhiguo received his BEng in electrical engineering from Beijing University of Posts and Telecommunications, China in 2000, and the PhD degree in electrical engineering from Imperial College London, UK in 2005. From Jul. 2005 to Aug. 2014, he worked in Queen’s University Belfast, Imperial College and New⁃ castle University, UK. Since Sept. 2014, he has been with Lan⁃ caster University, UK as a chair professor. From Oct. 2012 to Sept. 2017, he has also been an academic visitor in Princeton University, USA. His research interests are 5G networks, game theory, cooperative and energy harvesting networks, and statisti⁃ cal signal processing. He is serving as an editor for IEEE Transactions on Communica⁃ tions, IEEE Transactions on Vehicular Technology, IEEE Wireless Communication Let⁃ ters, IEEE Communication Letters, and Journal of Wireless Communications and Mo⁃ bile Computing. He received the best paper award in IET Comm. Conf. on Wireless, Mo⁃ bile and Computing, 2009, IEEE Communication Letter Exemplary Reviewer 2012, and the EU Marie Curie Fellowship 2012-2014. ▶ YUAN Zhifeng YUAN Zhifeng received his MS degree in signal and information processing from Nanjing University of Post and Telecommunica⁃ tions, China in 2005. He has been working at the Wireless Tech⁃ nology Advance Research Department, ZTE Corporation since 2006 and as the leader of the New Multi⁃Access (NMA) for 5G Wireless System Team since 2012. His research interests include wireless communications, MIMO systems, information theory, multiple access, error control coding, adaptive algorithm, and high⁃speed VLSI design.
related for 5G cellular mobile communications. The main focus is on the cutting⁃edge research, review and application on non⁃ orthogonal multiple access and related signal processing and coding methods for the air ⁃ interface of 5G enhanced mobile broadband (eMBB), mMTC, and ultra reliable and low latency communication (URLLC). Papers for this issue were invited, and after peer review, six were selected for publication. The se⁃ lected papers cover reviews of various uplink and downlink NOMA schemes, novel designs for MIMO⁃FBMC systems, re⁃ view and new designs on multiple access technologies for cellu⁃ lar M2M communications and IoT applications. This issue is intended to be a timely, high⁃quality forum for scientists and engineers. In“Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems”by Taheri, Ghoraishi, XIAO, CAO and GAO, the authors discuss a candidate waveform design for fu⁃ ture wireless communications based on MIMO⁃FBMC and tack⁃ le the challenging problem of channel estimation facing the waveform design. Specifically, they propose a novel channel es⁃ timation method which employs intrinsic interference cancella⁃ tion at the transmitter side. Their research results demonstrate that the proposed novel technique incurs less pilot ⁃ overhead compared to the well⁃known intrinsic approximation methods (IAM). In addition, it also has a better PAPR, BER and MSE performance. In“Non ⁃ Orthogonal Multiple Access Schemes for 5G,” YAN, YUAN, LI, and YUAN provide a comprehensive review of six potential multiple access schemes for 5G, including MU⁃ SA, RSMA, SCMA, PDMA, interleaver ⁃ division multiple ac⁃ cess (IDMA) and NOMA. The principles, advantages and dis⁃ advantages of these multiple access schemes are discussed. More importantly, this review offers a comprehensive compari⁃ son of these solutions from the perspective of user overload, re⁃ ceiver type, receiver complexity, performance and grant ⁃ free transmission. In“A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks”by WEI, YUAN, Ng, Elkashlan and DING, the authors use a simple downlink model with two users served by a single⁃carrier to illustrate the basic principles of NOMA and its performance. The related questions and designs for a more general model with an arbi⁃ trary number of users and multiple carriers are discussed. In addition, an overview of existing works on performance analy⁃ sis, resource allocation, and multiple ⁃ input multiple ⁃ output NOMA are summarized and discussed. The key features of NO⁃ MA and its potential research challenges in future networks are raised. In“Unified Framework Towards Flexible Multiple Access Schemes for 5G”, SUN, WANG, HAN and I provide a compre⁃ hensive overview for the multiple access schemes proposed for 5G networks. The authors distinguish three types of multiple access techniques in power, code and interleaver based solu⁃ tions, respectively. The key features of these multiple access techniques are highlighted, and the authors also provide com⁃ parison among these multiple access techniques. Another im⁃ portant contribution of this paper is that a unified framework of the aforementioned multiple access techniques is provided. In“Multiple Access Rateless Network Coding for Machine⁃ to ⁃ Machine Communications” by JIAO, Abbas, LI and ZHANG, the authors propose a novel multiple access rateless network coding scheme for machine⁃to⁃machine (M2M) com⁃ munications. The scheme is capable of increasing transmission efficiency by reducing occupied time slots yet with high decod⁃ ing success rates. In addition, in contrast to existing state⁃of⁃ the⁃art coding schemes, the novel rateless network coding is able to dynamically recode, making it suitable for M2M multi⁃ cast networks with heterogeneous erasure features. In“Multiple Access Technologies for Cellular M2M Commu⁃ nications”, Shirvanimoghaddam and Johnson provide a com⁃ prehensive survey of the multiple access techniques for ma⁃ chine ⁃ to ⁃ machine (M2M) communications in future wireless cellular networks. In particular, the overview highlights the multiple access strategies and explains their limitations when used for M2M communications. The throughput efficiency of different multiple access techniques when used in coordinated and uncoordinated scenarios are illustrated. The authors dem⁃ onstrate that in uncoordinated scenarios, NOMA can support a larger number of devices compared to orthogonal multiple ac⁃ cess techniques. We thank all authors for their valuable contributions and all reviewers for their timely and constructive comments on the submitted papers. We hope the content of this issue is informa⁃ tive and helpful to all readers. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS02 Special Topic Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng
Evaluation of Preamble Based Channel Estimation forEvaluation of Preamble Based Channel Estimation for MIMO⁃FBMC SystemsMIMO⁃FBMC Systems Sohail Taheri1 , Mir Ghoraishi1 , XIAO Pei1 , CAO Aijun2 , and GAO Yonghong2 (1. 5G Innovation Centre, Institute for Communication Systems (ICS), University of Surrey, Guildford, Surrey GU2 7XH, United Kingdom; 2. ZTE Wistron Telecom AB, Kista, Stockholm 164 51, Sweden) Abstract Filter⁃bank multicarrier (FBMC) with offset quadrature amplitude modulation (OQAM) is a candidate waveform for future wireless communications due to its advantages over orthogonal frequency division multiplexing (OFDM) systems. However, because of or⁃ thogonality in real field and the presence of imaginary intrinsic interference, channel estimation in FBMC is not as straightforward as OFDM systems especially in multiple antenna scenarios. In this paper, we propose a channel estimation method which employs intrinsic interference cancellation at the transmitter side. The simulation results show that this method has less pilot overhead, less peak to average power ratio (PAPR), better bit error rate (BER), and better mean square error (MSE) performance compared to the well⁃known intrinsic approximation methods (IAM). channel estimation; filter⁃bank multicarrier (FBMC); multiple⁃input multiple⁃output (MIMO); offset quadrature amplitude modula⁃ tion (OQAM); wireless communication Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 001 http://www.cnki.net/kcms/detail/34.1294.TN.20161014.0955.002.html, published online October 14, 2016 Special Topic This work is supported by ZTE Industry⁃Academia⁃Research Cooperation Funds under Grant No. Surrey⁃Ref⁃9953. 1 Introduction rthogonal frequency division multiplexing (OFDM) has been widely used in communication systems in the last decade. This is because of its immunity to multipath fading and simplicity of channel estimation and data recovery with a low complexity single⁃tap equalization, and also suitability for multiple⁃input multiple⁃output (MIMO) systems [1]. However, it suffers from disadvantages such as sensitivity to carrier frequency offset (CFO), significant out⁃of⁃band radiation, and cyclic prefix over⁃ head. In the presence of CFO, there is loss of orthogonality be⁃ tween subcarriers leading to inter carrier interference (ICI). Moreover, to efficiently use the available spectrum, a waveform with very low spectral leakage is needed. Because of the OFDM shortcomings, filter⁃bank multicarrier (FBMC) modulation combined with offset quadrature ampli⁃ tude modulation (OQAM) has drawn attention in the last de⁃ cade [2], [3]. Regardless of the higher complexity compared to OFDM, FBMC (known as OFDM/OQAM and FBMC/OQAM in the literature) provides significantly reduced out⁃of⁃band emis⁃ sions, robustness against CFO [4], and under certain condi⁃ tions, better spectral efficiency as there is no need to use cy⁃ clic prefix (CP) [5]. These advantages come from well localized prototype filters in time and frequency domain for pulse shap⁃ ing. Accordingly, FBMC can be a promising alternative to con⁃ ventional radio access techniques to improve wireless access capacity. On the other hand, as orthogonality in FBMC systems only holds in the real field, received symbols are contaminated with an imaginary intrinsic interference term coming from the neigh⁃ bouring real symbols. The interference becomes a source of problem in channel estimation and equalization processes, es⁃ pecially in MIMO systems. The pilot symbols used for channel estimation should be protected from interference as the receiv⁃ er has no knowledge about their neighbours to estimate the amount of interference. These protections cause overheads when designing a transmission frame. In a preamble⁃based ap⁃ proach, the preamble should be protected from the subsequent data transmission and the previous frame by inserting null sym⁃ bols, which causes longer preamble and thus more overhead compared to OFDM. This is also true for scattered pilots where the neighbouring data symbols contribute to the interference on the pilots [6]. In this scenario, typically one or two time⁃fre⁃ quency points adjacent to the pilots are used to cancel the in⁃ terference on the pilots [7]-[10]. Interference Approximation Method (IAM) for preamble ⁃ O October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 03
Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS04 based channel estimation in single⁃input, single⁃output (SISO) systems was first introduced in [11]. The preamble was named IAM⁃R in the literature, where R denotes real⁃valued pilots. Alternatively, IAM⁃I and IAM⁃C were introduced in [12], [13], where I and C stand for imaginary and complex pilots. Those preamble based channel estimation schemes were extended to FBMC⁃MIMO systems in [14]. In IAM⁃I and IAM⁃C, pilots on each subcarrier interfere with their adjacent subcarriers in a constructive way. That is, these methods use the intrinsic inter⁃ ference to enhance amplitude of the pilots. As a result, better performance of channel estimation is achieved. Despite good performance, IAM methods suffer from increased pilot over⁃ head, i.e., a number of zero symbols are required to protect the pilot symbols from the interference of their adjacent symbols. While the number of pilot symbols is equal to the number of antennas, the total number of symbols in the preamble will be more than twice the number of transmit antennas. This paper proposes a channel estimation method with re⁃ duced preamble overhead compared to the IAM family. The idea was first introduced in [15] for MIMO⁃OFDM. Applying this method to MIMO⁃FBMC with spatial multiplexing needs further consideration to cancel intrinsic interference. By using basic idea of zero forcing from single antenna, this method has modest computation complexity, while it can outperform IAM methods in terms of peak to average power ratio (PAPR), bit er⁃ ror rate (BER), and mean square error (MSE) under perfect syn⁃ chronization conditions and in presence of carrier frequency offset. The rest of this paper is organized as follows: Section 2 re⁃ views the MIMO⁃FBMC systems, the effect of intrinsic interfer⁃ ence, and the conventional channel estimation methods. In Sec⁃ tion 3, the new method for channel estimation is proposed and Section 4 shows the results and comparisons with IAM meth⁃ ods. Finally, conclusions are drawn in Section 5. 2 MIMO⁃FBMC System 2.1 System Model FBMC systems are implemented by a prototype filter g( )t and synthesis and analysis filter⁃banks in transmitter and re⁃ ceiver side respectively. The real and imaginary parts of com⁃ plex symbols are separated in two different branches where they are modulated in FBMC modulators as real symbols. Therefore, at a specific time, each subcarrier in this system car⁃ ries a real⁃valued symbol. Denoting T0 as symbol duration and F0 as subcarrier spacing in OFDM systems, duration and sub⁃ carrier spacing in FBMC are either τ0 = T0 2 , ν0 = F0 or τ0 = T0 , ν0 = F0 2 [16]. For the system model in this paper, the former approach is adopted. That is, subcarrier spacing re⁃ mains the same as OFDM, while symbol duration is reduced by half. Assuming a multiple antenna scenario with P transmit anten⁃ nas, Q receive antennas, and M subcarriers, the baseband sig⁃ nal to be transmitted over the p th branch in general form is expressed as s ( )p ( )t = ∑n = -∞ +∞ ∑m = 0 M - 1 a ( )p m,n gm,n( )t , (1) where a ( )p m,n is the real⁃valued symbol, and gm,n( )t is the shift⁃ ed version of the prototype filter on the m th subcarrier and at n th symbol duration: gm,n( )t = jm + n e j2πmν0t g( )t - nτ0 . (2) The prototype filter g( )t is designed to keep its shifted ver⁃ sions are orthogonal only in the real field [17], i.e., R æ è ç ö ø ÷∫gm,n( )t g* m0,n0 ( )t dt = δm,m0 δn,n0 , (3) where R( ). denotes the real⁃part of a complex number. As a consequence, the outputs of the analysis filter⁃bank have a so⁃ called intrinsic interference term which is pure imaginary. The demodulated signal on the q th receive antenna at a particular subcarrier and symbol point ( )m0,n0 is given by y ( )q m0,n0 =∑p = 1 P h q,p m0,n0 a ( )p m0,n0 + jI ( )q m0,n0 + η ( )q m0,n0 , (4) where h q,p m0,n0 is channel frequency response at ( )m0,n0 be⁃ tween qth receive and pth transmit antenna, η ( )q m0,n0 is the noise component at qth receive antenna, and the interference term I ( )q m0,n0 is formed as jI ( )q m0,n0 =∑p = 1 P ∑ ( )m,n ≠ ( )m0,n0 h q,p m,na ( )p m,n g m0,n0 m,n . (5) In (5), g m0,n0 m,n is expressed as g m0,n0 m,n = ∫gm,n( )t g* m0,n0 ( )t dt . (6) Having the prototype filter g( )t well localized in time and frequency, it can be assumed that the intrinsic interference is mostly due to the first ⁃ order neighbouring points. That is, ( )m,n in (5) can take the values of Ω* as follows [6]: Ω* ={ }( )m0,n0 ± 1 ,( )m0 ± 1,n0 ,( )m0 ± 1,n0 ± 1 , (7) which covers the ( )m0,n0 point first⁃order neighbours. By as⁃ suming constant channel frequency response over ( )m0,n0 and Ω* , we can simplify (5) as jI ( )q m0,n0 =∑p = 1 P h p,q m0,n0 ∑ ( )m,n ∈ Ω* a ( )p m,n g m0,n0 m,n . (8)
Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 05 Consequently, (4) can be written as y ( )q m0,n0 =∑p = 1 P h p,q m0,n0 æ è ç ç ç çç ç ö ø ÷ ÷ ÷ ÷÷ ÷       a ( )p m0,n0 + ju ( )p m0,n0 c ( )p m0,n0 + η ( )p m0,n0 , (9) where ju ( )p m0,n0 = ∑ ( )m,n ∈ Ω* a ( )p m,n g m0,n0 m,n . (10) Table 1 shows the number of g m0,n0 m,n coefficients on the first⁃ order neighbours of the point ( )m0,n0 . The weights of interfer⁃ ence, β , γ , and δ, depend on the prototype filter and have been derived in [18]. In this work, the isotropic orthogonal transform algorithm (IOTA) [19] filter is employed. It exploits the symmetrical property of Gaussian function in time and fre⁃ quency. Therefore, the amount of interference out of first⁃order neighbouring points is negligible. The weights of interference for this filter are β = 0.2486 , γ = 0.5755 , and δ = 0.1898 (Table 1) . The MIMO⁃FBMC signal model can be represented as æ è ç ç çç ö ø ÷ ÷ ÷÷ y (1) m0,n0 ⋮ y (Q) m0,n0 = æ è ç ç çç ö ø ÷ ÷ ÷÷ h 1,1 m0,n0 ⋯ h 1,P m0,n0 ⋮ ⋱ ⋮ h Q,1 m0,n0 ⋯ h Q,P m0,n0 æ è ç ç çç ö ø ÷ ÷ ÷÷ c (1) m0,n0 ⋮ c (Q) m0,n0 + æ è ç ç çç ö ø ÷ ÷ ÷÷ η (1) m0,n0 ⋮ η (Q) m0,n0 (11) where c ( )p m0,n0 is defined in (9). To retrieve the transmitted sy⁃ mbols from the system above, it is necessary to have an evalua⁃ tion of the channel coefficients, which are used to detect the linearly combined demodulated complex symbols c ( )p m0,n0 at each receiver branch using zero forcing (ZF), minimum mean square error (MMSE), or maximum likelihood (ML). In c ( )p m0,n0 , the imag⁃ inary parts are intrinsic interference terms. By taking R{}. op⁃ eration, the transmitted symbols a ( )p m0,n0 = R{ }c ( )p m0,n0 are recovered. 2.2 Channel Estimation To obtain the channel information over one frame duration on each receive antenna, we need to know the transmitted pilot symbols. The number of these pilot symbols should be equal to P to form a linear equation system with the least square estima⁃ tion method. For simplicity, let us consider a 2⁃by⁃2 antenna scenario. By allocating two pilot symbols at times n = n0 and n = n1 on each antenna, the equation set of the system on su⁃ bcarrier m is given by æ è ç ç ö ø ÷ ÷ y ( )1 m,n0 y ( )1 m,n1 y ( )2 m,n0 y ( )2 m,n1 = æ è ç ç ö ø ÷ ÷ h 1,1 m,n0 h 1,2 m,n0 h 2,1 m,n0 h 2,2 m,n0 æ è ç ç ö ø ÷ ÷ x ( )1 m,n0 x ( )1 m,n1 x ( )2 m,n0 x ( )2 m,n1 + æ è ç ç ö ø ÷ ÷ η ( )1 m,n0 η ( )1 m,n1 η ( )2 m,n0 η ( )2 m,n1 . (12) In (12), x ( )p m,n are pilot symbols. We have assumed that there is no significant variations in the channel between time slots n0 and n1 . Hence, we can drop the time subscript and express (12) as Ym = HmXm + ηm. (13) Thus, channel coefficients can be calculated by the least square estimation method: Ĥ m = Ym( )XH mXm -1 XH m = Hm + ηm( )XH mXm -1 XH m , (14) or in a special case with the equal number of transmit and re⁃ ceive antenna: Ĥ m = YmX-1 m = Hm + ηmX-1 m . (15) The preamble in the IAM methods is composed of 2P + 1 symbols. That is, the length of the preamble grows linearly with P. The symbols with even time indices are pilots, while other symbols are all zeros to protect pilots from intrinsic inter⁃ ference. Based on the values of pilot symbols, i.e. real, imagi⁃ nary, or complex valued pilots, IAM ⁃ R, IAM ⁃ I and IAM ⁃ C were proposed. In these approaches, the channel coefficients can be obtained using (12). For P=2, pilot symbols in (12) are set as x ( )1 m,n0 = x ( )1 m,n1 = x ( )2 m,n0 = -x ( )2 m,n1 = xm . Hence, they form a system based on (12) as Ym = xmHm( )1 1 1 -1 + ηm = xmHmA2 + ηm, (16) where A2 = A-1 2 is an orthogonal matrix if omitting the co⁃ nstant coefficient of the inverse [14]. Finally, the channel coef⁃ ficients are obtained as follows: Ĥ m = 1 xm YmA2 = Hm + 1 xm ηmA2. (17) The length of the preamble in this method is 2P+1=5 with just two pilot symbols. As a result, this approach suffers from significant pilot overhead which reduces the spectral efficien⁃ cy. Furthermore, the periodic nature of the pilots in these pre⁃ ambles results in high PAPR at the output of the synthesis filter⁃ ▼Table 1. Weights of interference on the first⁃order neighbours m0 - 1 m0 m0 + 1 n0 - 1 ( )-1 m0 δ -( )-1 m0 γ ( )-1 m0 δ n0 -β 1 β n0 + 1 ( )-1 m0 δ ( )-1 m0 γ ( )-1 m0 δ
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS06 bank [14]. 3 Proposed Method In order to reduce the preamble overhead and accordingly increase the spectral efficiency, a novel channel estimation ap⁃ proach with modest computation complexity is proposed. Since there is no need to have an estimation of the channel on each subcarrier, we can reduce the number of pilot symbols to one. In this way, each subcarrier is allocated to only one branch to transmit pilot. That is, while a branch is transmitting pilot on a subcarrier, the other branches remain silent. Therefore, the channel parameters between the receive branch and the pilot transmitting branch on that specific subcarrier can be ob⁃ tained. This method enables the increase of transmit branches with a constant length of the preamble. To elaborate the system more precisely, we assume a 2x2 MIMO system where preambles for branches 1 and 2 are shown in Fig. 1. It can be seen that the first and third symbols are all zero to protect the preamble from intrinsic interference from data section and previous frame. In the middle symbol for branch 1, complex pilots are placed on odd subcarriers, while the other subcarriers carry zeros. On branch 2, orthogonal pi⁃ lots to branch 1 are sent, i.e., even subcarriers carry complex pilots and the rest are zero valued. On a particular subcarrier m = m0 , the system equations is written as follows: æ è ç ç ö ø ÷ ÷ y ( )1 m0 y ( )2 m0 = æ è ç ç ö ø ÷ ÷ h 1,1 m0 h 1,2 m0 h 2,1 m0 h 2,2 m0 æ è ç ç ö ø ÷ ÷ x ( )1 m0 x ( )2 m0 + æ è ç ç ö ø ÷ ÷ η ( )1 m0 η ( )2 m0 . (18) On odd subcarriers m0 = 2k + 1, we have x ( )1 m0 = Xm0 , while x ( )2 m0 = 0 . Then, the channel coefficients h 1,1 m0 and h 2,1 m0 are o⁃ btained as h 1,1 m0 = y ( )1 m0 Xm0 |​ x ( )2 m0 = 0 h 2,1 m0 = y ( )2 m0 Xm0 |​ x ( )2 m0 = 0. (19) Likewise, on even subcarriers the channel coefficients of h 1,2 m0 and h 2,2 m0 are given by h 1,2 m0 = y ( )1 m0 Xm0 |​ x ( )1 m0 = 0 h 2,2 m0 = y ( )2 m0 Xm0 |​ x ( )1 m0 = 0. (20) Hence, we have calculated the channel parameters between each pair of antennas on alternative subcarriers. Channel Coef⁃ ficients on the rest of subcarriers can be obtained by interpola⁃ tion. Due to short distance between pilots in this system, linear interpolation provides enough accuracy with the advantage of low complexity. The technique works perfectly for MIMO ⁃ OFDM systems [15]. When applying this method to MIMO⁃FBMC, intrinsic in⁃ terference degrades the channel estimation performance, i.e., transmitted pilots from one branch interfere with the received pilots on other branch. Consequently, the conditions in (19) and (20) no longer hold. To tackle this problem, we propose a precoding approach in which the interference is calculated at the transmitter side. Then, the zero points in pilot symbols are replaced by Im,n , so that there are no interference on the corre⁃ sponding points at the receiver side. That is, the pilots are re⁃ ceived without any interference from other branches. Fig. 2 shows the precoded pilots. The value of cancelling in⁃ terference on subcarrier m is calculated by using (10) as Im,n = - ∑ ( )m' ,n' ∈ Ω* a ( )p m,n g m' ,n' m,n . (21) Moreover, the adjacent points of the pilot Xm are filled with pre⁃calculated values to maximize the received signal energy, thereby to enhance the estimation accuracy [18]. Defining Xm = XR m + jXI m , These values would be X' m = - jXI m X″ m = - XR m. (22) Consequently, the amplitude of the real and imaginary parts Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong ▲Figure 1. The basic preamble for two antennas. ▲Figure 2. The preambles for two antennas after interference cancellation of the first and third time symbols that helps the pilots become stronger. 0 0 0 0 0 0 Xm - 3 0 Xm - 1 0 Xm + 1 0 Branch 1 0 0 0 0 0 0 0 0 0 0 0 0 0 Xm - 2 0 Xm 0 X_(m + 2) Branch 2 0 0 0 0 0 0 X' m - 3 0 X' m - 1 0 X' m + 1 0 Xm - 3 Im - 2,1 Xm - 1 Im,1 Xm + 1 Im + 2,1 Branch 1 X″ m - 3 0 X″ m - 1 0 X″ m + 1 0 0 X' m - 2 0 X' m 0 X' m + 2 Im - 3,1 Xm - 2 Im - 1,1 Xm Im + 1,1 Xm + 2 Branch 2 0 X″ m - 2 0 X″ m 0 X″ m + 2
of the received pilots becomes ||X̂ R m = ||XR m + γ ||X' m + γ ||XI m ||X̂ I m = ||XI m + γ ||X″ m + γ ||XR m , (23) where γ is the interference weight shown in Table 1. The com⁃ plete design of the preambles is displayed in Fig. 2. The pilots can take arbitrary values. In this work, the maximum ampli⁃ tude of the used QAM modulation is used so that XR m = XI m . In order to avoid high PAPR, the sign of the pilots should be changed alternatively after a number of repetitions. The final value of the received pilots in (23) with XR m = XI m is X̂ m = ( )1 + 2γ Xm. (24) The extension to P⁃branch MIMO system is straightforward. In this case, one subcarrier of every P subcarriers carries a pi⁃ lot (non⁃zero), while each branch’s pilot symbol is orthogonal to other branches. The more transmit branches, the more dis⁃ tance between pilot subcarriers. Consequently, for larger num⁃ ber of branches, the quality of channel estimation degrades. 4 Simulation Results In this section, different preamble⁃based channel estimators for a 2x2 MIMO⁃FBMC system are simulated and compared. The simulations are performed using 7⁃tap EPA⁃5Hz and 9⁃tap ETU⁃70Hz channel models with low spatial correlations. Per⁃ fect synchronization is assumed for BER and MSE comparison, i.e., there is no timing or frequency offset errors. In order to de⁃ tect symbols, MMSE equalizer is used. Table 2 summarizes the simulation parameters. The results are compared with IAM⁃R and IAM⁃C methods introduced in [14]. For fair comparison, the transmission power is kept equal for all methods. In this system, Eb N0 is defined by Eb N0 = Q SNR α × log2( )M , (25) where M = 16 is the modulation order, SNR is signal⁃to⁃noise ratio, and α = Ns - Np Ns with the frame length Ns = 14 and the preamble length Np . The length of preamble Np in the pr⁃ oposed method is three symbols resulting in 40% overhead re⁃ duction compared to IAMs. As a result, a performance gain is expected due to shorter preamble. The extra symbols generated by the synthesis filter⁃banks can be dropped before transmis⁃ sion, but one of them with the most power should be kept to avoid filtering errors after demodulation, i.e., Ns + 1 symbols are transmitted. To consider this extra symbol, α can be changed to α = Ns - Np Ns + 1. 4.1 PAPR Comparison Fig. 3 shows the comparison between the proposed method and IAMs in terms of PAPR. The plots show the squared mag⁃ nitude of the preambles at the output of the synthesis filter ⁃ bank on branch 1. Evidently, from the point of practical imple⁃ mentations, the proposed method is preferable. Whereas in the others, the signal level should be kept very low to avoid A/D saturations. The PAPR levels for the pilot symbols are com⁃ pared in Table 3 for the three methods. 4.2 Channel Estimation Performance Comparison Fig. 4 shows the MSE comparison of the channel estimation methods. To calculate MSE, the channel tap on the second symbol in frame is considered as reference and it is assumed constant during the symbol duration. Then, the MSE is calcu⁃ Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 07 ▼Table 2. Simulation parameters EPA: Extended Pedestrian A model ETU: Extended Typical Urban model FFT: fast Fourier transform QAM: quadrature amplitude modulation Modulation type FFT size Used subcarriers Sampling frequency Symbols per frame Channel M⁃QAM, M =16 256 144 3.84 MHz 14 EPA 5 Hz, ETU 70 Hz ▼Table 3. PAPR comparison for the three methods IAM⁃C: Interference Approximation Method⁃complex pilots IAM⁃R: Interference Approximation Method⁃real valued pilots PAPR: peak to average power ratio PAPR IAM⁃C 17.5 IAM⁃R 9.3 Proposed 7.2 ▲Figure 3. Squared magnitude of the preambles on output of the branch 1. 150010005000 15 10 5 0 The proposed preamble 150010005000 15 10 5 0 IAM⁃R preamble 150010005000 15 10 5 0 IAM⁃C preamble Time (Samples) Transmitsignalcomparison IAM⁃C: Interference Approximation Method⁃complex pilots IAM⁃R: Interference Approximation Method⁃real valued pilots
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS08 lated using the estimated channel Ĥ as Eæ è ö ø ( )H - Ĥ H ( )H - Ĥ . It can be seen that the proposed preamble outperforms IAM⁃R and has approximately the same performance as IAM⁃C in both channel models. In the EPA⁃5Hz scenario, the proposed meth⁃ od gradually reaches an error floor. This is due to domination of errors from ISI and interference cancellation residual. How⁃ ever, the performance is still as good as IAM⁃C. In the ETU⁃ 70Hz scenario, because of rapid variation of the channel taps, the assumption of constant channel over Ω* in (8) is invalid. Consequently, the performance of all the methods degrades and reaches an error floor in higher SNRs. This is a general problem in channel estimation for FBMC systems where the re⁃ ceiver should necessarily have an estimation of intrinsic inter⁃ ferences for channel estimation. However, the degradation on IAMs is more significant as the channel is estimated using two symbols with one zero symbol in between. Therefore, as the channel is not constant over the two pilot symbols, degradation is higher than the proposed method with only one symbol for channel estimation. The Cramer⁃Rao lower bound (CRLB) for the proposed method, derived in Appendix A has also been plotted in the figure for benchmark comparison. It can be seen that the proposed scheme achieves closest performance to the theoretical lower bound in comparison to the other schemes. Fig. 5 shows the MSE comparisons in terms of residual CFO. It is assumed that the CFO has been estimated and com⁃ pensated before channel estimation. As the estimated CFO is not perfect, the residual CFO affects the quality of channel esti⁃ mation. Therefore, the methods are compared in presence of re⁃ sidual CFO in the two channel scenarios without added white Gaussian noise. When the CFO is zero, the MSEs show the er⁃ ror floor of the methods in Fig. 4 at very high SNRs. It can be seen that in EPA channel, the error floor of the proposed meth⁃ od is higher than IAM⁃C, while it has the best performance un⁃ der ETU channel. This is also true for the other values of CFO, where the degradation of MSE in the proposed method is lower than the other two in both channels. 4.3 Bit Error Rate Performance Comparison The BER performance comparison with respect to Eb N0 is i⁃ llustrated in Fig. 6. Evidently, the proposed method performs Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong CRLB: Cramer⁃Rao lower bound EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method SNR: signal⁃to⁃noise ratio CFO: carrier frequency offset EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method ▲Figure 4. MSE performance of the channel estimation methods. ▲Figure 5. MSE performance of the channel estimation methods in presence of residual CFO. ▲Figure 6. BER performance of the channel estimation methods. 30 100 SNR (dB) Meansquareerror 2520151050 10-1 10-2 10-3 10-4 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz Proposed⁃CRLB 150 10-1 Residual CFO (Hz) Meansquareerror 10-2 10-3 100500-50-100-150 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz 22 100 Eb /NO (dB) Biterrorrate 10-1 10-2 10-3 2018161412108642 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz
better compared to the others in low mobility EPA⁃5Hz scenar⁃ io. In the high mobility ETU⁃70Hz channel, the performance deteriorates as the channel varies significantly during the frame time. Consequently, the preamble⁃based channel estima⁃ tion is not a proper choice for high mobility applications and there is an error floor for all the curves showing around six per⁃ cent bit error rate. 5 Conclusions In this paper, we proposed a novel channel estimation algo⁃ rithm with much reduced pilot overhead compared to the exist⁃ ing IAM based approaches. Our results show that the proposed method has better PAPR property. The system performance un⁃ der low mobility and high mobility channels, as well as in the presence of CFO, has been simulated and compared. Accord⁃ ing to the results, the proposed method achieves comparable channel estimation performance to the IAM methods, and bet⁃ ter BER performance due to shorter preamble. Appendix A Cramer⁃Rao Lower Bound for the Proposed Channel Es⁃ timation In this section, a lower bound for the proposed channel esti⁃ mator is derived. We simplify the system using equations (13), (18), (19), and (20) as Y = XH + η, (26) where Y =[ ]y1 y2 is the received signal vector, η =[ ]η1 η2 is the noise vector, H =[ ]h1 h2 is the channel vector to be esti⁃ mated, X is the pilot symbol. The subcarrier index has also been dropped for simplicity. The CRLB is a bound on the smallest covariance matrix that can be achieved by an unbiased estimator, Ĥ , of a parameter vector H as J-1 ≤ CĤ = E{ }( )H - Ĥ ( )H - Ĥ * ; J = E ì í î ï ï ü ý þ ï ï æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H * , (27) where ( )∙ * denotes conjuagate transpose operation, J is the Fisher information matrix and ln p( )Y; H is the log⁃likelihood function of the observed vector Y . The vector Y is a complex Gaussian random vector, i.e., Y ∼ CN( )XH,N0I with likeli⁃ hood function and log⁃likelihood function as where K is a constant. Taking the complex gradient [20] of ln p( )Y; H with respect to H yields ∂ ln p( )Y; H ∂H = - 1 N0 [ ]X* XH - X* Y * . (29) The above equality holds since ∂Y2 ∂H = 0; ∂H* X* Y ∂H = 0; ∂Y* XH ∂H = ( )X* Y * ; ∂H* X* XH ∂H = ( )X* XH * . (30) Thus we can derive, ∂ ln p( )Y; H ∂H* = æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H * = X* Y - X* XH N0 = X* X N0 { }( )X* X -1 X* Y - H = J( )H [ ]Ĥ - H . (31) This proves that the minimum variance unbiased estimator of H is Ĥ = ( )X* X -1 X* Y = Y X . (32) It is efficient in that it attains the CRLB. The Fisher informa⁃ tion matrix J( )H and covariance matrix CĤ of this unbiased estimator are J( )H = E é ë êê ù û úú X* XI2 N0 = E[ ]X* X I2 N0 = Ex N0 I2 CĤ = J-1 ( )H = N0 Ex I2. (33) In (33), Ex is the pilot energy. The CRLB for each diagonal element of J-1 ( )H is var( )ĥ 1 = var( )ĥ 2 = diag[ ]CĤ i = N0 Ex . (34) As the pilots in this system are amplified exploiting intrinsic interference by the factor of 1 + 2γ , Ex should be replaced by E' x = ( )1 + 2γ 2 Ex . Assuming Ex N0 is approximately equal to SNR and considering (25), (34) becomes var( )ĥ 1 = var( )ĥ 2 = N0 Ex 1 ( )1 + 2γ 2 . (35) Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 09 ( )Y; H = 1 ( )πN0 2 exp é ë êê ù û úú- ( )Y - XH * ( )Y - XH N0 = 1 ( )πN0 2 exp é ë ê ù û ú- Y2 - H* X* Y - Y* XH + H* X* XH N0 ; ln p( )Y; H = K - Y2 - H* X* Y - Y* XH + H* X* XH N0 , (28) References [1] A. Sahin, I. Guvenc, and H. Arslan,“A survey on multicarrier communications: Prototype filters, lattice structures, and implementation aspects,”IEEE Commu⁃ nications Surveys Tutorials, vol. 16, no. 3, pp. 1312-1338, Mar. 2014.
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS10 Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong [2] B. Farhang ⁃ Boroujeny,“OFDM versus filter bank multicarrier,”IEEE Signal Processing Magazine, vol. 28, no. 3, pp. 92-112, May 2011. [3] F. Schaich and T. Wild,“Waveform contenders for 5G—OFDM vs. FBMC vs. UFMC,”in 6th International Symposium on Communications, Control and Sig⁃ nal Processing, Athens, Greece, 2014, pp. 457 - 460. doi: 10.1109/ISCC⁃ SP.2014.6877912. [4] Q. Bai and J. Nossek,“On the effects of carrier frequency offset on cyclic prefix based OFDM and filter bank based multicarrier systems,”in IEEE Eleventh In⁃ ternational Workshop on Signal Processing Advances in Wireless Communica⁃ tions, Marrakech, Morocco, Jun. 2010, pp. 1-5. doi: 10.1109/SPAWC.2010. 5670999. [5] M. Sriyananda and N. Rajatheva,“Analysis of self interference in a basic FBMC system,”in IEEE 78th Vehicular Technology Conference, Las Vegas, USA, Sept. 2013, pp. 1-5. doi: 10.1109/VTCFall.2013.6692102. [6] J. Javaudin and Y. Jiang,“Channel estimation in MIMO OFDM/OQAM,”in IEEE 9th Workshop on Signal Processing Advances in Wireless Communications, Recife, Brazil, Jul. 2008, pp. 266-270. doi: 10.1109/SPAWC.2008.4641611. [7] J. Javaudin, D. Lacroix, and A. Rouxel,“Pilot ⁃ aided channel estimation for OFDM/OQAM,”in 57th IEEE Semiannual Vehicular Technology Conference, Je⁃ ju, South Korea, Apr. 2003, pp. 1581-1585. doi: 10.1109/VETECS.2003. 1207088. [8] C. Lele, R. Legouable, and P. Siohan,“Channel estimation with scattered pilots in OFDM/OQAM,”in IEEE 9th Workshop on Signal Processing Advances in Wireless Communications, Recife, Brazil, Jul. 2008, pp. 286-290. doi: 10.1109/ SPAWC.2008.4641615. [9] Z. Zhao, N. Vucic, and M. Schellmann,“A simplified scattered pilot for FBMC/ OQAM in highly frequency selective channels,”in 11th international symposium on Wireless communications systems, Barcelona, Spain, Oct. 2014, pp. 819-823. doi: 10.1109/ISWCS.2014.6933466. [10] J. Bazzi, P. Weitkemper, and K. Kusume,“Power efficient scattered pilot chan⁃ nel estimation for FBMC/OQAM,”in 10th International ITG Conference on Sys⁃ tems, Communications and Coding, Hamburg, Germany, Feb. 2015, pp. 1-6. [11] C. Lélé, J. Javaudin, R. Legouable, A. Skrzypczak, and P. Siohan,“Channel es⁃ timation methods for preamble ⁃ based OFDM/OQAM modulations,”Transac⁃ tions on Emerging Telecommunications Technologies, pp. 741-750, Sept. 2008. doi: 10.1002/ett.1332. [12] C. Lélé, P. Siohan, and R. Legouable,“2 dB better than CP⁃OFDM with OFDM/ OQAM for preamble⁃based channel estimation,”in IEEE International Confer⁃ ence on Communications, Beijing, China, 2008, pp. 1302-1306. doi: 10.1109/ ICC.2008.253. [13] J. Du and S. Signell,“Novel preamble⁃based channel estimation for OFDM/ OQAM systems,”in IEEE International Conference on Communications, Dres⁃ den, Germany, 2009, pp. 1-6. doi: 10.1109/ICC.2009.5199226. [14] E. Kofidis and D. Katselis,“Preamble ⁃ based channel estimation in MIMO ⁃ OFDM/OQAM systems,”in IEEE International Conference on Signal and Im⁃ age Processing Applications, Kuala Lumpur, Malaysia, 2011, pp. 579-584. doi: 10.1109/ICSIPA.2011.6144161. [15] J. Siew, R. Piechocki, A. Nix, and S. Armour. (2002).“A channel estimation method for MIMO⁃OFDM systems,”London Communicaitons Symposium (LCS) [Online]. Available: http://www.ee.ucl.ac.uk/lcs/previous/LCS2002/LCS087.pdf [16] J. Du, P. Xiao, J. Wu, and Q. Chen,“Design of isotropic orthogonal transform algorithm⁃based multicarrier systems with blind channel estimation,”IET com⁃ munications, vol. 6, no. 16, pp. 2695- 2704, Nov. 2012. doi: 10.1049/iet ⁃ com.2012.0029. [17] P. Siohan, C. Siclet, and N. Lacaille,“Analysis and design of OFDM/OQAM systems based on filterbank theory,”IEEE Transactions on Signal Processing, vol. 50, no. 5, pp. 1170-1183, May 2002. [18] E. Kofidis and D. Katselis,“Improved interference approximation method for preamble⁃based channel estimation in FBMC/OQAM,”in 19th European sig⁃ nal processing conference (EUSIPCO⁃2011), Barcelona, Spain, 2011. pp. 1603- 1607. [19] J. Du and S. Signell,“Time frequency localization of pulse shaping filters in OFD/OQAM systems,”in 6th International Conference on Information, Commu⁃ nications Signal Processing, Singapore, 2007, pp. 1-5. [20] S. Kay, Fundamentals of Statistical Signal Processing. Upper Saddle River, USA: Prentice Hall, 1998. Manuscript received: 2016⁃04⁃04 Sohail Taheri (s.taheri@surrey.ac.uk) received his BS degree in electronic engineer⁃ ing and MSc degree in digital electronics from Amirkabir University of Technology, Iran in 2010 and 2012 respectively. He is currently working towards his PhD degree from the Institute for Communication Systems (ICS), University of Surrey, United Kingdom. His current research interests include signal processing for wireless com⁃ munications, waveform design for 5G air interface and physical layer for 5G net⁃ works. Mir Ghoraishi (m.ghoraishi@surrey.ac.uk) is a senior research fellow in the Insti⁃ tute for Communication Systems (ICS), University of Surrey. He joined the Institute in 2012 and is currently leading 5GIC testbed and proof⁃of⁃concept projects. This work area includes several implementation and proof⁃of⁃concept projects, e.g. 5G air⁃ interface proof⁃of⁃concept, distributed massive MIMO implementation, wireless in⁃ band full⁃duplex, millimeter wave hybrid beamforming system, and millimeter wave wireless channel analysis and modelling. He was involved in EU FP7 DUPLO proj⁃ ect as work package leader. He has previously worked in Tokyo Institute of Technol⁃ ogy as assistant professor and senior researcher from 2004 to 2012, after getting his PhD from the same institute. In Tokyo Tech he was involved in several national and small scale projects in planning, performing, implementation, analysis and model⁃ ling different aspect of wireless systems in physical layer, propagation channel and signal processing. He has co⁃authored 100 publications including refereed journals, conference proceedings and three book chapters. XIAO Pei (p.xiao@surrey.ac.uk) received the BEng, MSc and PhD degrees from Huazhong University of Science & Technology, Tampere University of Technology, Chalmers University of Technology, respectively. Prior to joining the University of Surrey in 2011, he worked as a research fellow at Queen’s University Belfast and had held positions at Nokia Networks in Finland. He is a Reader at University of Surrey and also the technical manager of 5G Innovation Centre (5GIC), leading and coordinating research activities in all the work areas in 5GIC. Dr Xiao’s research in⁃ terests and expertise span a wide range of areas in communications theory and sig⁃ nal processing for wireless communications. He has published 160 papers in refer⁃ eed journals and international conferences, and has been awarded research funding from various sources including Royal Society, Royal Academy of Engineering, EU FP7, Engineering and Physical Sciences Research Council as well as industry. CAO Aijun (cao.aijun@zte.com.cn) is a principal architect in ZTE R&D Center, Sweden (ZTE Wistron Telecom AB). He has over 17 years of experience in wireless communications research and development from baseband processing to network ar⁃ chitecture, including design and optimization of commercial UMTS/LTE base⁃sta⁃ tion and handset products, HetNet and small cell enhancement, etc. He has also been involved in standardization works and contributed to several 3GPP technical reports. He is also active in academic and industrial workshops and conferences re⁃ lated to the future wireless networks being as panelists or (co⁃)authors of published papers in refereed journals and international conferences. In addition, he holds more than 50 granted or pending patents. His current focus is 5G technologies relat⁃ ed to the new energy⁃efficient unified air⁃interface and network architecture, e.g., new waveform design, non⁃orthogonal multiple access schemes, random access chal⁃ lenges and innovative signaling architecture for 5G networks. GAO Yonghong (gao.yonghong@zte.com.cn) received his BEng degree in electronic engineering from Tsinghua University, China in 1989, and PhD degree in electronic systems from Royal Institute of Technology, Sweden in 2001. In 1996, he was a visit⁃ ing scientist at Royal Institute of Technology and Ericsson Sweden. In 1999, he joined Ericsson Sweden to develop 3G base stations, baseband algorithms, and base⁃ band ASICs. He joined ZTE European Research Institute (ZTE Wistron Telecom AB, Sweden) in 2002 and has been the CTO of ZTE European Research Institute till now, leading and participating the development of 3G/4G commercial base sta⁃ tions, baseband/RRM algorithms, and baseband ASICs, 3GPP small cell enhance⁃ ment, and from 3 years ago focusing on 5G pre⁃study, 5G standardization, and 5G re⁃ search projects in Europe. He has filed 40+ patents as a main author or co⁃author. His research interests include mobile communication standards/systems, and solu⁃ tions and algorithms for commercial wireless products. BiographiesBiographies
Non⁃Orthogonal Multiple Access Schemes forNon⁃Orthogonal Multiple Access Schemes for 55GG YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei (ZTE Corporation, Shengzhen 518057, China) Abstract Multiple access scheme is one of the key techniques in wireless communication systems. Each generation of wireless communica⁃ tion is featured by a new multiple access scheme from 1G to 4G. In this article we review several non⁃orthogonal multiple access schemes for 5G. Their principles, advantages and disadvantages are discussed, and followed by a comprehensive comparison of these solutions from the perspective of user overload, receiver type, receiver complexity and so on. We also discuss the applica⁃ tion challenges of non⁃orthogonal multiple access schemes in 5G. 5G; non⁃orthogonal multiple access; mMTC Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 002 http://www.cnki.net/kcms/detail/34.1294.TN.20161008.1038.002.html, published online October 8, 2016 Special Topic 1 Introduction ultiple access scheme is the key technique of wireless communications. In 3rd generation (3G) code division multiple access is applied. In 4G orthogonal frequency division multiplex⁃ ing access (OFDMA) is employed. In the coming 5G, non⁃or⁃ thogonal multiple access schemes are hot topics because they can achieve high system capacity. Moreover, massive machine type communication (mMTC) is one of the key scenarios for 5G in which massive connection is required. In this paper, we mainly focus on the non⁃orthogonal multiple access schemes supporting mMTC which has the rapidest growing speed and the urgent deploy demand. Several non ⁃ orthogonal multiple access schemes are pro⁃ posed for 5G, which include multi⁃user shared multiple access (MUSA) [1]-[4], resource spread multiple access (RSMA) [5], sparse code multiple access (SCMA) [6]-[8], pattern division multiple access (PDMA) [9]-[11], interleaver⁃division multiple access (IDMA) [12], [13], and non⁃orthogonal multiple access (NOMA) by power domain [14]. In this paper, the principles, merits and demerits of these schemes are discussed to let read⁃ ers have a full overview on that. 2 Features of 5G 5G has three main technical features, including enhanced mobile broadband (eMBB), mMTC and ultra reliable and low latency communication (URLLC). The eMBB is the evolution of MBB targeting for high data rate and can support high mobil⁃ ity The mMTC is characterized by massive connection with low cost terminals. High reliability and ultra ⁃ low latency are the goals of URLLC. With the development of Internet of things, a large number of terminals will have access to the network. Therefore, mMTC needs to support one million of connections per square kilome⁃ ter. The mMTC, which has the fastest growing speed and the most urgent deployment demand, will create new chances in 5G. The non⁃orthogonal multiple access should support at least mMTC where high user overload is the key requirement. In LTE there are several interactive processes between base station and terminal before the data is transmitted from termi⁃ nal to the base station. This makes sense for long time and con⁃ tinuous data transmission because signaling overhead is small by averaging over a long time. In mMTC each terminal only transmits small data and massive terminals would sporadically transmit their data to the base station. When the same access procedure like in LTE ⁃ A is applied, the signaling overhead will be comparably large and the access efficiency is very low, thus grant⁃free for mMTC is needed in which multiple termi⁃ nals can send their data on the same resource block without multi⁃step negotiations with base station. 3 Non⁃Orthogonal Multiple Access Schemes for 5G Several non⁃orthogonal multiple access schemes have been proposed for 5G. Based on their properties, they can be catego⁃ rized to different types. Most non⁃orthogonal multiple access schemes use spreading codes. When such schemes have other M October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 11
Special Topic Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei predominant properties, such as SCMA and PDMA use code matrix to illustrate how multiple users share the same resource block, and IDMA uses interleaver for user separation, we cate⁃ gorize them as other kind of schemes. In the following joint de⁃ tection denotes message passing algorithm (MPA) based schemes. 3.1 Non􀆼Orthogonal Multiple Access Schemes Based on Spreading Sequences 3.1.1 MUSA MUSA is a non⁃orthogonal multiple access scheme operat⁃ ing in code domain and power domain. Spreading code with short length is applied in MUSA to support a large number of users that share the same resource block. When the number of users is large and the length of the spreading code is small, it is difficult to design large number of spreading code with low correlation when binary element of the spreading code is as⁃ sumed. For binary spreading code the element of the spreading code belongs to the set {1, ⁃1}. Only two values are employed in the spreading code. To overcome this drawback, non⁃binary and complex⁃value spreading code is proposed in MUSA. Ei⁃ ther the real or the image element of the non⁃binary spreading code belongs to the set {1, 0, ⁃1}, there are nine values for se⁃ lection. This provides much more flexibility of spreading code design. Because the real and image elements of the spreading code are 1, 0 or ⁃1, the multiplication operation can be imple⁃ mented by addition operation which will reduce the implemen⁃ tation complexity. Fig. 1 shows the basic features of MUSA, where multiple users could transmit data on the same resourc⁃ es by using randomly selected non⁃orthogonal complex spread⁃ ing codes with short length (e.g. 4). In this example 12 users share 4 resource blocks, and the user overload is 300%. MU⁃ SA is always modeled by multiple spreading codes superposed on the same resource block. It can also be modeled by a code matrix. The code matrix of MUSA with 300% overload is given by In 5G, mMTC is one important application scenario. In this scenario MUSA is preferred since grant⁃free transmission can be readily supported. A device terminal autonomously access⁃ es the communication system without base station (BS) sched⁃ uling. Blind detection is applied at BS for MUSA in which ac⁃ tive user, user spreading code and user channel would not be known before hand. Because the spreading code length is rela⁃ tive short and its elements have limited values, BS can gener⁃ ate numerous local spreading codes with low correlation. By us⁃ ing these local spreading codes and the received signal, we can closely approximate the optimal performance of MMSE estima⁃ tor. Then the user signal with the highest signal⁃to⁃interference ⁃plus⁃noise ratio (SINR) can be detected and decoded. After that user’s signal is successfully decoded, it can be employed for channel estimation. After interference cancellation, the us⁃ er signal with the second highest SINR is detected and decod⁃ ed. During this process no pilots or preamble are needed for channel estimation, which facilitates MUSA application in mMTC because most other schemes rely on additional over⁃ head for channel estimation. The blind detection for MUSA is verified over flat fading channel and multi⁃path fading channel [3], [15]. The main advantages of MUSA are reflected by high over⁃ loading factor, robust blind detection and true sense of grant⁃ free transmission. Due to frequency ⁃ diversity gain achieved, 700% user overload can be achieved by MUSA over multi⁃path fading channel [15]. User detection can be carried out without the knowledge of the spreading code. User transmitted signal can be applied for enhanced channel estimation once it has been correctly decoded. Users can transmit their signals ac⁃ cording to their demand. The possibility of collision due to the same spreading code applied is small since large number of the spreading codes can be accommodated. Successive interference cancelation (SIC)⁃based receiver is applied for MUSA. It works well when there is SINR difference among the received signals. However, when the difference is small, there would be certain performance loss due to error propagation. While there is inherent SINR different in mMTC due to free power control, it is not a so serious problem for the signal detection of MUSA. The SINR difference is small, so it can be solved by using more advanced receiver, such as joint detection and decoding scheme. 3.1.2 RSMA In RSMA (Fig. 2), a group of users’signals are superposed on the same resource blocks, and each user’s signal is spread over the entire frequency/time resource blocks. Different users’ October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS12 G = é ë ê êê ê ù û ú úú ú 1 + i 1 + i 1 + i 1 1 - i 1 + i i -i -1 + i -1 + i -1 i i -i 1 1 + i -i -i 1 + i 1 - i -1 - i -1 + i 1 i 1 -1 1 + i -1 - i -1 1 1 -1 1 + i -i 0 1 1 -1 + i 0 0 1 - i 1 0 0 0 0 0 1 + i SIC: successive interference cancelation ▲Figure 1. An example of MUSA with 300% user overload [4]. Elements of complex spreading codeR-1 0 1 -1 I Complex spreading code set ··· Each user randomly picks one code for spreading Codeword⁃level SIC receiver C1 C2 C12 S1 + S2 + +··· S12 = 1
Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei Special Topic signals within the resource blocks may be not orthogonal. Low code rate channel codes are employed to achieve large coding gain. Relative long spreading codes with good correlation prop⁃ erty are applied to reduce the multi⁃user interference. Scram⁃ blers can be employed with the same purpose as the spreading codes. Interleaver is optional for RSMA according to the sys⁃ tem requirements. Depending on the application scenarios, it includes single carrier RSMA and multi⁃carrier RSMA. For the former it is op⁃ timized for battery power consumption and coverage extension for small data transactions by utilizing single carrier wave⁃ forms, very low peak⁃to⁃average⁃power⁃ratio (PAPR) modula⁃ tions. It allows grant ⁃ less transmission and potentially allow asynchronous access. While for the latter it is optimized for low latency access for radio resource connection (RRC) con⁃ nected users (i.e., timing with eNB already acquired) and al⁃ lows for grant⁃less transmission. The advantage of RSMA is that it supports asynchronous and grant ⁃ less transmission, so the signaling overhead is re⁃ duced. The disadvantage is that its user overload is limited when rake receiver is applied. By using advanced receiver, such as SIC based receiver, the overload can be enhanced. 3.2 Non􀆼Orthogonal Multiple Schemes Based on Structured Coding Matrix 3.2.1 SCMA Sparse codebook is applied at SCMA to reduce the detection complexity. At the same time joint detection is employed for SCMA to achieve excellent performance. The codewords are composed of multi⁃dimensional com⁃ plex symbols, and the codewords in the same codebook have the same sparse pattern. Sparse codeword mapping utilizes low density spreading and could be referred to as sparse spreading. At the receiver, iterative multi⁃user detection based on MPA is used. Fig. 3 shows an exam⁃ ple of SCMA, where the coded bits of a data stream are directly mapped to a codeword with sparse non⁃zero ele⁃ ments from a codebook. With 6 sparse codewords transmitted over 4 orthogonal resources, the user overload is 150% . The coding matrix of Fig. 3 is given by G = é ë ê êê ê ù û ú úú ú 1 1 0 0 1 0 1 0 1 0 0 1 0 1 1 0 0 1 0 1 0 0 1 1 To reduce the multi⁃user interference and the de⁃ tection complexity, sparse signature sequence is ap⁃ plied in SCMA for spreading. User signal is modu⁃ lated by a codebook in which multidimensional modulation maps of the input coded bits to the points in the multiple complex dimensions [6]. By such operation shaping gain is achieved, which is claimed as one major property of SCMA. The main disadvantage of SCMA is its high detection and de⁃ coding complexity even sparse signature sequence is applied. The detection and decoding complexity is even higher when large size constellation and a large number of users are em⁃ ployed. And additional pilots or preambles are needed for multi ⁃user channel estimation, which may reduce system spectral ef⁃ ficiency. Because the size of the codebook is limited, if two us⁃ ers choose the same codeword, collision will happen. Collision is a serious problem for SCMA, which limits its overload capa⁃ bility. For example, with 6 users transmitted over 4 units, the user overload is only 150% . Although the overloading factor can be enhanced by using longer spreading codes, the detec⁃ tion complexity will increase significantly since the size of the codebook and the searching space is enlarged. 3.2.2 PDMA For PDMA, the code in a code matrix is used to define map⁃ ping from data to a group of resources. Each element in the code corresponds to a resource in the resource group. PDMA can be detected by SIC type receiver. It also can be detected by MPA based scheme in the receiver. PDMA is designed for SIC⁃based receiver originally. The different diversity orders of different users by carefully design the code matrix facilitate the multi⁃user signal detection. The user with the largest diver⁃ October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 13 CP: cyclic prefix IFFT: inverse fast fourier transform OFDM: orthogonal frequency division multiplexing PAPR: peak⁃to⁃average⁃power⁃ratio RSMA: resource spread multiple access TDM: time division multiplexing MUD: multiple user detection MPA: message passing algorithm ▲Figure 2. RSMA block diagrams [5]. ▲Figure 3. An example of SCMA with 150% user overload [8]. Variable rate encoder TDM pilot insertion Spreader/ scrambler Low PAPR modulation Optional CP (a) Single carrier RSMA Pilot insertion Spreader/ scrambler Coder Serial to parallel IFFT Parallel to serial Cyclic prefix e j2πfct (b) OFDM RSMA e j2πfct Codebook 1 Codebook 2 Codebook 3 Codebook 4 Codebook 5 Codebook 6 (0,0) (1,0) (0,1) (1,1) (1,1) (0,0) Bit streams are mapped to sparse codewords MUD based on MPA 6 sparse codewords are transmitted over 4 orthogonal resources
sity order is detected first, and then the user with the largest di⁃ versity order among the remaining users is detected; in this way, all users’signals will be detected. To further improve the performance of PDMA, joint detec⁃ tion based scheme is proposed. In this case the unbalance weight of each column is interpreted as the irregular code weight. As we know irregular low density parity check (LDPC) code has better performance than that of the regular one. By carefully designing the code matrix with joint detection, even better performance can be obtained by PDMA compared with regular code matrix (for example non⁃orthogonal multiple ac⁃ cess with low density signatures can be regards as regular code). The main disadvantage of PDMA is its low user overload (us⁃ er overload is defined by the number of user over the resource block that all users share). It is difficult to achieve overload of 400% with the 4⁃row code matrix (when the row of the code ma⁃ trix is K, the largest user number it supported is 2K ⁃1 [10]).The complexity is high for high order modulation when joint⁃detec⁃ tion scheme is applied. Additional pilots or preamble are need⁃ ed for channel estimation. Because the number of patterns is limited, there is high probability of collision when users are al⁃ lowed to randomly select the patterns. 3.3 Non􀆼Orthogonal Multiple Schemes Based on Interleaver IDMA was proposed by [12], [13], in which users are sepa⁃ rated by different interleavers. Low ⁃ rate channel decoding is applied and the coded bits are repeated multiple times to in⁃ crease the SINR after accumulating the received signals. After channel coding and repetition, interleaver is employed to make the transmission bits randomly distributed. A block diagram of IDMA is shown in Fig. 4 where C represents channel encod⁃ ing, S denotes repetition and π is the interleaver. The strategy of user separation for IDMA is different from other non⁃orthogo⁃ nal multiple access schemes. Interleaver is used for user sepa⁃ ration and the length of the interleaver is very large (the length of the interleaver equals to the number of the bits after channel coding and repetition), thus this provides good base for a large number of users access by using IDMA. It is reported that 64 users can be supported by IDMA which share the same re⁃ source block [12]. This goal can never be achieved by other non⁃ orthogonal multiple access schemes at present. At the receiver side each user’s signal is detected, demodu⁃ lated and de⁃interleaved according to its own interleaver pat⁃ terns. The soft information of decoded bits is input to elementa⁃ ry signal estimator (ESE) for soft information updating. After soft information updating new soft information is input to the decoder for channel decoding again. Several iterative detec⁃ tions between ESE and channel decoder are needed to achieve the best performance. The detection and decoding complexity does not increase exponentially with the user number and total spectral efficiency. The complexity increases linearly, which is also different from other non ⁃ orthogonal multiple access schemes which use joint detection and decoding scheme. The main advantages of IDMA are its high user overload and excellent performance. And high spectral efficiency can be achieved by IDMA (as high as 8 b/s/Hz). The performance gap between IDMA simulation result and the system capacity bound is almost the same from the spectral efficiency 1 b/s/Hz to 8 b/s/Hz (this means the detection and decoding scheme is very robustness) [12]. These two merits are seldom achieved by other non⁃orthogonal multiple schemes simultaneously. The main disadvantage of IDMA may be its large decoding complexity and decoding latency, especially when a large num⁃ ber of users are supported. The reason is that when large num⁃ ber of iterative detection and decoding are needed with the in⁃ creasing of user number. For example, tens of channel decoder procedures are needed in the signal detection and tens of inter⁃ active actions between channel decoder and ESE detector are required. Thus high convergence algorithm is needed in the signal detection for IDMA in future. To solve the problem of large decoding complexity and decoding latency, interleaver patterns can be pre⁃allocate to small number of users, i.e., the relatively small pool size, so that the complexity of blind decod⁃ ing and channel decoding latency can be maintained below cer⁃ tain level. Another disadvantage is that additional pilots or long preamble is needed to estimate the users’channels. 3.4 Non􀆼Orthogonal Multiple Access (NOMA) Scheme Based on Power􀆼Domain Division Multi⁃user signals can be superposed together in NOMA. In NOMA, capacity or throughput improvement can be expected by sharing the same radio resources among multiple user equipments (UEs) as shown in Fig. 5a and Fig. 5b. A typical application scenario of NOMA is that a cell⁃center user and a cell⁃edge user are serviced by NOMA. Due to small path loss of cell center user, in the signal detection it is detected first and the signal of cell edge user is treated as interference. In the signal detection of cell edge user, the signal of cell center user is detected and decoded first. Then the signal of the cell center user is cancelled from the received signal and signal of cell edge user is detected and decoded. The main advantage of NOMA is that excellent performance can be achieved when a cell center user and cell edge user are scheduled with moderate computational complexity (SIC detec⁃ Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS14 ▲Figure 4. IDMA block diagram [13]. Multiple access channel π1 x1 SC d1 Transmitter for user 1 Transmitter for user K πK xK SC dK … … Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei
tor is always applied). And a user overload of 200% is easily achieved. The main disadvantage of NOMA is that there is re⁃ striction on the scheduled users. Usually a cell center user and a cell edge user should be scheduled on the same resource block. When two cell center users or two cell edge users are scheduled and SIC ⁃ type receiver is applied, there is perfor⁃ mance loss because one user always has low SINR due to inter⁃ ference from another user’s signal. The NOMA is designed for eMBB originally. Thus when it is applied for mMTC, the re⁃ ceived SINR would not be high and the number of supported users is very limited (two or three users are supported on the same resource block, which is much smaller than other non⁃or⁃ thogonal multiple access schemes). And additional pilots or long preamble is needed to estimate the users’channels. A summary of these non ⁃ orthogonal multiple schemes are shown on Table 1. They are compared in terms of multiplexing domain, user overload, receiver type, receiver complexity and so on. Among these schemes MUSA achieves a good balance between performance and complexity, such as high user over⁃ load, low implementation complexity and flexible in grant⁃free transmission. 4 Application Challenges of Non⁃Orthogonal Multiple Access Schemes in 5G Followings are the requirements for the non⁃orthogonal mul⁃ tiple access schemes. These factors should be considered when we design the non⁃orthogonal multiple access schemes. 4.1 Coverage Coverage is an important issue for mMTC since terminals may distribute over a large area, thus it is crucial for non⁃or⁃ thogonal multiple access schemes to support terminals with dif⁃ ferent received power due to path loss. And the non⁃orthogonal multiple access schemes should have the ability of robustness to the high interference. To increase the coverage, low code rate channel coding or large spreading factor could be consid⁃ ered. High efficiency power amplifier is appealing for coverage extension, which requires transmit signals with low PAPR. 4.2 PAPR When the non⁃orthogonal multiple access scheme is applied for uplink, PAPR should be considered to increase the trans⁃ mission efficiency and reduce the transmission power thus save the battery life. The battery life is desired to be 10 years for mMTC, so it puts a big challenge on the non ⁃ orthogonal multiple access scheme. The signal of the non⁃orthogonal mul⁃ tiple access schemes which have low PAPR will be preferred in practical implementation. Filtered π/2 ⁃ binary phase shift keying (BPSK) and Gaussian filtered minimum shift keying (GMSK) have good property of low PAPR and are employed for PAPR reduction in RSMA [16]. 4.3 Implementation Complexity The implementation complexity includes two parts: transmit⁃ ter implementation complexity and receiver implementation complexity. Because multi⁃user detection is carried out at re⁃ ceiver side, which has the highest complexity over the entire signal processing chain, the main implementation complexity is at the receiver side. Two types of receivers are always ap⁃ plied for non⁃orthogonal multiple access schemes: SIC⁃based receiver and joint ⁃ detection ⁃ based receiver. The former can achieve a good balance between performance and complexity. As the number of user increases, the complexity only increases linearly. While it suffers performance loss in some cases, such as the path⁃losses among different users are the same. Joint⁃de⁃ tection ⁃ based receiver achieves excellent performance at the Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 15 ▲Figure 5. NOMA block diagram. NOMA: non⁃orthogonal multiple access ▼Table 1. Summary of different non⁃orthogonal multiple access schemes MUSA: multi⁃user shared multiple access RSMA: resource spread multiple access SCMA: sparse code multiple access PDMA: pattern division multiple access IDMA: interleaver⁃division multiple access NOMA: non⁃orthogonal multiple access SIC: successive interference cancelation BS: base station Multiplexing domain User overload Receiver type Receiver complexity Grant⁃free transmission MUSA Spreading High SIC Low Users can randomly pick up spreading sequence RSMA Spreading/ scramble Low Raker or SIC Low Power control needed SCMA Codebooks Middle Joint detection High Codeword for each user is predefined and known at BS. Codeword collision is a problem due to limited number of codewords PDMA Pattern Middle SIC or joint detection Low for SIC High for joint detection Pattern is predefined and known at BS. User collision is a problem due to limited number of patterns IDMA Interleaver High Iterative detection and decoding High* Interleaver patterns are known at BS NOMA Power Low SIC Low Grant⁃ based (a) NOMA transmission (b) Signal strength for NOMA Base station Cell center user Cell edge user Strength of cell edge user signal Strength of cell center user signal Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei * Unlike joint detection scheme whose complexity increases exponentially as the number of the users and spectral efficiency increases, the complexity of IDMA only linear in⁃ creases with the number of users and the spectral efficiency. The high complexity is due to large number of iterative detection and decoding.
cost of high computational complexity. Although by some de⁃ signs, such as sparse coding matrix, the decoding complexity is reduced significantly, however, as the constellation size and the number of users increase, the decoding complexity grows exponentially. This bottleneck should be solved before such scheme is employed in practical systems. 4.4 Combination with Multiple􀆼Input Multiple􀆼Output (MIMO) By applying MIMO technique large system capacity or high transmission/receiver reliability can be achieved. It had been proved that MIMO is a very effective technique in wireless communication systems. The non ⁃ orthogonal multiple access schemes should be amiable for MIMO. As the first step, SISO is assumed in the research of the new non⁃orthogonal multiple access schemes. However, compatibility with MIMO should be considered in the next research step. 4.5 Flexibility The non ⁃ orthogonal multiple access schemes should have flexibility. It can change its parameters to support different use scenarios. For example, in some cases high user overload is the system design target, while in other cases coverage is the most important factor. This imposes requirements on the non⁃ orthogonal multiple access scheme design. By changing the pa⁃ rameter of the non⁃orthogonal multiple access schemes, differ⁃ ent targets can be achieved. Another example is that non⁃or⁃ thogonal multiple access schemes should support both multi⁃ carrier system and single⁃carrier systems to facilitate its appli⁃ cation scenarios. 5 Conclusion This article reviews the main non ⁃ orthogonal multiple ac⁃ cess schemes for 5G. Their principles and unique properties are discussed. MUSA can support high user overload with low implementation complexity and is more suitable for grant⁃free transmission. RSMA is suitable for single⁃carrier system and multi ⁃ carrier system. It has good property of large coverage. SCMA can achieve additional shaping gain and PDMA has the flexibility in the patterns design. IDMA can accommodate very high user overload and support high spectral efficiency at the cost of large decoding complexity and decoding latency. NO⁃ MA works well for large SINR difference among the non ⁃ or⁃ thogonal multiple users. At the same time they have their own disadvantages. It is important to integrate the advantages of dif⁃ ferent schemes to make the final designed scheme fulfill the challenging requirements of coming 5G. Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS16 References [1] Discussion on Multiple Access for New Radio Interface, 3GPP R1⁃162226, Apr. 2016. [2] Z. Yuan, G. Yu, W. Li, Y. Yuan, and X. Wang,“Multi⁃user shared access for in⁃ ternet of things,”in IEEE Vehicular Technology Conference, Nanjing, China, May 2016, pp 1-5. doi: 10.1109/VTCSpring.2016.7504361. [3] Receiver Implementation for MUSA, 3GPP R1⁃164270, May 2016. [4] Contention ⁃ Based Non ⁃ Orthogonal Multiple Access for UL mMTC, 3GPP R1 ⁃ 164269, May 2016. [5] Resource Spread Multiple Access, 3GPP R1⁃164688, May 2016. [6] M. Taherzadeh, H. Nikopour, A. Bayesteh, H. Baligh,“SCMA codebook de⁃ sign”, in IEEE Vehicular Technology Conference, Vancouver, Canada, Sept. 2014, pp.1-5, doi: 10.1109/VTCFall.2014.6966170. [7] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE Internation⁃ al Symposium On Personal, Indoor And Mobile Radio Communications, London, UK, Sept. 2013, pp. 332-336. doi: 10.1109/PIMRC.2013.6666156. [8] Future Mobile Communication Forum. (2016, Jul. 7). 5G white paper v2.0, part d—alternative multiple access v1 [Online]. Available: http://www.future ⁃ forum. org/dl/151106/whitepaper.rar [9] Candidate Solution for New Multiple Access, 3GPP R1⁃163383, Apr. 2016. [10] X. Dai, S. Chen, S. Sun, et al.,“Successive interference cancelation amenable multiple access (SAMA) for future wireless communications,”in Proc. IEEE In⁃ ternational Conference on Communication Systems, Macau, China, Nov. 2014, pp. 222-226. doi: 10.1109/ICCS.2014.7024798. [11] X. Dai,“Successive interference cancellation amenable space⁃time codes with good multiplexing⁃diversity tradeoff,”Wireless Personal Communications, vol. 55, no. 4, pp. 645-654, Dec. 2010. doi: 10.1007/s11277⁃009⁃9826⁃9. [12] P. Li, L. Liu, K. Wu, and W. K. Leung,“On interleave⁃division multiple⁃ac⁃ cess,”in IEEE International Conference on Communications, Paris, France, Jun. 2004, pp. 2869-2873. doi: 10.1109/ICC.2004.1313053. [13] P. Li, L. Liu, K. Wu, and W. K. Leung,“Interleave division multiple⁃access,” IEEE Transactions on Wireless Communications, vol. 5, no. 4, pp. 938-947, Apr. 2006. doi: 10.1109/TWC.2006.1618943. [14] Y. Saito, Y. Kishiyama, A. Benjebbour, et al.,“Non⁃orthogonal multiple access (NOMA) for cellular future radio access,”in IEEE Vehicular Technology Con⁃ ference, Dresden, Germany, Jun. 2013, pp. 1-5. doi: 10.1109/VTC Spring.2013. 6692652. [15] Receiver Details and Link Performance for MUSA, 3GPP R1⁃166404, Aug. 2016. [16] Resource Spread Multiple Access, 3GPP R1⁃166359, Aug. 2016. Manuscript received: 2016⁃07⁃07 Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei YAN Chunlin (yan.chunlin@zte.com.cn) received his PhD degree from University of Electronic Science and Technology of China (UESTC), China in 2004. He worked at DOCOMO Beijing communications lab from 2005 to 2016. Since 2016 he has been with ZTE Corporation. He has published tens of papers in IEEE ICC, Globecom, VTC, PIMRC and other international conferences. His main research interests in⁃ clude synchronization, binary and non⁃binary channel coding, MIMO detection and non⁃orthogonal multiple access technique for 5G. YUAN Zhifeng (yuan.zhifeng@zte.com.cn) received his MS degree in signal and in⁃ formation processing from Nanjing University of Post and Telecommunications (NUPT), China in 2005. He has been worked with the Wireless Technology Ad⁃ vance Research Department of ZTE Corporation since 2006 and the leader of the team for new multi⁃access (NMA) for 5G wireless systems since 2012. His research interests include wireless communication, MIMO systems, information theory, multi⁃ ple access, error control coding, adaptive algorithm, and high⁃speed VLSI design. LI Weimin (li.weimin6@zte.com.cn) received his master degree from NUPT, China. He joined in ZTE Corporation in 2010, and is responsible for technology research of power control and interference control in wireless communications. His current re⁃ search focuses on multiple access technology for 5G system. YUAN Yifei (yuan.yifei@zte.com.cn) received his master degree from Tsinghua Uni⁃ versity, China, and PhD from Carnegie Mellon University, USA. He was with Alca⁃ tel⁃Lucent from 2000 to 2008, working on 3G/4G key technologies. Since 2008, he has been with ZTE as the technical director of standards research on LTE⁃advanced physical layer and 5G new radio. His research interests include MIMO, channel cod⁃ ing, resource scheduling, multiple access, and NB⁃IoT. He was admitted to Thou⁃ sand Talent Plan Program of China in 2010. He has extensive publications, includ⁃ ing two books on LTE⁃Advanced. BiographiesBiographies
A Survey of Downlink Non⁃Orthogonal MultipleA Survey of Downlink Non⁃Orthogonal Multiple Access forAccess for 55G Wireless Communication NetworksG Wireless Communication Networks WEI Zhiqiang 1 , YUAN Jinhong 1 , Derrick Wing Kwan Ng 1 , Maged Elkashlan2 , and DING Zhiguo3 (1. The University of New South Wales, Sydney, NSW 2052, Australia; 2. Queen Mary University of London, London E1 4NS, UK; 3. Lancaster University, Lancaster LA1 4YW, UK) Abstract Non⁃orthogonal multiple access (NOMA) has been recognized as a promising multiple access technique for the next generation cel⁃ lular communication networks. In this paper, we first discuss a simple NOMA model with two users served by a single⁃carrier si⁃ multaneously to illustrate its basic principles. Then, a more general model with multicarrier serving an arbitrary number of users on each subcarrier is also discussed. An overview of existing works on performance analysis, resource allocation, and multiple⁃in⁃ put multiple⁃output NOMA are summarized and discussed. Furthermore, we discuss the key features of NOMA and its potential re⁃ search challenges. non⁃orthogonal multiple access (NOMA); successive interference cancellation (SIC); resource allocation; multiple⁃input multiple⁃ output (MIMO) Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 003 http://www.cnki.net/kcms/detail/34.1294.TN.20161019.0829.002.html, published online October 19, 2016 Special Topic 1 Introduction and Background he fifth generation (5G) communication system is on its way. It is widely believed that 5G is not just an incremental version of the fourth generation (4G) communication systems [1] due to the in⁃ creasing demand of data traffic and the expected new services and functionalities, such as internet⁃of⁃things (IoT) and cloud⁃ based architectural applications [2]. These envisioned services pose challenging requirements for 5G wireless communication systems, such as much higher data rates (100-1000 times fast⁃ er than current 4G technology), lower latency (1 ms for a roundtrip latency), massive connectivity and support of diverse quality of service (QoS) (106 devices/km2 with diverse QoS re⁃ quirements) [1]. From a technical perspective, to meet the aforementioned challenges, some potential technologies, such as massive multiple⁃input multiple⁃output (MIMO) [3], [4], mil⁃ limeter wave [5], [6], and ultra densification and offloading [7]- [9], have been discussed extensively. Besides, it is expected to employ a future radio access technology for 5G, which is flexi⁃ ble, reliable [10], and efficient in terms of energy and spec⁃ trum [11], [12]. Radio access technologies for cellular commu⁃ nications are characterized by multiple access schemes, such as frequency⁃division multiple access (FDMA) for the first gen⁃ eration (1G), time⁃division multiple access (TDMA) for the sec⁃ ond generation (2G), code ⁃ division multiple access (CDMA) used by both 2G and the third generation (3G), and orthogonal frequency division multiple access (OFDMA) for 4G. All these conventional multiple access schemes are categorized as or⁃ thogonal multiple access (OMA) technologies, where different users are allocated to orthogonal resources in either time, fre⁃ quency, or code domain in order to mitigate multiple access in⁃ terference (MAI). However, OMA schemes are not sufficient to support the massive connectivity with diverse QoS require⁃ ments. In fact, due to the limited degrees of freedom (DoF), some users with better channel quality have a higher priority to be served while other users with poor channel quality have to wait to access, which leads to high unfairness and large laten⁃ cy. Besides, it is inefficient when allocating DoF to users with poor channel quality. In this survey, we focus on one promising technology, non⁃orthogonal multiple access (NOMA), which in our opinion will contribute to disruptive design changes on ra⁃ dio access and address the aforementioned challenges of 5G. In contrast to conventional OMA, NOMA transmission tech⁃ niques intend to share DoF among users via superposition and consequently need to employ multiple user detection (MUD) to separate interfered users sharing the same DoF, as illustrated in Fig. 1. NOMA is beneficial to enlarge the number of connec⁃ tions by introducing controllable symbol collision in the same DoF. Therefore, NOMA can support high overloading transmis⁃ T October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 17
Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS18 sion and further improve the system capacity given limited re⁃ source (spectrum or antennas). In addition, multiple users with different types of traffic request can be multiplexed to transmit concurrently on the same DoF to improve the latency and fair⁃ ness. The comparison of OMA and NOMA is summarized in Table 1. As a result, NOMA has been recognized as a promis⁃ ing multiple access technique for the 5G wireless networks due to its high spectral efficiency, massive connectivity, low la⁃ tency, and high user fairness [13]. For example, multiuser su⁃ perposition transmission (MUST) has been proposed for the third generation partnership project long ⁃ term evolution ad⁃ vanced (3GPP⁃LTEA) networks [14]. Three kinds of non⁃orthogonal transmission schemes have been proposed and studied in the MUST study item. Through the system⁃level performance evaluation, it has been shown that the MUST can increase system capacity as well as improve user experience. Recently, several NOMA schemes have been proposed and received significant attention. According to the domain of mul⁃ tiplexing, the authors in [13] divided the existing NOMA tech⁃ niques into two categories, i.e., code domain multiplexing (CDM) and power domain multiplexing (PDM). The CDM⁃NO⁃ MA techniques, including low⁃density spreading (LDS) [15]- [17], sparse code multiple access (SCMA) [18], pattern divi⁃ sion multiple access (PDMA) [19], etc, introduce redundancy via coding/spreading to facilitate the users separation at the re⁃ ceiver. For instance, LDS ⁃ CDMA [15] intentionally arranges each user to spread its data over a small number of chips and then interleave uniquely, which makes optimal MUD affordable at receiver and exploits the intrinsic interference diversity. LDS⁃ OFDM [16], [17], as shown in Fig. 2, can be interpreted as a system which applies LDS for multiple access and OFDM for multicarrier modulation. Besides, SCMA is a generalization of LDS methods where the modulator and LDS spreader are merged. On the other hand, PDM⁃NOMA exploits the power do⁃ main to serve multiple users in the same DoF, and performs successive interference cancellation (SIC) at users with better channel conditions. In fact, the non⁃orthogonal feature can be introduced either in the power domain only or in the hybrid code and power domain. Although DM⁃NOMA has the poten⁃ tial code gain to improve spectral efficiency, PDM⁃NOMA has a simpler implement since there is almost no big change in the physical layer procedures at the transmitter side compared to current 4G technologies. In addition, PDM⁃NOMA paves the way for flexible resource allocation via relaxing the orthogonali⁃ ty requirement to improve the performance of NOMA, such as spectral efficiency [20], [21], energy efficiency [22], and user fairness [23]. Therefore, this paper will focus on the PDM⁃NO⁃ MA, including its basic concepts, key features, existing works, and future research challenges. 2 Fundamentals of NOMA This section presents the basic model and concepts of single⁃ antenna downlink NOMA. The first subsection presents a sim⁃ ple downlink single⁃carrier NOMA (SC⁃NOMA) system serving two users simultaneously, while the second subsection pres⁃ ents a more general multi⁃carrier NOMA (MC⁃NOMA) model for serving an arbitrary number of users in each subcarrier. 2.1 Two􀆼User SC􀆼NOMA1 Benjebbour, Saito et al. [24], [25] proposed the system mod⁃ el of downlink NOMA with superposition transmission at the base station (BS) and successive interference cancellation DoF: degrees of freedom ▲Figure 1. From OMA to NOMA via power domain multiplexing. AWGN: additive white Gaussian noise FEC: forward error correction LDS: low⁃density spreading OFDM: orthogonal frequency⁃division multiplexing 1 In this paper, a two⁃user NOMA system means that two users are multi⁃ plexed on each subcarrier simultaneously. Similarly, a multiuser NOMA system means that an arbitrary number of users are multiplexed on each subcarrier simultaneously. ▼Table 1. Comparison of OMA and NOMA NOMA: non⁃orthogonal multiple access OMA: orthogonal multiple access QoS: quality of service OMA NOMA Advantages Simpler receiver detection •Higher spectral efficiency •Higher connection density •Enhanced user fairness •Lower latency •Supporting diverse QoS Disadvantages •Lower spectral efficiency •Limited number of users •Unfairness for users •Increased complexity of receivers •Higher sensitivity to channel uncertainty ▲Figure 2. Block diagram of an uplink LDS⁃OFDM system. Power Orthogonality between users Power Superposition & power allocation DoF DoF FEC encoder b1 Symbol mapper LDS spreader S1 OFDM modulator OFDM channel FEC encoder bM Symbol mapper LDS spreader SM OFDM modulator OFDM channel … AWGN FEC decoder b1 Symbol demapper S1 FEC decoder bM Symbol demapper SM LDS detector … … … y1 yK OFDM demodulator …
A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 19 (SIC) at the user terminals, which is illustrated in Fig. 3 in case of one BS and two users. The BS transmits the messages of both user 1 and user 2, i.e., s1 and s2 , with different trans⁃ mit powers p1 and p2 , on the same subcarrier, respectively. The corresponding transmitted signal is represented by x = p1 s1 + p2 s2 , (1) where transmit power is constrained by p1 + p2 = 1. The re⁃ ceived signal at user i is given by yi = hi x + vi , (2) where hi denotes the complex channel coefficient including the joint effect of large scale fading and small scale fading. Variable vi denotes the additive white Gaussian noise (AW⁃ GN), and vi ∼ CN(0,σ2 i ) , where CN(0,σ2 i ) denotes the circula⁃ rly symmetric complex Gaussian distribution with mean zero and variance σ2 i . We assume that user 1 is the cell⁃center u⁃ ser with a better channel quality (strong user), while user 2 is the cell⁃edge user with a worse channel quality (weak user), i.e., ( ||h1 2 σ2 1)≥( ||h2 2 σ2 2) . According to the NOMA protocol [26], the BS will allocate more power to the weak user to pro⁃ vide fairness and facilitate the SIC process, i.e., p1 ≤ p2 . In downlink SC⁃NOMA, the SIC process is implemented at the receiver side. The optimal SIC decoding order is in the de⁃ scending order of channel gains normalized by noise. It means that user 1 will decode s2 first and remove the inter⁃user inter⁃ ference of user 2 by subtracting s2 from the received signal yi before decoding its own message s1 . On the other hand, u⁃ ser 2 does not perform interference cancellation and directly decodes its own message s2 with interference from user 1. For⁃ tunately, the power allocated to user 2 is larger than that of us⁃ er 1 in the aggregate received signal y2, which will not intro⁃ duce much performance degradation compared to allocating us⁃ er 2 on this subcarrier exclusively. The rate region of SC⁃NO⁃ MA is illustrated in Fig. 4 in comparison with that of OMA, where it has been proved that NOMA schemes are very likely to outperform OMA schemes in [27]. It is noted that the rate re⁃ gion of NOMA only covers a part of the capacity region of broadcast channel with SIC receiver [28] due to the power con⁃ straint p1 ≤ p2 . 2.2 Multiuser MC􀆼NOMA For a downlink MC⁃NOMA system with one BS serving an arbitrary number of users, such as N, the available bandwidth is divided into a set of K subcarriers, where N> K, i.e., an over⁃ loading scenario that OFDMA cannot afford. The channel be⁃ tween user n and the BS on subcarrier k is denoted by hk,n , and is assumed to be perfectly known at both the transmitter and receiver side. The BS schedules all users across all subcar⁃ riers by ξk and ζn , where ξk denotes a user set allocated on subcarrier k and ζn denotes a subcarrier set occupied by user n. Without loss of generality, the channel gains of all users allo⁃ cated on subcarrier k are sorted as ||hk,b(1) 2 ≥ ||hk,b(2) 2 ≥ … ≥ | | | |hk,b( ||ξk ) 2 , where ||ξk denotes the card⁃ inality of the user set ξk and b( )∙ indicates the mapping b⁃ etween the sorted channel gain order and the original one. For instance, for subcarrier k occupied by three users ξk ={ }1,2,3 and ||hk,2 2 ≥ ||hk,3 2 ≥ ||hk,1 2 , we will have b(1)= 2 , b(2)= 3 , and b(3)= 1, respectively. It is noted that the mapping func⁃ tions are various on different subcarriers due to users’differ⁃ ent frequency selective fading patterns. According to NOMA protocol [26], all users in ξk share sub⁃ carrier k by different transmission power pk,b(l) based on the gi⁃ ven channel gain, where l = 1,2,…, ||ξk and pk,b(1) ≤ pk,b(2) ≤ … ≤ pk,b( ||ξk ) . The sharing strategy saves the su⁃ bcarriers those might be wasted by only transmitting the mes⁃ sages of the weak users and accommodates more users with di⁃ SIC: successive interference cancellation ▲Figure 3. A downlink NOMA model with one base station and two users. ▲Figure 4. The rate region of two⁃user SC⁃NOMA in comparison with that of OMA. User 1 is a strong user with ( ||h1 2 σ2 1 )= 100 , while user 2 is a weak user with ( ||h2 2 σ2 2)= 1 . NOMA: non⁃orthogonal multiple access OMA: orthogonal multiple access Power … … Frequency User 1 User 2 Base station SIC of user 2’s signal User 2’s signal decoding User 1’s signal decoding 7 Rateofuser1(bit/s/Hz) 1.0 Rate of user 2 (bit/s/Hz) 6 5 4 3 2 1 0 0.90.80.70.60.50.40.30.20.10 Capacity gain NOMA OMA
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS20 Special Topic verse QoS requirements, which is favorable to massive connec⁃ tivity and IoT in 5G networks. All messages of users in ξk are superimposed on subcarrier k, where the transmitted signal is given by xk =∑l = 1 ||ξk pk,b(l) Sk,b(l), (3) where Sk,b(l) and pk,b(l) denote the message and allocated pow⁃ er of user n(l) on subcarrier k, respectively. Assuming the independent and identically distributed (IID) AWGN over all subcarriers and all users for simplicity, the us⁃ er scheduling, power allocation, and the SIC decoding order on⁃ ly depend on the channel gain order. At the receiver side, the received signal of user b(l) on subcarrier k can be represented by yk,b(l) = hk,b(l) xk + v (4) = hk,b(l)∑l′ = 1 ||ξk pk,b(l′) Sk,b(l′) + v, ∀l ∈{ }1,2,…, ||ξk , (5) where v denotes the AWGN, i.e., vi ∼ CN(0,σ2 ) , and σ2 de⁃ notes the noise power. On subcarrier k, the scheduled users in ξk perform SIC to eliminate inter⁃user interference. Similar to the case of two⁃us⁃ er NOMA, the optimal SIC decoding order is in the descending channel gain order, i.e., { }b(1),b(2),…,b( )||ξk . It means that the user b(l) first decodes and subtracts the message sk,b(l′) , ∀l′ > l , in descending order from ||ξk to l + 1, and then de⁃ codes its own message sk,n(l) by treating sk,n(l′) , ∀l′ > l , as in⁃ terference. 3 Performance and Key Features of NOMA In this section, we present the performance characteristics of NOMA in existing works, and then discuss the pros and cons of NOMA schemes. 3.1 Performance of NOMA It has been shown that NOMA offers considerable perfor⁃ mance gain over OMA in terms of spectral efficiency and out⁃ age probability [25]-[27], [29]-[31]. Initially, the performance of NOMA was evaluated through simulations given perfect CSI by utilizing the proportional fairness scheduler [25], [29], frac⁃ tional transmission power allocation (FTPA) [25], and tree ⁃ search based transmission power allocation (TTPA) [30]. These works showed that the overall cell throughput, cell⁃edge user throughput, and the degrees of proportional fairness achieved by NOMA are all superior to those of OMA. In [27], the author analyzed a two ⁃ user SC ⁃ NOMA system under statistical CSI from an information theoretic perspective, where it was proved that NOMA outperforms native TDMA with high probability in terms of both the sum rate and individual rates. In [26], for a fixed power allocation, the performance of a multiuser SC⁃NO⁃ MA system in terms of outage probability and ergodic sum rates under statistical CSI was investigated in a cellular down⁃ link scenario with randomly deployed users. With the proposed asymptotic analysis, it showed that user n experiences a diver⁃ sity gain of n and NOMA is asymptotically equivalent to the op⁃ portunistic multiple access technique. Furthermore, the au⁃ thors in [32] analyzed the performance degradation of a mul⁃ tiuser SC ⁃ NOMA system on outage probability and average sum rates due to partial CSI. It showed that NOMA based on second order statistical CSI always achieves a better perfor⁃ mance than that of NOMA based on imperfect CSI, while it can achieve similar performance to the NOMA with perfect CSI in the low SNR region. In summary, most of the existing works on performance anal⁃ ysis of NOMA focused on a SC⁃NOMA system since the user scheduling in MC⁃NOMA complicates the analysis due to its combinatorial nature. A remarkable work in [31] characterized the impact of user pairing on the performance of a two⁃user SC⁃ NOMA system with fixed power allocation and cognitive radio inspired power allocation, respectively. The authors proved that, for fixed power allocation, the performance gain of NOMA over OMA increases when the difference in channel gains be⁃ tween the paired users becomes larger. However, further explo⁃ ration on performance analysis of MC⁃NOMA system should be carried out in the future since user scheduling is critical for performance of NOMA. 3.2 Pros 1) Higher spectral efficiency By exploiting the power domain for user multiplexing, NO⁃ MA systems are able to accommodate more users to cope with system overload. In contrast to allocate a subcarrier exclusive⁃ ly to a single user in OMA scheme, NOMA can utilize the spec⁃ trum more efficiently by admitting strong users into the subcar⁃ riers occupied by weak users without compromising much their performance via utilizing appropriate power allocation and SIC techniques. 2) Better utilization of heterogeneity of channel conditions As we mentioned before, NOMA schemes intentionally mul⁃ tiplex strong users with weak users to exploit the heterogeneity of channel condition. Therefore, the performance gain of NO⁃ MA over OMA is larger when channel gains of the multiplexed users become more distinctive [31]. 3) Enhanced user fairness By relaxing the orthogonal constraint of OMA, NOMA en⁃ ables a more flexible management of radio resources and offers an efficient way to enhance user fairness via appropriate re⁃ source allocation [23]. 4) Applicability to diverse QoS requirements NOMA is able to accommodate more users with different types of QoS requests on the same subcarrier. Therefore, NO⁃ MA is a good candidate to support IoT which connects a great A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo
October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 21 Special Topic number of devices and sensors requiring distinctive targeted rates. 3.3 Cons 1) The BS needs to know the perfect channel state informa⁃ tion (CSI) to arrange the SIC decoding order, which increases the CSI feedback overhead. 2) The SIC process introduces a higher computational com⁃ plexity and delay at the receiver side, especially for multicarri⁃ er and multiuser systems. 3) The strong users have to know the power allocation of the weaker users in order to perform SIC, which also increases the system signalling overhead. 4) Allocating more power to the weak users, who are general⁃ ly in the cell⁃edge, will introduce more inter⁃cell interferences into the whole system. 4 Design of NOMA Schemes Due to the remarkable performance gain of NOMA over con⁃ ventional OMA, a lot of works on design of NOMA schemes have been proposed in literatures. In this section, we present the existing works on resource allocation of NOMA and MIMO⁃ NOMA, and then briefly introduce other works associated with NOMA. 4.1 Resource Allocation Resource allocation has received significant attention since it is critical to improve the performance of NOMA. However, optimal resource allocation is very challenging for MC⁃NOMA systems, since user scheduling and power allocation couple with each other severely. Some initial works on resource alloca⁃ tion in [25], [29], [30] have been reported, but they are far from optimal. In [33], [34], the authors studied a two⁃user MC⁃NO⁃ MA system by minimizing the number of subcarriers assigned under the constraints of maximum allowed transmit power and requested data rates, and further introduced a hybrid orthogo⁃ nal ⁃ nonorthogonal scheme. Furthermore, the authors in [21] studied a joint power and subcarrier allocation problem for a two ⁃ user MC ⁃ NOMA system. They proposed an optimal scheme and a suboptimal scheme with close⁃to⁃optimal perfor⁃ mance based on monotonic optimization and difference of con⁃ vex function programming, respectively. Besides, there are also several works on resource allocation for multiuser MC⁃NOMA systems. In [35], the authors formulat⁃ ed the resource allocation problem to maximize the sum rate, which is a non⁃convex optimization problem due to the binary constraint and the existence of the interference term in the ob⁃ jective function. Interestingly, they proposed a suboptimal solu⁃ tion by employing matching theory and water⁃filling power allo⁃ cation. In [36], the authors presented a systematic approach for NOMA resource allocation from a mathematical optimization point of view. They formulated the joint power and channel al⁃ location problem of a downlink multiuser MC⁃NOMA system, and proved its NP⁃hardness based on [37] via defining a spe⁃ cial user. Furthermore, they proposed a competitive subopti⁃ mal algorithm based on Lagrangian duality and dynamic pro⁃ gramming, which significantly outperforms OFDMA as well as NOMA with FTPA. Most of works aforementioned focus on the optimal resource allocation for maximizing the sum rate. However, fairness is an⁃ other objective to optimize for resource allocation of NOMA. Proportional fairness (PF) has been adopted as a metric to bal⁃ ance the transmission efficiency and user fairness in many works [38], [39]. In [40], the authors proposed a user pairing and power allocation scheme for downlink two⁃user MC⁃NO⁃ MA based on the PF objective. A prerequisite for user pairing was given and a closed⁃form optimal solution for power alloca⁃ tion was derived. Apart from PF, max⁃min or min⁃max methods are usually adopted to achieve user fairness. Given a preset user group, the authors in [23] studied the power allocation problem from a fairness standpoint by maxi⁃ mizing the minimum achievable user rate with instantaneous CSI and minimizing the maximum outage probability with aver⁃ age CSI. Although the resulting problems are non⁃convex, sim⁃ ple low⁃complexity algorithms were developed to provide close⁃ to⁃optimal solutions. Similarly, another paper [41] studied the outage balancing problem of a downlink multiuser MC⁃NOMA system to maximize the minimum weighted success probability with and without user grouping. Joint power allocation and de⁃ coding order selection solutions were given, and the inter ⁃ group power and resource allocation solutions were also provid⁃ ed in the paper. In summary, many existing works focus on the resource allo⁃ cation for NOMA systems under perfect CSI at the transmitter side. However, there are only few works on the joint user scheduling and power allocation problem for MC⁃NOMA sys⁃ tems under imperfect CSI, not to mention the SIC decoding or⁃ der selection problem. In fact, under imperfect CSI, the SIC de⁃ coding order cannot be determined by channel gain order, and some other metrics, such as distance, priority, and target rates, are potential criteria to decide the SIC decoding order. 4.2 MIMO􀆼NOMA The application of MIMO techniques to NOMA systems is important for enhancing the performance gains of NOMA. Therefore, MIMO⁃NOMA is another hot topic that has been re⁃ searched, where the BS and users are equipped with multiple antennas, and multiple users in the same beam are multi⁃ plexed on power domain. Fig. 5 illustrates a simple MIMO⁃NO⁃ MA system with one base station and four users. Initially, the concept of MIMO ⁃ NOMA was proposed in [30], [42], [43], which demonstrated that MIMO ⁃ NOMA outperforms conven⁃ tional MIMO OMA. The authors in [44] proposed a two⁃user MIMO⁃NOMA scheme with a clustering and power allocation algorithm, where the correlation and channel gain difference A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS22 Special Topic were taken into consideration to reduce intra ⁃ beam interfer⁃ ence and inter⁃beam interference simultaneously. In [45], the authors proposed a minimum power multicast beamforming scheme and applied to two⁃user NOMA systems for multi⁃reso⁃ lution broadcasting. The proposed two ⁃ stage beamforming method outperforms the zero ⁃ forcing beamforming scheme in [44]. The design of precoding and detection algorithms also re⁃ ceived considerable attention since they are the key to elimi⁃ nate or reduce inter⁃cluster interference. The authors in [46] studied the ergodic sum capacity maximization problem of a two⁃user MIMO⁃NOMA system under statistical CSI with the total power constraint and minimum rate constraint for the weak user. This paper derived the optimal input covariance ma⁃ trix, and proposed the optimal power allocation scheme as well as a low complexity suboptimal solution. Furthermore, in [47], the authors studied the sum rate optimization problem of two⁃ user MIMO ⁃ NOMA under perfect CSI with the same con⁃ straints, while different precoders were assigned to different us⁃ ers. The optimal precode covariance matrix was derived by uti⁃ lizing the duality between uplink and downlink, and a low com⁃ plexity suboptimal solution based on singular value decomposi⁃ tion (SVD) was also provided. In [48], the authors proposed a new design of precoding and detection matrices for a downlink multiuser MIMO⁃NOMA system, then analyzed the impact of user pairing as well as power allocation on the sum rate and outage probability of MIMO ⁃ NOMA system. Furthermore, in [49], a transmission framework based on signal alignment was proposed for downlink and uplink two⁃user MIMO⁃NOMA sys⁃ tems. The authors in [20] studied the sum rate maximization problem of a downlink multiuser multiple⁃input single⁃output (MISO) NOMA system. The MISO NOMA transmission outper⁃ forms conventional OMA schemes, particularly when the trans⁃ mit SNR is low, and the number of users is greater than the number of BS antennas. Recently, a multiuser MIMO⁃NOMA scheme based on limited feedback was proposed and analyzed in [50]. In summary, most of the existing works on MIMO⁃NOMA fo⁃ cused on design of precoding and detection algorithms, and their performance analyses. However, user scheduling and power allocation were rarely discussed in the spatial domain, which play important roles in improving the spatial efficiency of MIMO⁃NOMA. 4.3 Other Works on NOMA In addition to the above two aspects, there are many other works associated NOMA. We will not discuss further in detail due to the limited space. Compared to downlink NOMA, up⁃ link NOMA was also studied in several works [51]-[57]. More⁃ over, asynchronous NOMA has also been investigated in up⁃ link scenarios [58], [59]. Cooperative NOMA, where strong us⁃ ers serve as relays for weak users, was studied in [60], [61]. In addition, several works on NOMA combined with other tech⁃ niques were also reported, such as energy harvesting [62], [63], cognitive radio networks [64], visible light communication [65], and physical layer security [66]. 5 Research Challenges As discussed above, NOMA can be employed to improve the spectral efficiency, user fairness, as well as to support massive connections with diverse QoS requirements. Based on our over⁃ view of existing works on NOMA and its potential applications in practical systems, we present the research challenges of NO⁃ MA in the following three aspects. 5.1 Resource Allocation under Imperfect CSI Most of existing works on resource allocation of NOMA are based on the assumption of perfect CSI at the transmitter side, which is difficult to obtain in practice due to either the estima⁃ tion error or the feedback delay. Therefore, it is nature to inves⁃ tigate how CSI error affects the performance of NOMA and to consider robust resource allocation under imperfect CSI. Since NOMA is expected to offer lower latency in order to support de⁃ lay⁃sensitive applications in 5G, one promising solution is the outage⁃based robust approach for designing the resource allo⁃ cation of NOMA. In this direction, the SIC decoding order un⁃ der imperfect CSI is still an open problem. Furthermore, it is important to study the joint optimization of power allocation, user scheduling and SIC decoding order selection of NOMA under imperfect CSI. 5.2 Cooperative NOMA A key feature of NOMA is that the strong users have prior in⁃ formation of the weak users, which has not been fully exploited in existing works. In cooperative NOMA, the strong users can serve as relays for the weak users, which has the potential to utilize the spatial DoF even for users with a single antenna. Some preliminary works showed that cooperative NOMA can achieve the maximum diversity gain for all the users [60], [61]. It is important to study the optimal resource allocation for coop⁃ A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo SIC: successive interference cancellation ▲Figure 5. A downlink MIMO⁃NOMA model with one base station and four users. Base station User 3 User 1 User 2 SIC of user 2’s signal User 1’s signal decoding User 2’s signal decoding SIC of user 3’s signal User 3’s signal decoding User 4’s signal decoding User 4 Power Power … … Frequency Frequency … …
October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 23 Special Topic erative NOMA. Besides, distributed beamforming can be em⁃ ployed in cooperative NOMA to harvest the spatial DoF with⁃ out much signalling overhead. Considering that cooperative NOMA will introduce more complexity and extra delay into sys⁃ tems, it is important to investigate the tradeoffs among the sys⁃ tem performance, complexity, and delay. 5.3 QoS􀆼Based NOMA As we mentioned before, NOMA has great potential to sup⁃ port diverse QoS requirements. The heterogeneity of QoS re⁃ quirements might in turn facilitate the power allocation and us⁃ er scheduling of NOMA, which is also an interesting topic to explore in the future. For example, users in NOMA systems can be categorized according to their QoS requirements, in⁃ stead of their channel conditions, which offers two following benefits. One is that the SIC decoding order, power allocation, and user scheduling can be designed more appropriately to meet the users QoS requests. The other is to make NOMA com⁃ munications more general, e.g., applicable to scenarios in which users channel conditions are the same. 6 Conclusions In this article, a promising multiple access technology for 5G networks, NOMA, is discussed. A two ⁃ user SC ⁃ NOMA scheme and a multiuser MC ⁃ NOMA scheme were presented and discussed to illustrate the basic concepts and principles of NOMA. A literature review about performance analyses of NO⁃ MA, resource allocation for NOMA, and MIMO⁃NOMA was dis⁃ cussed. Furthermore, we presented the key features and poten⁃ tial research challenges of NOMA. A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo References [1] J. Andrews, S. Buzzi, W. Choi, et al.,“What will 5G be?”IEEE Journal on Se⁃ lected Areas in Communications, vol. 32, no. 6, pp. 1065-1082, Jun. 2014. doi: 10.1109/JSAC.2014.2328098. [2] G. Wunder, P. Jung, M. Kasparick, et al.,“5GNOW: non⁃orthogonal, asynchro⁃ nous waveforms for future mobile applications,”IEEE Communications Maga⁃ zine, vol. 52, no. 2, pp. 97-105, Feb. 2014. doi: 10.1109/MCOM.2014.6736749. [3] T. Marzetta,“Noncooperative cellular wireless with unlimited numbers of base station antennas,”IEEE Transactions on Wireless Communications, vol. 9, no. 11, pp. 3590-3600, Nov. 2010. doi: 10.1109/TWC.2010.092810.091092. [4] J. Zhu, D. W. K. Ng, N. Wang, R. Schober, and V. K. Bhargava,“Analysis and design of secure massive MIMO systems in the presence of hardware impair⁃ ments,”arXiv preprint arXiv:1602.08534, 2016. [5] Z. Pi and F. Khan,“An introduction to millimeter⁃wave mobile broadband sys⁃ tems,”IEEE Communications Magazine, vol. 49, no. 6, pp. 101-107, Jun. 2011. doi: 10.1109/MCOM.2011.5783993. [6] T. Rappaport, S. Sun, R. Mayzus, et al.,“Millimeter wave mobile communica⁃ tions for 5G cellular: It will work!”IEEE Access, vol. 1, pp. 335-349, May 2013. doi: 10.1109/ACCESS.2013.2260813. [7] J. Andrews, H. Claussen, M. Dohler, S. Rangan, and M. Reed,“Femtocells: past, present, and future,”IEEE Journal on Selected Areas in Communications, vol. 30, no. 3, pp. 497-508, Apr. 2012. doi: 10.1109/JSAC.2012.120401. [8] J. Andrews,“Seven ways that HetNets are a cellular paradigm shift,”IEEE Com⁃ munications Magazine, vol. 51, no. 3, pp. 136-144, Mar. 2013. doi: 10.1109/ MCOM.2013.6476878. [9] D. Ramasamy, R. Ganti, and U. Madhow,“On the capacity of picocellular net⁃ works,”in IEEE International Symposium on Information Theory, Istanbul, Tur⁃ key, Jul. 2013, pp. 241-245. doi: 10.1109/ISIT.2013.6620224. [10] D. W. K. Ng, E. S. Lo, and R. Schober,“Robust beamforming for secure com⁃ munication in systems with wireless information and power transfer,”IEEE Transactions on Wireless Communications, vol. 13, no. 8, pp. 4599-4615, Aug. 2014. doi: 10.1109/TWC.2014.2314654. [11] D. W. K. Ng, E. S. Lo, and R. Schober,“Energy⁃efficient resource allocation in OFDMA systems with large numbers of base station antennas,”IEEE Transac⁃ tions on Wireless Communications, vol. 11, no. 9, pp. 3292-3304, Sept. 2012. doi: 10.1109/TWC.2012.072512.111850. [12] D. W. K. Ng, E. S. Lo, and R. Schober,“Wireless information and power trans⁃ fer: Energy efficiency optimization in OFDMA systems,”IEEE Transactions on Wireless Communications, vol. 12, no. 12, pp. 6352- 6370, Dec. 2013. doi: 10.1109/TWC.2013.103113.130470. [13] L. Dai, B. Wang, Y. Yuan, S. Han, I. Chih⁃Lin, and Z. Wang,“Non⁃orthogonal multiple access for 5G: solutions, challenges, opportunities, and future re⁃ search trends,”IEEE Communications Magazine, vol. 53, no. 9, pp. 74- 81, Sept. 2015. doi: 10.1109/MCOM.2015.7263349. [14] J. M. Meredith,“Study on downlink multiuser superposition transmission (MUST) for LTE (Release 13),”3GPP Tech. Rep. TR 36.859, Dec. 2015. [15] R. Hoshyar, F. P. Wathan, and R. Tafazolli,“Novel low⁃density signature for synchronous CDMA systems over AWGN channel,”IEEE Transactions on Sig⁃ nal Processing, vol. 56, no. 4, pp. 1616- 1626, Apr. 2008. doi: 10.1109/ TSP.2007.909320. [16] R. Hoshyar, R. Razavi, and M. Al⁃Imari,“LDS⁃OFDM an efficient multiple ac⁃ cess technique,”in IEEE Vehicular Technology Conference, Taiwan, China, May 2010, pp. 1-5. doi: 10.1109/VETECS.2010.5493941. [17] R. Razavi, M. Al⁃Imari, M. A. Imran, R. Hoshyar, and D. Chen,“On receiver design for uplink low density signature OFDM (LDS⁃OFDM),”IEEE Transac⁃ tions on Communications, vol. 60, no. 11, pp. 3499- 3508, Nov. 2012. doi: 10.1109/TCOMM.2012.082812.110284. [18] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE Personal, Indoor and Mobile Radio Communications Symposium, London, United King⁃ dom, Sept. 2013, pp. 332-336. doi: 10.1109/PIMRC.2013.6666156. [19] X. Dai, S. Chen, S. Sun, et al.,“Successive interference cancelation amenable multiple access (SAMA) for future wireless communications,”in IEEE Interna⁃ tional Conference on Communication Systems, Macau, China, Nov. 2014, pp. 222-226. doi: 10.1109/PIMRC.2013.6666156. [20] M. F. Hanif, Z. Ding, T. Ratnarajah, and G. K. Karagiannidis,“A minorization⁃ maximization method for optimizing sum rate in the downlink of non⁃orthogonal multiple access systems,”IEEE Transactions on Signal Processing, vol. 64, no. 1, pp. 76-88, Jan. 2016. doi: 10.1109/TSP.2015.2480042. [21] Y. Sun, D. W. K. Ng, Z. Ding, and R. Schober,“Optimal joint power and sub⁃ carrier allocation for MC⁃NOMA systems,”arXiv preprint, arXiv:1603.08132, 2016. [22] Q. Sun, S. Han, C.⁃L. I, and Z. Pan,“Energy efficiency optimization for fading MIMO non⁃orthogonal multiple access systems,”in IEEE International Confer⁃ ence on Communications, London, United Kingdom, Jun. 2015, pp. 2668- 2673. doi: 10.1109/ICC.2015.7248728. [23] S. Timotheou and I. Krikidis,“Fairness for non⁃orthogonal multiple access in 5G systems,”IEEE Signal Processing Letters, vol. 22, no. 10, pp. 1647-1651, Oct. 2015. doi: 10.1109/LSP.2015.2417119. [24] A. Benjebbour, A. Li, Y. Saito, Y. Kishiyama, A. Harada, and T. Nakamura, “System⁃level performance of downlink NOMA for future LTE enhancements,” in IEEE Global Communications Conference, Atlanta, USA, Dec. 2013, pp. 66- 70. doi: 10.1109/GLOCOMW.2013.6824963. [25] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura,“System⁃level perfor⁃ mance evaluation of downlink non ⁃ orthogonal multiple access (NOMA),”in IEEE Personal, Indoor and Mobile Radio Communications Symposium, London, United Kingdom, Sept. 2013, pp. 611- 615. doi: 10.1109/GLO⁃ COMW.2013.6824963. [26] Z. Ding, Z. Yang, P. Fan, and H. Poor,“On the performance of non⁃orthogonal multiple access in 5G systems with randomly deployed users,”IEEE Signal Processing Letters, vol. 21, no. 12, pp. 1501-1505, Dec. 2014. doi: 10.1109/ LSP.2014.2343971. [27] P. Xu, Z. Ding, X. Dai, and H. V. Poor,“A new evaluation criterion for non⁃or⁃ thogonal multiple access in 5G software defined networks,”IEEE Access, vol. 3, pp. 1633-1639, Sept. 2015. doi: 10.1109/ACCESS.2015.2480117. [28] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cam⁃
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS24 Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo bridge, United Kingdom: Cambridge University Press, 2005. [29] N. Otao, Y. Kishiyama, and K. Higuchi,“Performance of non⁃orthogonal access with SIC in cellular downlink using proportional fair ⁃ based resource alloca⁃ tion,”in IEEE International Symposium on Wireless Communication Systems, Paris, France, Aug. 2012, pp. 476-480. doi: 10.1109/ISWCS.2012.6328413. [30] Y. Saito, Y. Kishiyama, A. Benjebbour, et al.,“Non⁃orthogonal multiple access (NOMA) for cellular future radio access,”in IEEE Vehicular Technology Con⁃ ference, Dresden, Germany, Jun. 2013, pp. 1- 5. doi: 10.1109/VTC⁃ Spring.2013.6692652. [31] Z. Ding, P. Fan, and V. Poor,“Impact of user pairing on 5G non⁃orthogonal multiple access downlink transmissions,”IEEE Transactions on Vehicular Technology, vol. 65, no. 8, pp. 6010- 6023, 2016. doi: 10.1109/ TVT.2015.2480766. [32] Z. Yang, Z. Ding, P. Fan, and G. K. Karagiannidis,“On the performance of non⁃ orthogonal multiple access systems with partial channel information,”IEEE Transactions on Communications, vol. 64, no. 2, pp. 654-667, Feb. 2016. doi: 10.1109/TCOMM.2015.2511078. [33] M. R. Hojeij, J. Farah, C. Nour, and C. Douillard,“Resource allocation in downlink non⁃orthogonal multiple access (NOMA) for future radio access,”in IEEE Vehicular Technology Conference, Glasgow, Scotland, May 2015, pp. 1- 6. doi: 10.1109/VTCSpring.2015.7146056. [34] M. R. Hojeij, J. Farah, C. Nour, and C. Douillard,“New optimal and subopti⁃ mal resource allocation techniques for downlink non⁃orthogonal multiple ac⁃ cess,”Wireless Personal Communications, vol. 87, no. 3, pp. 837- 867, May 2015. doi:10.1007/s11277⁃015⁃2629⁃2. [35] B. Di, S. Bayat, L. Song, and Y. Li,“Radio resource allocation for downlink non ⁃orthogonal multiple access (NOMA)networks using matching theory,”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1-6. doi: 10.1109/GLOCOM.2015.7417643. [36] L. Lei, D. Yuan, C. K. Ho, and S. Sun,“Joint optimization of power and chan⁃ nel allocation with non⁃orthogonal multiple access for 5G cellular systems,”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1- 6. doi: 10.1109/GLOCOM.2015.7417761. [37] Y.⁃F. Liu and Y.⁃H. Dai,“On the complexity of joint subcarrier and power allo⁃ cation for multi⁃user OFDMA systems,”IEEE Transactions on Signal Process⁃ ing, vol. 62, no. 3, pp. 583-596, Feb. 2014. doi: 10.1109/TSP.2013.2293130. [38] H. Kim, K. Kim, Y. Han, and S. Yun,“A proportional fair scheduling for multi⁃ carrier transmission systems,”in IEEE Vehicular Technology Conference, Los Angeles, USA, Sept 2004, vol. 1, pp. 409- 413. doi: 10.1109/ VETECF.2004.1400034. [39] C. Wengerter, J. Ohlhorst, and A. G. E. von Elbwart,“Fairness and throughput analysis for generalized proportional fair frequency scheduling in OFDMA,”in IEEE Vehicular Technology Conference, Stockholm, Sweden, May 2005, vol. 3, pp. 1903-1907. doi: 10.1109/VETECS.2005.1543653. [40] F. Liu, P. Mahonen, and M. Petrova,“Proportional fairness⁃based user pairing and power allocation for non⁃orthogonal multiple access,”in IEEE Personal, In⁃ door and Mobile Radio Communications Symposium, Hong Kong, China, Aug. 2015, pp. 1127-1131. doi: 10.1109/PIMRC.2015.7343467. [41] S. Shi, L. Yang, and H. Zhu,“Outage balancing in downlink non⁃orthogonal multiple access with statistical channel state information,”IEEE Transactions on Wireless Communications, vol. 15, no. 7, pp. 4718-4731, Jul. 2016. doi: 10.1109/TWC.2016.2544922. [42] Y. Lan, A. Benjebboiu, X. Chen, A. Li, and H. Jiang,“Considerations on down⁃ link non⁃orthogonal multiple access (NOMA) combined with closed⁃loop SU⁃ MIMO,”in IEEE International Conference on Signal Processing and Communi⁃ cation System, Gold Coast, Australia, Dec. 2014, pp. 1- 5. doi: 10.1109/IC⁃ SPCS.2014.7021086. [43] X. Chen, A. Benjebbour, Y. Lan, A. Li, and H. Jiang,“Impact of rank optimiza⁃ tion on downlink non⁃orthogonal multiple access (NOMA) with SU⁃MIMO,”in IEEE International Conference on Communication Systems, Macau, China, Nov. 2014, pp. 233-237. doi: 10.1109/ICCS.2014.7024800. [44] B. Kim, S. Lim, H. Kim, et al.,“Non⁃orthogonal multiple access in a downlink multiuser beamforming system,”in IEEE Military Communications Conference, San Diego, USA, Nov. 2013, pp. 1278-1283. doi: 10.1109/MILCOM.2013.218. [45] J. Choi,“Minimum power multicast beamforming with superposition coding for multiresolution broadcast and application to NOMA systems,”IEEE Transac⁃ tions on Communications, vol. 63, no. 3, pp. 791-800, Mar. 2015. doi: 10.1109/ TCOMM.2015.2394393. [46] Q. Sun, S. Han, C.⁃L. I, and Z. Pan,“On the ergodic capacity of MIMO NOMA systems,”IEEE Wireless Communications Letters, vol. 4, no. 4, pp. 405-408, Aug. 2015. doi: 10.1109/LWC.2015.2426709. [47] Q. Sun, S. Han, Z. Xu, S. Wang, C.⁃L. I, and Z. Pan,“Sum rate optimization for MIMO non⁃orthogonal multiple access systems,”in IEEE Wireless Communica⁃ tions and Networking Conference, New Orleans, USA, Mar. 2015, pp. 747-752. doi: 10.1109/WCNC.2015.7127563. [48] Z. Ding, F. Adachi, and H. V. Poor,“The application of MIMO to non⁃orthogo⁃ nal multiple access,”IEEE Transactions on Wireless Communications, vol. 15, no. 1, pp. 537-552, Jan. 2016. doi: 10.1109/TWC.2015.2475746. [49] Z. Ding, R. Schober, and H. V. Poor,“A general MIMO framework for NOMA downlink and uplink transmission based on signal alignment,”IEEE Transac⁃ tions on Wireless Communications, vol. 15, no. 6, pp. 4438-4454, Jun. 2016. doi: 10.1109/TWC.2016.2542066. [50] Z. Ding and H. V. Poor,“Design of massive⁃MIMO⁃NOMA with limited feed⁃ back,”IEEE Signal Processing Letters, vol. 23, no. 5, pp. 629-633, May 2016. doi: 10.1109/LSP.2016.2543025. [51] T. Takeda and K. Higuchi,“Enhanced user fairness using non⁃orthogonal ac⁃ cess with SIC in cellular uplink,”in IEEE Vehicular Technology Conference, San Francisco, USA, Sept. 2011, pp. 1- 5. doi: 10.1109/ VETECF.2011.6093272. [52] M. Al⁃Imari, P. Xiao, M. A. Imran, and R. Tafazolli,“Uplink non⁃orthogonal multiple access for 5G wireless networks,”in IEEE International Symposium on Wireless Communications Systems, Barcelona, Spain, Aug. 2014, pp. 781- 785. doi: 10.1109/ISWCS.2014.6933459. [53] S. Chen, K. Peng, and H. Jin,“A suboptimal scheme for uplink NOMA in 5G systems,”in IEEE International Wireless Communications and Mobile Comput⁃ ing Conference, Dubrovnik, Croatia, Aug. 2015, pp. 1429-1434. doi: 10.1109/ IWCMC.2015.7289292. [54] Y. Chen, J. Schaepperle, and T. Wild,“Comparing IDMA and NOMA with su⁃ perimposed pilots based channel estimation in uplink,”in IEEE Personal, In⁃ door and Mobile Radio Communications Symposium, Aug. 2015, pp. 89- 94. doi: 10.1109/PIMRC.2015.7343274. [55] M. Al⁃Imari, P. Xiao, and M. A. Imran,“Receiver and resource allocation opti⁃ mization for uplink NOMA in 5G wireless networks,”in IEEE International Symposium on Wireless Communication Systems, Brussels, Belgium, Aug. 2015. doi: 10.1109/ISWCS.2015.7454317. [56] B. Kim, W. Chung, S. Lim, et al.,“Uplink NOMA with multi ⁃ antenna,”in IEEE Vehicular Technology Conference, Glasgow, Scotland, May 2015, pp. 1- 5. doi: 10.1109/VTCSpring.2015.7146149. [57] N. Zhang, J. Wang, G. Kang, and Y. Liu,“Uplink non⁃orthogonal multiple ac⁃ cess in 5G systems,”IEEE Communications Letters, vol. 20, no. 3, pp. 458- 461, Mar. 2016. doi: 10.1109/LCOMM.2016.2521374. [58] H. Haci, H. Zhu, and J. Wang,“A novel interference cancellation technique for non ⁃ orthogonal multiple access (NOMA),”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1- 6. doi: 10.1109/GLO⁃ COM.2015.7417128. [59] H. Haci,“Non⁃orthogonal multiple access (NOMA) with asynchronous interfer⁃ ence cancellation,”Ph.D. dissertation, Dept. Electron. Eng., Univ. of Kent, Kent, England, Mar. 2015. [60] Z. Ding, M. Peng, and H. Poor,“Cooperative non⁃orthogonal multiple access in 5G systems,”IEEE Communications Letters, vol. 19, no. 8, pp. 1462- 1465, Aug. 2015. doi: 10.1109/LCOMM.2015.2441064. [61] Z. Ding, H. Dai, and H. V. Poor,“Relay selection for cooperative NOMA,” IEEE Wireless Communications Letters, vol. 5, no. 4, pp. 416-419, Jun. 2016. doi: 10.1109/LWC.2016.2574709. [62] Y. Liu, Z. Ding, M. Eikashlan, and H. Poor,“Cooperative non⁃orthogonal multi⁃ ple access in 5G systems with SWIPT,”in European Signal Processing Confer⁃ ence, Nice, France, Aug. 2015, pp. 1999- 2003. doi: 10.1109/EUSIP⁃ CO.2015.7362734. [63] P. D. Diamantoulakis, K. N. Pappi, Z. Ding, and G. K. Karagiannidis,“Optimal design of non⁃orthogonal multiple access with wireless power transfer,”arXiv preprint arXiv:1511.01291, 2015. [64] Y. Liu, Z. Ding, M. Elkashlan, and J. Yuan,“Non⁃orthogonal multiple access in large⁃scale underlay cognitive radio networks,”IEEE Transactions on Vehicu⁃ lar Technology, vol. PP, no. 99, pp. 1- 1, Feb. 2016. doi: 10.1109/ TVT.2016.2524694. [65] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat,“Non⁃or⁃ thogonal multiple access for visible light communications,”IEEE Photonics Technology Letters, vol. 28, no. 1, pp. 51- 54, Jan. 2016. doi: 10.1109/ LPT.2015.2479600. [66] Y. Zhang, H. M. Wang, Q. Yang, and Z. Ding,“Secrecy sum rate maximization in non⁃orthogonal multiple access,”IEEE Communications Letters, vol. 20, no. 5, pp. 930-933, Mar. 2016. doi: 10.1109/LCOMM.2016.2539162. Manuscript received: 2016⁃08⁃15
October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 25 Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo WEI Zhiqiang (zhiqiang.wei@unsw.edu.au) received the BE degree from Northwes⁃ tern Polytechnical University, China in 2012. He is currently pursuing the PhD de⁃ gree in Wireless Communications Laboratory, University of New South Wales, Aus⁃ tralia. His research interests include non⁃orthogonal multiple access and resource allocation. YUAN Jinhong (j.yuan@unsw.edu.au) received the BE and PhD degrees in electro⁃ nics engineering from Beijing Institute of Technology, China in 1991 and 1997, re⁃ spectively. From 1997 to 1999, he was a reasearch fellow with the School of Electri⁃ cal Engineering, University of Sydney, Australia. In 2000, he joined the School of Electrical Engineering and Telecommunications, University of New South Wales, Australia, where he is currently a professor of telecommunications. He has authored two books, three book chapters, more than 200 papers in telecommunications jour⁃ nals and conference proceedings, and 40 industrial reports. His research interests include error control coding and information theory, communication theory, and wireless communications. He is a co⁃inventor of one patent on MIMO systems and two patents on low⁃density parity⁃check codes. He is currently serving as an associ⁃ ate editor for the IEEE TCOM. He served as the IEEE NSW Chair of Joint Commu⁃ nications/Signal Processions/Ocean Engineering Chapter from 2011 to 2014. He was the co⁃recipient of three best paper awards and one best poster award, includ⁃ ing the Best Paper Award from the IEEE Wireless Communications and Networking Conference, Cancun, Mexico in 2011, and the Best Paper Award from the IEEE In⁃ ternational Symposium on Wireless Communications Systems, Trondheim, Norway in 2007. Derrick Wing Kwan Ng (w.k.ng@unsw.edu.au) received the bachelor degree with first class honors and the Master of Philosophy (M.Phil.) degree in electronic engi⁃ neering from the Hong Kong University of Science and Technology (HKUST) in 2006 and 2008, respectively. He received his PhD degree from the University of British Columbia (UBC) in 2012. He was a senior postdoctoral fellow at the Institute for Digital Communications, University of Erlangen ⁃ Nuremberg, Germany. He is now working as a lecturer at the University of New South Wales, Australia. Dr. Ng has published more than 80 journal and conference papers and his publications have been cited over 2000 times in Google Scholar with an h⁃index of 20. Dr. Ng is currently an editor of IEEE Communications Letters and IEEE Transactions on Green Communications and Networking. He served as a Co⁃Chair for the Wireless Access Track of 2014 IEEE 80th Vehicular Technology Conference and 2016 IEEE GlobeCom Workshop on Wireless Energy Harvesting. He was also a co⁃organizer and guest editor of the special issue on Energy Harvesting Wireless Communica⁃ tions in EURASIP Journal on Wireless Communications and Networking in 2014. Maged Elkashlan (maged.elkashlan@qmul.ac.uk) received the PhD degree in electr⁃ ical engineering from the University of British Columbia, Canada in 2006. From 2007 to 2011, he was with the Wireless and Networking Technologies Laboratory, Commonwealth Scientific and Industrial Research Organization, Australia. During this time, he held an adjunct appointment with the University of Technology Syd⁃ ney, Australia. In 2011, he joined the School of Electronic Engineering and Comput⁃ er Science, Queen Mary University of London, U.K. He currently holds visiting fac⁃ ulty appointments with the University of New South Wales, Australia, and the Bei⁃ jing University of Posts and Telecommunications, China. His research interests fall into the broad areas of communication theory, wireless communications, and statisti⁃ cal signal processing for distributed data processing, heterogeneous networks, and massive MIMO. Dr. Elkashlan received the best paper award at the IEEE Interna⁃ tional Conference on Communications in 2014, the International Conference on Communications and Networking in China in 2014, and the IEEE Vehicular Tech⁃ nology Conference in 2013. He also received the Exemplary Reviewer Certificate of the IEEE CL in 2012. He serves as an editor of IEEE TWC, IEEE TVT, and IEEE CL. He also serves as a lead guest editor of the Special Issue on Green Media: The Future of Wireless Multimedia Networks of the IEEE Wireless Communications Mag⁃ azine and the Special Issue on Millimeter Wave Communications for 5G of the IEEE Communications Magazine, and a guest editor of the Special Issue on Energy Har⁃ vesting Communications of the IEEE Communications Magazine and the Special Is⁃ sue on Location Awareness for Radios and Networks of the IEEE JSAC. DING Zhiguo (z.ding@lancaster.ac.uk) received his BEng from the Beijing Univers⁃ ity of Posts and Telecommunications, China in 2000, and the PhD degree from Im⁃ perial College London, U.K. in 2005. From Jul. 2005 to Aug. 2014, he was working in Queen’s University Belfast, Imperial College and Newcastle University. Since Sept. 2014, he has been with Lancaster University as a Chair Professor in Signal Processing. From Sept. 2012 to Sept. 2017, he has also been an academic visitor in Princeton University. Dr Ding’s research interests are 5G networks, game theory, cooperative and energy harvesting networks and statistical signal processing. He is serving as an editor for IEEE TCOM, IEEE TVT, IEEE WCL, and IEEE CL. He was the TPC Co⁃Chair for ICWMMN2015, and Symposium Chair for ICNC 2016 and WOCC 2015. He received the best paper award in ICWOC 2009 and WCSP 2015, IEEE CL Exemplary Reviewer 2012, and the EU Marie Curie Fellowship 2012 ⁃ 2014. BiographiesBiographies
Unified Framework Towards Flexible MultipleUnified Framework Towards Flexible Multiple Access Schemes forAccess Schemes for 55GG SUN Qi, WANG Sen, HAN Shuangfeng, and Chih􀆼Lin I (China Mobile Research Institute, Beijing 100032, China) Abstract Non⁃orthogonal multiple access (NOMA) schemes have achieved great attention recently and been considered as a crucial compo⁃ nent for 5G wireless networks since they can efficiently enhance the spectrum efficiency, support massive connections and poten⁃ tially reduce access latency via grant free access. In this paper, we introduce the candidate NOMA solutions in 5G networks, com⁃ paring the principles, key features, application scenarios, transmitters and receivers, etc. In addition, a unified framework of these multiple access schemes are proposed to improve resource utilization, reduce the cost and support the flexible adaptation of multi⁃ ple access schemes. Further, flexible multiple access schemes in 5G systems are discussed. They can support diverse deployment scenarios and traffic requirements in 5G. Challenges and future research directions are also highlighted to shed some lights for the standardization in 5G. 5G; non⁃orthogonal multiple access; unified framework; flexible multiple access Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 004 http://www.cnki.net/kcms/detail/34.1294.TN.20161018.1014.002.html, published online October 18, 2016 1 Introduction orldwide initiatives on the 5th generation (5G) wireless communication have been extensive⁃ ly carried out, starting with an investigation on user demands, scenarios, key performance indicators (KPIs) and enabling technologies. A global consen⁃ sus is first forming that 5G network will be able to sustainably support 1000⁃fold mobile data traffic growth, improve energy efficiency (EE) and cost efficiency by over 100 times, provide fiber link access data rates and“zero”latency user experi⁃ ence, and be capable of connecting 100 billion devices and ca⁃ pable of delivering a consistent experience across a variety of scenarios including the cases of ultra⁃high traffic volume densi⁃ ty, ultra⁃high connection density and ultra⁃high mobility [1]. Three typical usage scenarios of 5G are also identified: en⁃ hanced mobile broadband (eMBB), massive machine type com⁃ munication (mMTC) and ultra ⁃ reliable low latency machine type communication (URLLC), targeting different 5G capabili⁃ ties. Beyond that, the standardization organizations, e.g. 3GPP has started the new research on 5G, studying the new access technology to meet a broad range of use cases. Multiple access schemes, the most fundamental aspect of the physical layer, to a large extent, are considered as the de⁃ fining technical feature of each wireless communication gener⁃ ation and have continually evolved in each cellular generation from frequency division multiple access (FDMA), time division multiple access (TDMA) in 1G and 2G to code division multi⁃ ple access (CDMA) in 3G and orthogonal frequency ⁃ division multiple access/single⁃carrier FDMA (OFDMA/SC⁃FDMA) for 4G. Facing the stringent demands of diverse scenarios in 5G, e. g., 1000x higher data rates, massive uplink connectivity and low access latency, the traditional pure orthogonal multiple ac⁃ cess is not a good option. Some alternative non⁃orthogonal mul⁃ tiple access schemes have attracted considerable attention and been identified as a crucial technology component in 5G since they can serve multiple users in the same frequency and time resources via code domain multiplexing and/or power domain multiplexing to enhance system access performance. The non⁃ orthogonal multiple access schemes are potentially able to sup⁃ port massive connections, improve spectrum efficiency and al⁃ so reduce access latency via the grant free access. Currently, some potential alternative multiple access schemes are being actively studied in 3GPP for 5G, including superposition cod⁃ ing based non ⁃ orthogonal multiple access (SPC ⁃ NOMA) [2], multi user shared access (MUSA) [3], sparse code multiple ac⁃ cess (SCMA) [4], pattern division multiple access (PDMA) [5], resource spread multiple access (RSMA) [6], non ⁃ orthogonal coded multiple access (NCMA) [7], and interleave⁃grid multi⁃ ple access (IGMA) [8]. In this paper, the principles, advantages and application sce⁃ narios of different multiple access techniques are discussed W Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS26
and compared. In addition, we introduce a unified framework that can merge a wide range of multiple access techniques, which helps to minimize the hardware functional module. Based on the unified framework, some initial work on flexible multiple access schemes is also introduced. Finally, the chal⁃ lenges and future directions are discussed. 2 Candidate Non⁃Orthogonal Multiple Access Solutions In this section, we introduce the typical candidate NOMA so⁃ lutions for 5G, which can be basically divided into three cate⁃ gories, i.e., the power domain based, code domain based and interleaver based. Their principles and key features are dis⁃ cussed. At last, we provide their comparison in terms of appli⁃ cation scenarios, system performance, receivers, etc. 2.1 Power Domain Based Solutions 2.1.1 SPC⁃NOMA NOMA based on superposition coding utilizes power domain for user multiplexing and can be applied for both downlink and uplink. Established by network information theory, non⁃orthog⁃ onal access with successive interference cancellation (SIC)/ dirty paper coding (DPC) can achieve the multiuser capacity region both in uplink and downlink. NOMA superposes multi⁃ ple users in power ⁃ domain and exploits channel gain differ⁃ ence between the multiplexed users with the aid of advanced receiver, e.g. the SIC receiver, for user separation. Fig. 1 shows signal transmission and receiving in downlink NOMA system with two users. Currently the NOMA technique is being discussed in the 3GPP under the study item of“study on down⁃ link multiuser superposition transmission (MUST)”for release 13 [9]. For the study in 3GPP, the study scope of NOMA is very limited, e.g. only about downlink transmission, only for the intra⁃cell usage and only for data channels. For 5G system, there are more application scenarios of NO⁃ MA technique, such as uplink and control channel, and more advanced NOMA techniques, such as combination with sophis⁃ ticated multiple⁃input multiple⁃output (MIMO) techniques and inter ⁃ cell techniques. In [10]- [12], MIMO NOMA schemes have been studied. Network NOMA which considering the multi⁃cell scenarios are also studied from EE⁃SE co⁃design per⁃ spective in [13]. 2.2 Code Domain Based Solutions 2.2.1 MUSA MUSA is a non⁃orthogonal multiple access scheme operat⁃ ing in code domain. Conceptually, each user’s modulated data symbols are spread firstly by a specially designed sequence which facilitates robust SIC implementation compared to the sequences employed by traditional direct⁃sequence CDMA (DS ⁃CDMA ). Then, each user’s spread symbols are transmitted concurrently on the same radio resource by means of“Shared Access”, which is essentially a superposition process. Finally, decoding of each user’s data from superimposed signal can be performed at the base⁃station side using SIC technology. The major processing blocks of MUSA transmitter and re⁃ ceiver are illustrated in Fig. 2. Symbols of each user are spread by a spreading sequence. Multiple spreading sequences constitute a pool from which each user can randomly pick one. Note that for the same user, different spreading sequences may also be used to different symbols. This may further improve the performance via interference averaging. Then, all spreading symbols are transmitted over the same time⁃frequency resourc⁃ es. The spreading sequences should have low cross⁃correlation and can be non⁃binary. At the receiver, codeword level SIC is used to separate data from different users. The complexity of codeword level SIC is less of an issue in the uplink as the re⁃ ceiver anyway needs to decode the data for all users. The only noticeable impact on the receiver implementation would be that the pipeline of processing may be changed in order to per⁃ form SIC operation. MUSA relies on a special family of complex spread sequenc⁃ es that can enjoy relatively low cross ⁃ correlation even when they are very short, say, 8 or even 4. The real and imaginary parts of the complex spread sequence can be drawn from an M⁃ ary real value set. For example, for a 3⁃value set {⁃1, 0, 1}, ev⁃ SIC: successive interference cancellation ▲Figure 1. Illustration of SPC ⁃NOMA transmission. ▲Figure 2. An example of MUSA with four resources shared by multiple users [1]. SIC: successive interference cancellation Power User 1 User 2 Time/frequency /spatial resources Base station User 2 User 2 signal detection User 1 signal detection User 1 SIC of user 1 signal Data of user1 SIC Using SIC receiver to decode each user’s data Each user’s spread symbols can be transmitted simultaneously User1 … C1 C2 Cn Each user’s modulated data symbols are spread by a specially designed sequence Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 27 User2 User n Data of user2 Data of usern
ery bit of the complex sequence is drawn from the constellation depicted in Fig. 3 with equal probability. It should be pointed out that the spread sequences used in MUSA are different from the spreading codes, in the sense that MUSA spreading does not have the low density property. Equipped with the well ⁃ optimized spreading sequence and state⁃of⁃the⁃art SIC technology, MUSA is capable of decou⁃ pling the multiuser mingled data even if those users are con⁃ tending to access the system. Potentially a large number of de⁃ vices are allowed to transmit data at their will, by randomly picking spread sequences, spread the data and send them. In other words, MUSA is suitable for the scenario where the up⁃ link transmissions are not tightly scheduled, and the grants for transmission are not signaled per user basis, and with a high overloading. The relaxed UL synchronization requirement for MUSA allows simple derivation of UL time from a DL synchro⁃ nization process, which can greatly cut down the battery con⁃ sumption. Lastly, the code domain superposition nature of MU⁃ SA can turn the near⁃far problem into a near⁃far advantage. The disparity in the received signal to noise ratio (SNR) across the simultaneously transmitting users can be exploited in MU⁃ SA to facilitate SIC. Tight transmit power control is no longer needed, which can further lower the device cost and its power consumption. 2.2.2 SCMA SCMA is a novel non ⁃ orthogonal multiple access scheme with sparse codebooks. The main idea of SCMA is to accommo⁃ date more users with limited resources and increase the total network throughput, without scarifying user experience, which can be overloaded to enable massive connectivity and support grant⁃free access. There are multiple layers in SCMA, which can be used for user multiplexing. Each layer has a predefined codebook, which consists of multiple codewords. The code⁃ words are composed of multi ⁃ dimensional complex symbols, and the codewords in the same codebook have the same sparse pattern. For each layer, the coded bits are directly mapped to codewords, which are selected from layer⁃specific SCMA code⁃ books. The codewords of different layers are overlaid in code and power domains and carried over shared time⁃frequency re⁃ sources. Typically, the layer multiplexing may become over⁃ loaded if the number of layers is more than the length of the codewords. Fig. 4 shows an example of bits to codewords mapping in a SCMA system. The codebook design of SCMA has been stud⁃ ied in [14]; it has been shown that with multi⁃dimensional con⁃ stellation, shaping and coding gain can be achieved. At the re⁃ ceiver, joint multiuser detection algorithms are needed. Due to the sparsity of the SCMA codeword structure, message passing algorithm (MPA) on factory graph with much lower complexity can be adopted to achieve a suboptimal performance. Some simplified algorithms are proposed [15]-[19] to further reduce the detection complexity. Besides the codebook and the receiver design, some other challenging issues of SCMA, e.g., the energy efficiency optimi⁃ zation, uplink grant free access, downlink multiuser transmis⁃ sion, and the multi⁃cell transmission based on SCMA have also been studied. The energy efficiency performance and optimiza⁃ tion of SCMA are investigated in [20] and [21]. In [22] and [23], uplink contention based grant⁃free access based on SC⁃ MA has been proposed for 5G radio access. [24] and [25] focus on the downlink multiuser SCMA (MU ⁃ SCMA) network. [24] theoretically derives the capacity for downlink Massive MIMO MU⁃SCMA systems. In [25], a weighted sum rate based user pairing and power sharing algorithm are introduced to the MU⁃ SCMA network. It shows that SCMA can significantly increase the downlink spectral efficiency of 5G wireless cellular net⁃ works. Further, SCMA has also been introduced into multi⁃cell transmission. SCMA based uplink inter⁃cell interference can⁃ cellation technique and open loop joint coordinated multiple point transmission are studied in [26] and [27], respectively. There are still many challenging issues for SCMA, which need to be solved in the future work. For example, the layer multiplexing in SCMA provides new degree⁃of⁃freedom for us⁃ er scheduling. The algorithms for user grouping and power allo⁃ cation need to be optimized. In addition, the combination of SCMA and MIMO can be further enhanced. 2.2.3 PDMA PDMA introduces reasonable diversity between multiple us⁃ ers to promote the capacity, which can obtain higher multiuser ◀Figure 3. The elements of the complex spreading sequence [3]. ▲Figure 4. Illustration of SCMA codebooks and the process of bit mapping [1]. R1-1 -1 1 0 I Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 Layer 6 (1,1)(b1,b2) (1,0) (0,1) (0,0) (0,1) (1,1) Sparsity pattern G = 0 1 0 1 1 0 1 0 1 1 0 0 0 0 1 1 1 0 0 1 0 1 1 0 October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS28 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
multiplexing and diversity gain. It considers the joint design of the transmitter and the receiver based on the optimization point of view for multiuser communication system. At the trans⁃ mitter side, the non⁃orthogonal characteristic pattern is used to distinguish users based on the multiple signals domain (includ⁃ ing time, frequency and the space domain). At the receiver side, sub⁃optimal multiuser detection by General SIC based on the features of the user pattern is utilized. To alleviate the error propagation problem of the SIC receiv⁃ er, the pattern used in PDMA is generally designed to ensure unequal transmission diversity for each user. In this way, the identical diversity order can be achieved after detection. In⁃ spired by the idea of unequal transmission diversity and sparse coding, an example of pattern and the related resource map⁃ ping has been proposed (Fig. 5). In the example, a code can also be seen as a pattern, which is used to define sparse mapping from data to a group of re⁃ sources. The code could be represented by a binary vector. The dimension of the vector equals to the number of resources in a group. Each element in the vector corresponds to a re⁃ source in a resource group. A“1”means that data shall be mapped to the corresponding resource. Actually, the number of “1”in the code is defined as its transmission diversity order. A code matrix is constructed by all codes sharing on the same resource group. Assuming six users multiplexing on four re⁃ source elements (REs). The data for User 1 are mapped to all the four resources in the group, and the data for User 2 are mapped to the first three resources, etc. The order of transmis⁃ sion diversity of the six users is 4, 3, 2, 2, 1, and 1, which is ob⁃ viously quite different from the SCMA scheme where all the us⁃ ers bear the same transmission diversity. Generally, if N is the size of resource group (the row num⁃ ber of code matrix), there are 2N - 1 possible binary vectors for a code matrix. Assuming K is the column number deter⁃ mined based on overload factor, we can thus choose K pat⁃ terns out from 2N - 1 candidates to construct code matrix. Se⁃ lection of codes also gives impacts on performance and com⁃ plexity. 2.2.4 RSMA RSMA combines the low rate channel code and the scram⁃ bling code (and optionally different interleavers) with good cor⁃ relation properties to separate different transmitters. In RSMA system, all users use the same frequency and time resources to transmit messages to the base station, regardless of the number of concurrent users. In other words, each user’s transmission power can be spread over all the available time and frequency resources. RSMA can be coupled with various waveforms/modulation schemes depending on the design target. Generally, it includes the single carrier RSMA and the multi⁃carrier RSMA. The sin⁃ gle carrier RSMA is optimized for battery power consumption and link budget extension by using single carrier waveforms. It allows grant⁃less transmission and potentially allows asynchro⁃ nous access. The grant⁃less transmission using RSMA reduces the signaling overhead, while the single carrier waveform fur⁃ ther reduces peak⁃to⁃average power ratio (PAPR) and achieves higher power amplifier efficiency. The pulse shaping block can further enhance the PAPR (e.g. potentially leading to constant envelope waveform), reducing out⁃of⁃band emission simultane⁃ ously. The multi⁃carrier RSMA is optimized for low latency ac⁃ cess, where reducing access delay is the design priority. It is suitable for the scenario where a connected state device is al⁃ ready synchronized to the base station and not link budget lim⁃ ited (e.g., close to the base station). Such a device can use RS⁃ MA with OFDM⁃based multi⁃carrier waveform for grant⁃less transmission to reduce overall access delay. 2.2.5 NCMA NCMA is a multiple access scheme based on the resource spreading by using non⁃orthogonal codewords, which is com⁃ posed of the codewords obtained by Grassmannian line pack⁃ ing problem [28]. To minimize the MUI theoretically, the spreading codes are designed with the minimum correlation. The non ⁃ orthogonal codebook is defined by C =[ ]c1 ⋯cK = é ë ê êê ê ù û ú úú ú c1 1 ⋯ cK 1 ⋮ ⋱ ⋮ c1 N ⋯ cK N ,C ∈ ℂN × K , where N is the sprea⁃ ding factor and K is the superposition factor. Then, the code⁃ book design problem can be posed in terms of maximizing the minimum chordal distance between codeword pairs minC æ è ç ö ø ÷max1 ≤ k ≤ j ≤ K 1 - |(ck )*.c j | where (ck )* is the conjugate cod⁃ eword of ck . NCMA can provide the additional throughput or improved connectivity with a small loss of block error rate (BLER) in spe⁃ cific environments, by exploiting additional layers through the superposed symbol, while satisfying QoS constraints. Since the receiver of NCMA system is available for parallel interference (b) resource mapping RE: resource element ▲Figure 5. Users sharing on four resource elements [5]. 1 G [4,6] Code = 1 1 1 1 1 1 0 1 0 1 0 0 1 0 1 0 0 1 0 0 0 0 1 (a) code matrix RE 1 RE 2 RE 3 RE 4 = + + + + + User 4User 3User 2User 1 User 5 User 6 Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 29 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
cancellation (PIC), the multiuser detection can be implement⁃ ed with low complexity. In addition, the MUI level between codeword pairs is always similar due to the correlation charac⁃ teristics mentioned above. Consequently, NCMA provides the potentials in terms of throughput or connectivity under special scenarios, e.g., huge connections with small packet in mMTC scenarios without changing the transmission block size, or for reducing the collision probability in contention based multiple access. 2.3 Interleaver Based Solution IGMA is an interleaver⁃based MA scheme, The typical trans⁃ mitter system structure using IGMA is shown in Fig. 6. Basi⁃ cally, the IGMA scheme distinguishes different users based on different bit⁃level interleavers, different grid mapping patterns and different combinations of bit ⁃ level interleaver and grid mapping pattern. Compared to the need of well⁃designed codewords or code sequences, the sufficient source of bit⁃level interleavers and/or grid mapping patterns are able to provide enough scalability for different connection densities, and also provide flexibility to achieve good balance between channel coding gain and ben⁃ efit from sparse resource mapping. By proper selection, the low correlated bit⁃level interleavers is achieved. In the grid map⁃ ping process, sparse mapping based on zero padding and sym⁃ bol⁃level interleaving is introduced, which provides another di⁃ mension for user multiplexing. Moreover, the density ρ of the grid mapping pattern is defined as the occupied RE numbers Nused dividing the total assigned RE numbers Nall , i.e. ρ = Nused /Nall . Different densities could be flexibly configured. It should be noted that the symbol sequence order is random⁃ ized after the grid mapping process due to symbol⁃level inter⁃ leaving, which may further bring benefit in terms of combating frequency selective fading and inter ⁃ cell interference, com⁃ pared to resource mapping using direct code ⁃ matrices/code⁃ books. At the receiver side, the low complexity multiuser detector (MUD) and the elementary signal estimator (ESE) that takes ad⁃ vantage of the special property of interleaving can be utilized with a simple de⁃mapping operation on the top. It should be noted that lower density of the grid mapping pattern further re⁃ duces detection complexity of ESE for IGMA. In addition, MAP and MPA detectors are also applicable for IGMA, which can improve the detection performance a lot comparing to ESE at the cost of complexity. The complexity of MAP/MPA for IG⁃ MA probably can be alleviated when spare grid mapping is used, due to the similar property of LDS. Fig. 7 shows an example of the grid mapping process of IG⁃ MA. The sparse symbol⁃to⁃RE mapping is performed based on an assigned grid mapping pattern. An exemplary operation can be mathematically formulated as a process by permutation ma⁃ trix αGM . According to the symbol⁃level interleave θk,2 assoc⁃ iated with the grid mapping pattern βk with density ρk(0 < ρk ≤ 1) , the corresponding permutation matrix αGM ∈ ℂN × L can be obtained. Thus, the kth user’s symbol s⁃ equence sk after zero padding and interleaving can be denoted by s' k = sk × αGM =[s' k,1,s' k,2,⋯,s' k,L ] , where L = N/ρk and ρk d⁃ ecides the number of zeros padded. 2.4 Summary of Multiple Access Techniques The pros and cons of the multiple access tech⁃ niques introduced above are summarized here in Table 1. It’s worth mentioning that some of these non⁃or⁃ thogonal schemes, such as SCMA MUSA and PD⁃ MA, can be implemented within a unified frame⁃ work, and each of them corresponds to a different codebook mapping module. In this way, the air interface can handover between different multiple access schemes in a flexi⁃ ble way, and all the other modules can be reused. This helps to improve the resource utilization and reduce the cost. In the fol⁃ lowing section, we will provide a unified framework for the mul⁃ tiple access schemes. 3 Unified Framework of Multiple Access Schemes Fig. 8 shows a unified framework of multiple access RF: Radio Frequency FEC: forward error correction ▲Figure 6. The IGMA transmitter [8]. βk = ì í î θk,2 ={4,0,2,0,0,3,0,1} ρk = 0.5 → αGM = é ë ê êê ê ù û ú úú ú 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 → s' k = sk × αGM = sk × αGM =[sk,4 0 sk,2 0 0 sk,3 0 sk,1] UE: user equipment ▲Figure 7. Example of the grid mapping process of IGMA when N = 4, ρk = 0.5 and L = 8 [8]. User data Channel coding FEC Repetition Modulation Bit⁃level interleaving Grid mapping Carrier modulation Baseband to RF Zero padding Sybol⁃level interleaving Zero⁃padding Symbol⁃level interleaving UE1 UE2 UE3 UE4 UEk f t + + + + … + = October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS30 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
schemes. The differences among these multiple access schemes lie in the different realization of interleaver, constella⁃ tion optimization, factor graph and multiplexing domain. The detailed explanations are listed in Table 2. 4 Flexible Multiple Access in 5G The above discussed advanced multiple access schemes as well as the traditional orthogonal multiple access scheme, e.g. OFDMA are all identified as potential candidates for 5G. There is no individual scheme can fulfill the requirements of all applications and scenarios in 5G system. A flexible adapta⁃ tion of these multiple access schemes is needed to support the diverse deployment scenarios and traffic requirements. For ex⁃ ample, in the case of massive connections, how to accommo⁃ date more users with limited resources has become a critical problem for next generation access network. With non⁃orthogo⁃ nal multiple access schemes, e.g., SCMA, MUSA, PDMA and RSMA, the same resources are shared and reused by multiple users, thus the number of connections increases. To support the traffic with low latency requirement, non⁃orthogonal multi⁃ ple access schemes help to realize grant⁃free multiple access, with which the latency is much lower, and the power consump⁃ tion of the devices can be reduced. In other scenarios, such as downlink machine type traffic, the simple orthogonal multiple access schemes are better due to the device cost and imple⁃ mentation complexity. OFDMA can be utilized for the cell⁃cen⁃ ter user with high data rate transmission applications. SIC: successive interference cancellation ▼Table 1. Summary of multiple access techniques DL: downlink eMBB: enhanced mobile broadband ESE: elementary signal estimator IGMA: interleave⁃grid multiple access mMTC: massive machine type communication MPA: message passing algorithm MUD: low complexity multiuser detector MUSA: multi user shared access NCMA: non⁃orthogonal coded multiple access PDMA: pattern division multiple access PIC: parallel interference cancellatio RSMA: resource spread multiple access SCMA: sparse code multiple access SIC: successive interference cancellation SPC⁃NOMA: superposition coding based non⁃orthogonal multiple access UL: uplink URLLC: ultra⁃reliable low latency machine type communication Category Scheme Scenario Multiplexing domain Transmitter Overloading Transmitter Spreading Transmitter multi ⁃ dimension constellation Receiver Power domain based SPC ⁃NOMA DL: eMBB Power Medium No No SIC Code domain based MUSA UL: mMTC, URLLC DL: eMBB Code/Power High Yes No SIC SCMA UL: mMTC, URLLC DL: eMBB Code/Power High Yes Yes MPA/SIC PDMA UL: mMTC, URLLC DL: eMBB Code/Power/Spatial High Yes No SIC/MPA RSMA UL: mMTC, URLLC Code/Power High Yes No SIC NCMA UL: eMBB, mMTC, URLLC code High Yes No PIC Interleaver based IGMA UL: eMBB, mMTC, URLLC Interleaver High Yes No MAP/MPA ESE MUD ▲Figure 8. Unified framework of multiple access schemes. Channel encoding, rate matching & scrambing Channel encoding, rate matching & scrambing Channel encoding, rate matching & scrambing … … …… Multiple access encoder (bit level) Interleaver Multiple access encoder (symbol level) Layer mapping & spatial precoding …… ……… Advanced receiver (IRC/SIC/R⁃ML) Advanced receiver (IRC/SIC/R⁃ML) … … Constellation optimization Factor graph Resource mapping+ + Example: u1 u2 u3 u4 u5 u6 r 1 r 2 r 3 r 4 Resource can be the time/frequency/space /code/power Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 31 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
Besides, the multiple access scheme should also be properly selected, taking the tradeoff of multiple conflicting objectives into account, e.g., complexity vs. performance, energy efficien⁃ cy (EE) vs. spectral efficiency (SE) and coverage. In addition, because the channel conditions and service load may also dy⁃ namically vary, the multiple access schemes and their related parameters such as the number of codewords, length of code⁃ word, spreading factor, max number of layers, need to be opti⁃ mized based on the instant services and the link conditions. In the following, we provide two potential adaptive multiple ac⁃ cess schemes in 5G. In [21], the adaptive multiple access scheme is studied from EE⁃SE co⁃design perspective, taking the detection complexity into consideration. The SCMA and OFDMA schemes are taken as the candidate uplink multiple access schemes in the study. The problem is formulated to choose the optimal multiple ac⁃ cess scheme and the related parameters simultaneously to max⁃ imize the EE under the total transmit power constraint, the quality of service (QoS) constraints and other specific require⁃ ments. The considered power consumption includes the trans⁃ mit power consumption, the static circuit power consumption, and the SCMA decoding power consumption related which is proportional to the SCMA decoding complexity order O(M df ) , where M is the constellation size, df = N K J , K and N denotes the codeword length and non⁃zero entries in each codeword in SCMA, respectively, and J = æ è ö ø K N is the maximum number of a⁃ ccess users. Fig. 9 shows the EE performance comparison of SCMA, OFDMA and the proposed link adaptation schemes with various cell radiuses. When the cell radius is small, the SCMA scheme has better EE performance; when the cell radi⁃ us is large, the OFDMA scheme performs better than SCMA scheme. The reason is that the SCMA can access more users than OFDMA, and the increment of the number of access users per resource can improve the system EE when the cell radius is small since the user transmit power efficiency is large when the path loss is small. When the cell radius increases, the user transmit power efficiency decreases and the increment of the number of access users per resource will decrease the system EE. The adaptation scheme can obtain the overall good EE per⁃ formance for all the cell radiuses (Fig. 9). Another example of the adaptive multiple access is between the spatial NOMA (also known as MIMO NOMA) scheme and orthogonal the multiuser MIMO (MU ⁃ MIMO). Fig. 10 shows ▼Table 2. Configuration methods of different multiple access schemes based on a unified framework IGMA: interleave⁃grid multiple access MUSA: multi user shared access MUST: Downlink Multiuser Superposition Transmission NCMA: non⁃orthogonal coded multiple access OMA: orthogonal multiple acce PDMA: pattern division multiple access RSMA: resource spread multiple access SCMA: sparse code multiple access SPC⁃NOMA: superposition coding based non⁃orthogonal multiple access OMA SPC ⁃NOMA MUSA (uplink) SCMA PDMA RSMA NCMA IGMA MUST Cat 1 [9] MUST Cat 2 [9] MUST Cat 3 [9] Interleaver Identity matrix Identity matrix Constraint permutation matrix Permutation matrix Identity matrix Identity matrix Identity matrix Optional Identity matrix Permutation matrix Constellation mapping Gray⁃mapped legacy constellation non⁃Gray⁃mapped superposed constellation Gray⁃mapped superposed constellation Gray⁃mapped legacy constellation Legacy constellation Joint optimization (multi ⁃dimensional modulation + Sparse matrix3 ) Legacy modulation Legacy modulation legacy modulation legacy modulation Factor graph Identity matrix Identity matrix Identity matrix Identity matrix Matrix composed of low cross⁃correlation and non⁃binary spreading sequence Sparse matrix Sparse matrix with unequal diversity order Matrix composed of scrambling code with good correlation properties Matrix obtained by Grassmannian line packing problem Sparse matrix Resource mapping (multiplexing domain) Time/frequency/code/space Power Power/bit Bit Code Code/power Code/power/space Code code code EE: energy efficiency OFDMA: orthogonal frequency ⁃division multiple access SCMA: sparse code multiple access ▲Figure 9. Average EE v.s. Cell Radiuses. 6.4 ×105 6.2 6.0 5.8 5.6 5.4 5.2 5.0 4.8 4.6 4.4 AverageEE(bits/Joule) 1.00.90.80.70.60.50.40.30.20.1 Cell radium (km) OFDMA Adaptation SCMA, N=2, K=4 SCMA, N=2, K=5 SCMA, N=2, K=6 October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS32 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
the concept of the orthogonal MU⁃MIMO and the spatial NO⁃ MA. In spatial NOMA, each two users can be served via one beam, and the interference between these two users will be large but can be cancelled via SIC decoding at the stronger us⁃ er. This is unlike the orthogonal MU⁃MIMO precoding (e.g. zero ⁃forcing MU⁃MIMO), in which there is no interference between users or it is usually small. Owing to this feature, when the channel has large correlations, spatial NOMA will have signifi⁃ cantly higher throughput over orthogonal MU⁃MIMO since the MU ⁃ MIMO precoding needs to reduce the interference be⁃ tween users and will suffer large gain loss in that case. In addi⁃ tion, when the user channel gain difference is large, the spatial NOMA will also have better performance compared to orthogo⁃ nal MU⁃MIMO due to the near⁃far effects. While in the high SNR regimes with low transmit correlation, the orthogonal MU⁃ MIMO is preferred since it can approach the capacity bound of MIMO broadcast channel in high SNR regimes. Considering the time varying fading characteristics of MIMO channels and the random distribution feature of active users, in a multiuser scenario, the adaptation between orthogonal MU ⁃ MIMO and spatial NOMA is desired for both the cell average and cell edge throughput enhancement. Fig. 11 shows the gain of the adaptation between orthogonal MU⁃MIMO based on zero⁃forcing scheme and spatial NOMA. When the transmit antenna correlation or the cell radius grows large, the spatial NOMA will have better performance than zero⁃ forcing based MU ⁃ MIMO, and the adaptation gain will in⁃ crease. The reason is that the higher transmit antenna correla⁃ tion will lead to the higher probability that the user channels have high correlation, and the larger cell radium will lead to the higher probability that the user channels have large gain difference. 5 Conclusions All the typical candidate NOMA solutions for 5G have differ⁃ ent strength points and weakness points. None of them can sur⁃ pass other schemes on all aspects. To fully exploit the advan⁃ tages of these candidate technologies and traditional orthogo⁃ nal multiple access solutions, a unified framework and a flexi⁃ ble multiple access schemes are required. Flexible switch among different NORMA schemes and the orthogonal multiple access technologies is expected to efficiently enhance the data rate and accommodate the necessary scalability for massive IoT connectivity and drastic reduction in access latency, and then to fully meet the diversified needs of 5G wireless commu⁃ nication systems. Some challenging problems need to be solved before NOMA schemes are put into use in 5G. In future, the impact of these candidate schemes on the existing systems, e.g. the grant free access procedure, reference signal, channel estimation and network assisted signaling, need to be carefully designed. What’s more, the performance tradeoff of the code mapping manners in these schemes and their implementation complexity may need further evaluated. The adaptive mecha⁃ nisms for part of these candidate schemes are also worth fur⁃ ther study to meet the diversified requirements of 5G. MU⁃MIMO: multiuser multiple⁃input multiple⁃output NOMA: non⁃orthogonal multiple access ▲Figure 11. Average cell throughput comparisons of various multiuser MIMO schemes (32 antenna base station at 6 GHz with four transceiver chains and four single antenna users) MU⁃MIMO: multiuser multiple⁃input multiple⁃output NOMA: non⁃orthogonal multiple access ▲Figure 10. Orthogonal MU⁃MIMO and Spatial NOMA. References [1] FuTURE Mobile Communication Forum 5G SIG. (Nov. 2014). Rethink mobile communications for 2020+ [Online]. Available: http://www.future ⁃forum.org/dl/ 141106/whitepaper.zip [2] A. Benjebbovu, Y. Saito, Y. Kishiyama, et al.,“Concept and practical consider⁃ ations of non ⁃ orthogonal multiple access for future radio access,”in IEEE ISPACS 2013, Naha, Japan, Nov. 2014, pp. 770- 774. doi: 10.1109/ ISPACS.2013.6704653. [3] 3GPP,“Discussion on multiple access for new radio interface,”TSG ⁃RAN WG1 #84bis, Busan Korea, R1 ⁃162226, Apr. 11-15, 2016. [4] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE PIMRC 2013, London, England, Sept. 2013, pp. 332- 336. doi: 10.1109/PIM⁃ RC.2013.6666156. User 1 User 2 Orthogonal beams User 2 User 1 (b) Spatial NOMA(a) Orthogonal MU⁃MIMO 20 Averagecellthroughput(bps/Hz) 0.8 Transmit antenna correlation 0.70.60.50.40.30.20.10 18 16 14 12 10 8 6 4 cell radius=500 m cell radius=1000 m MU⁃MIMO Spatial NOMA Adaptation Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 33 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
[5] 3GPP,“Candidate Solution for New Multiple Access”, TSG ⁃RAN WG1 #84bis, Busan, Korea, R1 ⁃163383, Apr. 11-15, 2016. [6] 3GPP,“Candidate NR multiple access schemes”, TSG ⁃ RAN WG1 #84bis, Busan Korea, R1 ⁃163510, Apr. 11-15, 2016. [7] 3GPP,“Considerations on DL/UL multiple access for NR,”TSG ⁃RAN WG1 # 84bis, Busan Korea, R1 ⁃162517, Apr. 11-15, 2016. [8] 3GPP,“Considerations on DL/UL multiple access for NR,”TSG ⁃RAN WG1 # 84bis, Busan Korea, R1 ⁃163992, Apr. 11-15, 2016. [9] Study on Downlink Multiuser Superposition Transmission (MUST) for LTE, 3GPP TR36.859 v13.0.0, Jan. 2016. [10] Z. Ding, F. Adachi, and H. V. Poor,“The application of MIMO to non ⁃orthogo⁃ nal multiple access,”IEEE Transactions on Wireless Communications., to be published. [11] Q. Sun, S. Han, C. I, and Z. Pan,“On the ergodic capacity of MIMO NOMA sys⁃ tems,”IEEE Wireless Communications Letters, vol. 4, pp. 405-408, Aug 2015, doi: 10.1109/LWC.2015.2426709 [12] Q. Sun, Chih ⁃Lin I, S. Han, et al.,“Sum rate optimization for MIMO non ⁃or⁃ thogonal multiple access systems,”accepted by IEEE WCNC, 2015. [13] S. Han, Chin ⁃Lin I, Z. Xu, et al.,“Energy efficiency and spectrum efficiency co ⁃design: from NOMA to network NOMA,”IEEE MMTC E ⁃Letter (Special Is⁃ sue on 5G), vol. 9, no. 5, pp. 21-24, Sept. 2014. [14] M. Taherzadeh, H. Nikopour, A. Bayesteh, et al.,“SCMA codebook design,”in IEEE VTC Fall 2014, Vancouver, Canada, Sept. 2013, pp. 1-5. doi: 10.1109/ VTCFall.2014.6966170. [15] K. Xiao, B. Xiao, S. Zhang, et al.,“Simplified multiuser detection for SCMA with sum ⁃ product algorithm,”in IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5. doi: 10.1109/WCSP.2015.7341328. [16] Y. Wu, S. Zhang, Y. Chen, et al.,“Iterative multiuser receiver in sparse code multiple access systems,”in Proc. IEEE ICC 2015, London, England, Jun. 2015, pp. 2918-2923. doi: 10.1109/ICC.2015.7248770. [17] A. Bayesteh, H. Nikopour, M. Taherzadeh, et al.,“Low complexity techniques for SCMA detection,”in Proc. IEEE Globecom Workshops 2015, San Diego, USA, Dec. 2015, pp. 1-6. doi: 10.1109/GLOCOMW.2015.7414184. [18] H. Mu, Z. Ma, M. Alhaji, et al.,“A fixed low complexity message pass algo⁃ rithm detector for uplink SCMA system,”IEEE Wireless Communications Let⁃ ters, vol. 4. no. 6, pp. 585-588, 2015. doi: 10.1109/LWC.2015.2469668. [19] Y. Du, B. Dong, Z. Chen, et al.,“A fast convergence multiuser detection scheme for uplink SCMA systems,”IEEE Wireless Communications Letters, to be published. [20] S. Zhang, X. Xu, L. Lu, et al.,“Sparse code multiple access: an energy efficient uplink approach for 5G wireless systems,”in Proc. IEEE Globecom 2014, Aus⁃ tin, USA, Dec. 2014, pp. 4782-4787. doi: 10.1109/GLOCOM.2014.7037563. [21] Q. Sun, Chih ⁃Lin I, S. Han, et al.,“Software defined air interface: a framework of 5G air interface,”in Proc. IEEE WCNC Workshop 2015, New Orleans, USA, Mar. 2015, pp. 6-11. doi: 10.1109/WCNCW.2015.7122520. [22] K. Au, L. Zhang, H. Nikopour, et al.,“Uplink contention based SCMA for 5G radio access,”in Proc. IEEE Globecom Workshop 2014, Austin, USA, Dec. 2014, pp. 900-905. doi: 10.1109/GLOCOMW.2014.7063547. [23] A. Bayesteh, E. Yi, H. Nikopour, et al.,“Blind detection of SCMA for uplink Grant ⁃Free Multiple Access,”in Proc. IEEE ISWCS 2014, Barcelona, Spain, Aug. 2014, pp. 853-857. doi: 10.1109/ISWCS.2014.6933472. [24] T. Liu, X. Li, and L. Qiu,“Capacity for downlink massive MIMO MU ⁃SCM.A system,”in Proc. IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5. doi: 10.1109/WCSP.2015.7341100. [25] H. Nikopour, E. Yi, A. Bayesteh, et al.,“SCMA for downlink multiple access of 5G wireless networks,”in Proc. IEEE Globalcom 2014, Austin, USA, Dec. 2014, pp. 3940-3945. doi: 10.1109/GLOCOM.2014.7037423. [26] U. Vilaipornsawai, H. Nikopour, and A. Bayesteh,“SCMA for open ⁃loop joint transmission CoMP,”in Proc. IEEE VTC Fall 2015, Boston, USA, Sept. 2015, pp. 1-5. doi: 10.1109/VTCFall.2015.7391126. [27] Y. Li, X. Lei, P. Fan, et al.,“An SCMA ⁃based uplink inter ⁃cell interference cancellation technique for 5G wireless systems,”in Proc. IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5, doi: 10.1109/WCSP.2015.7341306 [28] A. Medra and T. N. Davidson,“Flexible codebook design for limited feedback systems via sequential smooth optimization on the grassmannian manifold,” IEEE Transactions on Signal Processing, vol. 62, no. 5, pp. 1305-1318, Mar. 2014. doi:10.1109/TSP.2014.2301137. Manuscript received: 2016⁃06⁃30 SUN Qi (sunqiyjy@chinamobile.com) received the BSE and PhD degree in informa⁃ tion and communication engineering from Beijing University of Posts and Telecom⁃ munications, China in 2009 and 2014, respectively. After graduation, she joined the Green Communication Research Center of the China Mobile Research Institute. Her research focuses on 5G key technologies, including non ⁃orthogonal multiple access, new waveforms, flexible duplex and UDN. WANG Sen (wangsenyjy@chinamobile.com) received the PhD degree in informa⁃ tion and communication engineering from Beijing University of Posts and Telecom⁃ munications, China in 2013. He joined the Green Communication Research Center of the China Mobile Research Institute after graduation. He is now working on the 5G key technologies and standardization. His research interests include massive MI⁃ MO, non ⁃orthogonal multiple access, new waveforms and system level evaluation. HAN Shuangfeng (hanshuangfeng@chinamobile.com) received his MS and PhD de⁃ grees in electrical engineering from Tsinghua University, China in 2002 and 2006 respectively. He joined Samsung Electronics as a senior engineer in 2006 working on MIMO, MultiBS MIMO etc. From 2012, he is a senior project manager in the Green Communication Research Center of the China Mobile Research Institute. His research interests are green 5G, massive MIMO, full duplex, NOMA and EE ⁃SE codesign. Chih⁃Lin I (icl@chinamobile.com) received her PhD degree in electrical engineer⁃ ing from Stanford University, USA. She has been working at multiple world ⁃class companies and research institutes leading the R&D, including AT&T Bell Labs; Di⁃ rector of AT&T HQ, Director of ITRI Taiwan, and VPGD of ASTRI Hong Kong. She received the IEEE Trans. COM Stephen Rice Best Paper Award and is a winner of the CCCP National 1000 Talent Program. In 2011, she joined China Mobile as its Chief Scientist of wireless technologies, established the Green Communications Re⁃ search Center, and launched the 5G Key Technologies R&D. She is spearheading major initiatives including 5G, C ⁃ RAN, high energy efficiency system architec⁃ tures, technologies and devices; and green energy. She was an elected Board Mem⁃ ber of IEEE ComSoc, Chair of the ComSoc Meetings and Conferences Board, and Founding Chair of the IEEE WCNC Steering Committee. She is currently an Execu⁃ tive Board Member of GreenTouch, a Network Operator Council Member of ETSI NFV, a Steering Board Member of WWRF, and a Scientific Advisory Board Member of Singapore NRF. Her current research interests center around“Green, Soft, and Open” BiographiesBiographies October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS34 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
Multiple Access Rateless Network Coding forMultiple Access Rateless Network Coding for Machine⁃to⁃Machine CommunicationsMachine⁃to⁃Machine Communications JIAO Jian1,2 , Rana Abbas2 , LI Yonghui2 , and ZHANG Qinyu1 (1. Harbin Institute of Technology Shenzhen Graduate School, Shenzhen 518055, China; 2. Center of Excellence in Telecommunications, University of Sydney, Sydney, NSW 2006, Australia) Abstract In this paper, we propose a novel multiple access rateless network coding scheme for machine⁃to⁃machine (M2M) communications. The presented scheme is capable of increasing transmission efficiency by reducing occupied time slots yet with high decoding suc⁃ cess rates. Unlike existing state⁃of⁃the⁃art distributed rateless coding schemes, the proposed rateless network coding can dynami⁃ cally recode by using simple yet effective XOR operations, which is suitable for M2M erasure networks. Simulation results and analysis demonstrate that the proposed scheme outperforms the existing distributed rateless network coding schemes in the scenar⁃ io of M2M multicast network with heterogeneous erasure features. rateless network coding; multiple accesss; machine⁃to⁃machine communications (M2M) Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 005 http://www.cnki.net/kcms/detail/34.1294.TN.20161011.1511.002.html, published online October 11, 2016 Special Topic M 1 Introduction achine⁃to⁃machine (M2M) communication sys⁃ tem is expected to support a massive number of devices communicating with each other in a fully automated fashion with minimum or with⁃ out human intervention [1]. Equipped with networked and real⁃ time processing capabilities, these devices can implement a wide range of applications, such as intelligent transportation systems (ITS), healthcare monitoring, smart metering, energy management and smart grids. M2M communication system is generally characterized by a massive number of machine⁃type communication (MTC) devic⁃ es that have no/low mobility, low computational and storage ca⁃ pabilities, and low power budget [2]. Moreover, most MTC de⁃ vices suffer from severe congestion and access delay in an M2M system with a large number of devices [3]-[5]. Therefore, the main motivation behind this paper is to propose a coding strategy that exploits the interference in the channel to in⁃ crease data rates. Our work focuses on the cooperative joint network and coding strategy for MTC devices in multicast set⁃ tings. These MTC devices disseminate messages to multiple re⁃ ceivers simultaneously with the help of relay nodes. The rateless code, originally investigated in [6] for single source broadcasting in a single hop network, is deemed as a milestone for packet erasure codes. It can recover the original k information symbols from any n = k + O( k ln2 (k θ)) received coded symbols with the probability 1⁃θ and the decoding cost of O(k ln(k θ)) of operations, where θ is the allowable failure probability to recover the original message after n coded sym⁃ bols have been received. In addition, the encoding and decod⁃ ing process of the rateless code is complex, including logarith⁃ mic order for Luby Transform (LT) code and linear order for Raptor code. Furthermore, both LT and Raptor codes are able to provide practical capacity ⁃ achieving solutions, if their en⁃ coding degree distributions are sophisticated designed [7], [8]. The rateless code has been widely applied in cooperative communications [9]- [13]. In [9], the complexity, delay, and memory of different state⁃of⁃the⁃art rateless coding algorithms are analyzed for a multi⁃hop network. In [10], a superimposed on ⁃ the ⁃ fly recoding scheme is performed by each transport node in a multi⁃hop tree network, but it is difficult to imple⁃ ment due to the high decoding complexity. The first distributed LT (DLT) code is proposed in [11], and a new degree distribu⁃ tion, named deconvolved soliton distribution (DSD) is de⁃ signed. However, all the source nodes and relays are assumed This works was supported in part by Natural Scientific Research Innovation Foundation in Harbin Institute of Technology under Grant No. HIT. NSRIF 2017051, Shenzhen Basic Research Program under Grant Nos. JCYJ20150930150304185 and JCYJ2016 0328163327348, and National High Technology Research & Development Program of China under Grant No. 2014AA01A704. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 35
Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS36 to have exact information regarding the number of sources and encoding distributions to adapt relaying schemes. In [12], the authors utilized density evolution and linear programming frameworks to find an optimal combination at each relay node for any network architecture consisting of four sources. This manner can achieve the asymptotical error floor, but has intrac⁃ table calculation complexity. A relaying protocol for Y ⁃ net⁃ works, namely Soliton⁃Like Rateless Coding (SLRC), is intro⁃ duced in [13]. By enabling probabilistic forwarding and com⁃ bining packets to reduce the overhead between relay and desti⁃ nation, the aggregate distribution at the destination can still maintain a near⁃ideal distribution, even if one source left the network. However, the lack of buffer utilization in SLRC relay limits the total encoding and decoding overheads. The destination cooperation in interference channels is an⁃ other feature of M2M communication where one device can act as a relay for another device. A cooperative communication scheme for two mobile users is proposed in [4], which is poten⁃ tially able to receive and decode each other’s messages based on the signal⁃to⁃interference⁃plus⁃noise ratio (SINR). In [5], The received signal at the destination can be realized as a su⁃ perposition of coded symbols sent from the relay, which is ca⁃ pacity approaching if an appropriate successive interference cancellation (SIC) is used for decoding [14]. In this paper, we propose an adaptive rateless network cod⁃ ing scheme in an M2M erasure network. First, an simple de⁃ gree distribution is designed for rateless coding in all the source devices, and the collided devices transmit simultaneous⁃ ly. Then, an optimal relaying strategy is proposed to forward and combine the encoded packets with appropriate propor⁃ tions, according to different erasure probabilities over the un⁃ derlying edges. This is particularly suitable for M2M communi⁃ cations with strict power limitations, especially when the data size is small. By doing this, the total time slot of transmission is reduced obviously while the high decoding success rates are maintained. Moreover, we compared the current typical rate⁃ less coding relay schemes with our proposed scheme, with the aspects of the complexity and buffer memory. Simulation re⁃ sults show that the proposed rateless network coding scheme outperforms the existing distributed rateless coding schemes under various erasure probability scenarios. 2 Network Model and Rateless Code 2.1 Network Model In [12] and [13], authors have introduced and optimized the applications of rateless coding in the Y⁃network model. We at⁃ tempt to extend the Y⁃network model to a relay multicast mod⁃ el as shown in Fig. 1a. The relay R multicasts the data streams both to destination nodes D1 and D2 and guarantees the two source data to be recovered. Due to the special feature of rate⁃ less coding, the overheads at two destination nodes are appar⁃ ently the same as the one in the Y⁃network and accordingly the DLT and SLRC algorithms are also appropriate for the network model in Fig. 1a. We also consider a butterfly network model (Fig. 1b), which has two sources, one relay and two destinations. Two direct edges are added to send two separate data streams form S1 and S2 respectively. The encoded packets should be processed (re⁃ coding) at R, and they can be converged at both D1 and D2 in the end. The model uses multicast from source to relay and di⁃ rectly from source to destination as well. This is its remarkable difference from the Y⁃network model. We define the edges in this model as e1 to e6. Each edge has an erasure probability εi, which is described as an independent⁃identical⁃distribution (i.i. d) Bernoulli variable. We assume that the packet size and the transmission rate of all the edges are equivalent (one time slot one packet). The rateless coding transmission scheme in this network model is described as follows. 1) Step 1: S1 and S2 generate the encoded packets with a rate⁃ less coding degree distribution [7]; 2) Step 2: S1 multicasts the encoded stream both to R and to D1, and simultaneously, S2 multicasts its stream to R and to D2; 3) Step 3: R generates a new encoded stream by the relaying network coding (NC) scheme with the received encoded packets and multicasts them to D1 and D2 simultaneously; 4) Step 4: Once D1 and D2 receive enough encoded packets, they start to decode the encoded packets from the source and relay nodes to recover the two sources original packets; ▲Figure 1. The proposed network models: (a) relay multicast model; (b) butterfly model. (a) R: relay node S:source node D: destination node (b) S1 S2 D1 D2 R e3 e4 e1 e2 S1 S2 D1 D2 R e1 e2 e3 e4 e5 e6
Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 37 5) Step 5: After successful decoding, D1 (D2) transmits a single acknowledgment (ACK) packet indicating the termination of the session. 2.2 Rateless Coding Rateless coding is a modern forward error correction (FEC) technology. It protects from packet ⁃ loss, and can reduce the feedbacks for user acknowledgement due to rarely caring about the erasure probability of the channel. In a single⁃hop scenar⁃ io, the original packets are defined as k data symbols, and the encoded packets required by the decoder are defined as n en⁃ coded symbols. Furthermore, the overhead is defined as ( )n - k k and the number of encoded symbols sent by the send⁃ er is defined as N. Therefore, the expected code rate at the sender is conveyed as R = k N = ( )1 - ε k n , where ε is the era⁃ sure probability of the channel. On the other hand, the encod⁃ ing and decoding complexity of rateless codes is very low, which are expressed as O(ln k) for LT codes and O(k) for Rap⁃ tor codes. As an example of rateless coding, the LT code uses robust soliton distribution (RSD) to achieve the erasure channel ca⁃ pacity under the single hop network. The coding degree distri⁃ bution is a key element for the successful recovery of data sym⁃ bols. For a parameter δ and the length k RSD, μ(k) is defined as: μ(i)= ρ(i)+ τ(i) β , for i = 1,…,k , (1) where ρ(i)={1 i, for i = 1 1 i(i - 1) , for i = 2,…,k , (2) τ(i)= ì í î ïï ïï S ik, for i = 1,…,ë ûk S - 1 S ln(S δ) k, for i = ë ûk S 0, for i = ë ûk S + 1,…,k , (3) β =∑i = 1 k μ(i)+ τ(i) . (4) S is the average number of degree⁃one symbols, namely rip⁃ ple size, which is defined by S = c ln(k δ) k , where c>0. It is worth noting that the LT decoder performs the BP algo⁃ rithm with the prior knowledge of the degree and associated neighbors. Given a block of encoded symbols, the decoder re⁃ cursively decodes the data symbols from the bipartite graph connecting the information and encoded symbols. The BP algo⁃ rithm starts from degree⁃one symbols, by removing their contri⁃ butions from the graph in order to produce a smaller graph with another set of degree⁃one encoded symbols. Then, the new degree⁃one encoded symbols of this smaller graph are removed again, and iteratively the process continues to recover all data symbols, as described in [7] and [8]. 3 Analysis of Relaying Schemes As the rateless code is used in multi⁃source relay network, the erasure probability of different paths (multicast and uni⁃ cast) may influence the relaying strategy and the correspond⁃ ing performance with NC. Specifically, the relay R may receive no packet from S1 or S2 in a time slot due to packet loss in multi ⁃source relay network. Hence, it is an interesting and signifi⁃ cant topic to select proper rateless coding algorithms based on NC and relaying strategy for efficient transmissions on lossy network. In this section, we try to consider the conventional methods in Y ⁃ networks and butterfly networks, and compare their decoding performance at the destination nodes. Moreover, we propose a new optimized ⁃NC scheme to trade off the decod⁃ ing performance by selecting proper forwarding and combining probabilities. 3.1 Comparison of Typical Relaying Schemes We assume that the number of original packets is k and the number of encoded packets generated is N at both the source nodes. In two fixed time slots, the destination nodes D1/D2 of butterfly network can receive one encoded packet from S1/S2 in the first slot, and then receive one from the relay R in the sec⁃ ond slot. It is a limited condition that D1 and D2 only receive the maximum 2N packets when N encoded packets are sent from the source, if and only if all the edges are lossless. D1 and D2 use the BP decoding algorithm to decode the compilations of two encoded streams after 2N slots to recover 2k original pack⁃ ets, respectively. There are the following four typical relaying schemes in this butterfly network model: 1) Store⁃and⁃Forward (SF) The relay R immediately forwards the packets to the next hop as soon as it receives packets. If two packets arrive simul⁃ taneously, R randomly forwards one of them and stores another into the buffer. If the relay R receives no packet, it waits for the next slot. Due to the uncertain storage of packets, this scheme may easily make congestion on R. 2) DLT With S1 (S2) using DSD, R performs random decision proto⁃ col in [11] to combine and forward two received packets. Once the erasure event occurs at one of the edges between sources and relay, R directly forwards another received packet. If no packets arrive, the relay waits for the next slot. These waiting slots at relay lead to low efficiency due to a serious waste of sources. By using considerable low ⁃ degree encoded packets, this scheme could scarcely cover all the original packets of two sources, despite of its simple encoding complexity. 3) SLRC With S1 (S2) using RSD, R uses the SLRC relaying scheme to operate the two encoded packets. It forwards most of the low degree packets (degree⁃one or degree⁃two) directly and com⁃
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS38 bines other high degree packets, in order to assure the aggre⁃ gate distribution at destination nodes to be soliton⁃like. With the SLRC, the transmission time slot is not wasted if the pack⁃ et on e1 or e2 is dropped, by R’s choosing two packets from the buffers to combine and forward. Compared with the DLT scheme, this SLRC scheme obtains more gains. 4) eXclusive⁃OR (XOR) This scheme is based on the simple eXclusive ⁃ OR (XOR) operation of classic NC in the butterfly model. The relay R combines the two received packets into one new packet. If only one packet arrives, the relay R sends the received packet di⁃ rectly, and if no packet arrives, R waits for the next time slot. Since the relay carries the two received packets by forward⁃ ing and combining operations, the recoding complexity comes almost from XOR operation which depends mainly on the era⁃ sure probability of the edges. In addition, the relay requires two buffers to store the packets from different sources. There⁃ fore, we compare the four schemes (Table 1). We can find that SF has the least recoding complexity, and the complexity of SL⁃ RC and DLT depends significantly on their forwarding proba⁃ bility. On the other hand, DLT and XOR need the smallest buf⁃ fer size of only one packet for each source, while SLRC must store all the packets not forwarded for the next operation. 3.2 Analysis of Decoding Matrix In this section, we make an intuitive analysis on recovery performance at the destination nodes by the BP decoding ma⁃ trix, in which the BP decoding algorithm could be described as one unitization process, with the columns denoting the encod⁃ ed symbols and the rows denoting the original data symbols. We define A and B as the encoding matrix at S1 and S2 re⁃ spectively, and denote the dimension of the matrix as the sub⁃ scripts. The decoding matrices at D1 and D2 are the aggregation of the forwarding and combining sub⁃matrices A and B, with 2k rows due to the number of data symbols from two sources. Since the lost packets have been removed from the matrix, the number of columns reveals the received encoded symbols by destination nodes exactly. For the SF scheme, the decoding matrices at D1 and D2 are defined as: F D1 2k ×[ ](1 - ε5)N +[ ](1 - ε1)+(1 - ε2) (1 - ε3)N 2 = é ë ê ê ù û ú ú A S1 k ×(1 - ε5)N A S1 k ×(1 - ε1)(1 - ε3)N 2 0 0 0 B S2 k ×(1 - ε2)(1 - ε3)N 2 , (5) F D2 2k ×[ ](1 - ε6)N +[ ](1 - ε1)+(1 - ε2) (1 - ε4)N 2 = é ë ê ê ù û ú ú 0 A S1 k ×(1 - ε1)(1 - ε4)N 2 0 B S2 k ×(1 - ε6)N 0 B S2 k ×(1 - ε2)(1 - ε4)N 2 . (6) From the matrices F D1 and F D2 in (5) and (6), we know that the SF scheme is only appropriate for unicast like source to destination, since the dimensions of sub⁃matrices for A and B are extremely unequal. The lack of combination operations makes the destination nodes unable to recover the whole data symbols. Therefore, this scheme is inefficient for multicast in the butterfly network model. For the SLRC scheme, as an optimized DLT, we only pres⁃ ent its decoding matrices that are defined as: H D1 2k × ( )1 - ε5 + 1 - ε3 N = é ë ê ê êê ê ê ù û ú ú úú ú ú A S1 k ×(1 - ε5)N A S1 k × ν1(1 - ε3)N 0 A S1 k ×(1 -∑i = 1 2 νi)(1 - ε3)N 0 0 B S2 k × ν2(1 - ε3)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε3)N , (7) H D2 2k × ( )1 - ε6 + 1 - ε4 N = é ë ê ê êê ê ê ù û ú ú úú ú ú 0 A S1 k × ν1(1 - ε4)N 0 A S1 k ×(1 -∑i = 1 2 νi)(1 - ε4)N B S2 k ×(1 - ε6)N 0 B S2 k × ν2(1 - ε4)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε4)N , (8) where νi (i=1, 2) is the probability distribution of the relay for⁃ warding packets from S1 and S2, while ν͂ i = 1 - νi is the distribu⁃ tion of the packets into the buffer. For the XOR Scheme, the decoding matrices are defined as: G D1 2k ×[ ](1 - ε5)+ Max{ }(1 - ε1), (1 - ε2) (1 - ε3) N = é ë ê ê ù û ú ú A S1 k ×(1 - ε5)N A S1 k ×(1 - ε1)(1 - ε3)N 0 B S2 k ×(1 - ε2)(1 - ε3)N , (9) G D2 2k ×[ ](1 - ε6)+ Max{ }(1 - ε1), (1 - ε2) (1 - ε4) N = é ë ê ê ù û ú ú 0 A S1 k ×(1 - ε1)(1 - ε4)N B S2 k ×(1 - ε6)N B S2 k ×(1 - ε2)(1 - ε4)N . (10) In (9) and (10), G can be segmented into four sub⁃matrices: Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu ▼Table 1. The complexity and buffer for the four schemes * γi and νi are the distributions of the no⁃forwarding packets for DLT and SLRC DLT: distributed Luby transform SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR SF DLT SLRC XOR Relay recoding complexity 0 ( )1 - ε1 ( )1 - ε2 ( )1 - γi N (1 -∑i = 1 2 νi)N ( )1 - ε1 ( )1 - ε2 N Buffer size of relay N 2 packets for each source One packet for each source ( )1 - νi N packets for each source One packet for each source
October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 39 the two parts in the left are the forwarding matrix A or B, and the two right parts are the combined matrices. Since the com⁃ bined symbols of this scheme may be lack of degree⁃one and degree⁃two packets, it needs enough columns of single sub⁃ma⁃ trices A or B on the edge e5 or e6 in Fig. 1 to start the BP decod⁃ er. In the decoding process, only if all the data symbols of A have been recovered, B can begin to decode. By comparing these decoding matrices, we can find that the forwarding matrix A or B has occupied such numerous columns in H, as A S1 k × ν1(1 - ε3)N or B S2 k × ν2(1 - ε3)N . The combination part of en⁃ coded symbols as A and B é ë ê ê êê ê ê ù û ú ú úú ú ú A S1 k ×(1 -∑i = 1 2 νi)(1 - ε3)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε3)N in the SLRC scheme has much less columns than that of é ë ê ê ù û ú ú A S1 k ×(1 - ε1)(1 - ε3)N B S2 k ×(1 - ε2)(1 - ε3)N in the XOR scheme. SLRC can still recode new packets in the re⁃ lay to transmit, even if no packets received due to enough large ε1 and ε2. However, it has not fully utilized the packets directly from D1 and D2 on e5 and e6. Note that, only N encoded packets at most could be sent by the sources and relay, which con⁃ strains the required time slots to be only 2N in the network model, given one packet at each time slot. It renders that the decoding matrix H has many single forwarding columns in A or B which obviously reduces the relevance of encoded symbols from S1 and S2. As a result, the large proportion of forwarding packets by the relay cannot give much help to improve the BP decoding performance, especially on the condition of the rela⁃ tively small erasure probability ε5 and ε6. On the other hand, when ε5 and ε6 become larger, the decod⁃ ing performance is mostly decided by the proportion of forward⁃ ing the single packets and XOR combination of two sources’ packets in the relay. In the XOR scheme, the number of recov⁃ ery data symbols would decline very fast due to the lack of low degree encoded symbols for BP decoding. Besides, the XOR operations would be blocked and degraded since two separate packets from two sources could hardly arrive at the relay simul⁃ taneously in the large ε5 and ε6. 3.3 Proposed Optimized NC Scheme On the basis of above analysis, we have found that the relay⁃ ing schemes should forward the low⁃degree packets for starting BP decoder, by taking into consideration the erasure probabili⁃ ties of the direct edges e1 and e2. On the other side, the relay al⁃ so needs to remain the enough proportion of combinations of the packets in the buffers, in order to prevent from no received packets from S1 and S2 simultaneously. Accordingly, we pro⁃ pose a novel NC scheme with a self⁃adjusted forwarding proba⁃ bility associated with the variations of ε5 and ε6. The basic rule of the proposed scheme is as follows: if ε1 and ε2 increase, the relay immediately forwards more low⁃degree packets; if ε5 and ε6 decrease, the relay combines more packets. The proposed algorithm, named Opt ⁃ NC scheme, has the comparable complexity and buffer requirements with the SLRC scheme. We denote the forwarding probabilities λ and θ for en⁃ coded packets from S1 and S2, which are predetermined to be equivalent to the erasure probabilities ε5 and ε6, respectively. Algorithm 1 shows the steps of Opt⁃NC algorithm. Algorithm 1: Opt⁃NC Scheme (at one time slot) p1: received encoded packets from S1; p2: received encoded packets from S2; d1: degree of p1; d2: degree of p2; λ: forwarding probability of the low degree packets from S1; θ: forwarding probability of the low degree packets from S2; a=rand(); b=rand(); if d1=1˅2 and d2=1˅2 and a<λ and b <θ forward p1 or p2 with equal probability; put another packet into the buffer of another source; else if d1=1˅2 and a <λ forward p1 and put p2 into the buffer of S2; else if d2=1˅2 and b <θ forward p2 and put p1 into the buffer of S1; else put the packets received into the buffers respectively; pb1: random choose one packet in the buffer of S1; pb2: random choose one packet in the buffer of S2; pXOR = pb1XOR pb2; forward pXOR; end if * ˅ means logical operator of OR 4 Simulation Results and Discussion We analyze the performance of the above five algorithms in a butterfly network coding system as Fig. 1b. The encoding de⁃ gree distribution is selected to be RSD with parameters δ = 0.05, c=0.03. The number of data symbols k=100, and the num⁃ ber of encoded symbols from S1 and S2 is indicated to be the same as N. We emulate the encoding and decoding procedure using Monte Carlo experiments with 10,000 times. The ratio between the statistics of decoding failure times and total exper⁃ iment times is defined as decoding failure rate (DFR). In this work, the lowest displayable DFR in our simulation is 10⁃4 . Giv⁃ en the time slots and erasure probabilities of edges, the lower DFR of relaying schemes means outstanding decoding perfor⁃ mance. We give the unicast performance and multicast perfor⁃ Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic
October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS40 mance respectively to discuss the influences between the sin⁃ gle source and double sources. Fig. 2 gives the performance of five schemes in the erasure free network model. We can find that the Opt⁃NC scheme ap⁃ proaches the unicast and multicast curves of the XOR scheme, which obtain the lowest DFR of 10⁃4 at 400 time slots. The oth⁃ er schemes need at least 600 time slots to recover two sources’ data, due to their inefficiency of buffer utilization at the relay. This simulation result proves that the Opt⁃NC scheme can ap⁃ ply as a typical NC scheme to get the remarkable multicast throughput gains in lossless network data transmission. With vary erasure probabilities from 0 to 0.8 of all edges among ε1 to ε6, and the code length N=360, the decoding per⁃ formance of five schemes is shown in Fig. 3 . The unicast re⁃ sults of the schemes reveal the better performance than the multicast results, since the erasure probabilities of ε1, ε2, ε3, ε4 make the combination operations at the relay inappropriate. It is noted that the SLRC, XOR and Opt⁃NC schemes have simi⁃ lar multicast decoding performance, with the DFR lower than 10⁃ 4 and the erasure probability of 0.2. The simulation results indicate that our adaptive Opt⁃NC scheme integrates the advan⁃ tages of SLRC and XOR, which also reveals outstanding decod⁃ ing performance in lossy network. Fig. 4 shows the DFRs of five schemes with the encoded packets of N=250, ε1 to ε4 null, and ε5 and ε6 from 0.2 to 0.9. Since the multicasts of the SF and DLT are both restricted by the limited number of encoded packets from source, their de⁃ coding performance maintains at an inferior level in spite of ε5 and ε6 increasing. On the other side, the multicast DFR perfor⁃ mance of XOR and that of Opt⁃NC are almost consistency with their unicasts. If the erasure probabilities of e5 and e6 are lower than 0.25, the Opt⁃NC scheme can get the similar DFR with that of the XOR scheme. If ε5 and ε6 increase from 0.25 to 0.9, the Opt⁃NC scheme outperforms the XOR scheme by its dy⁃ namic property. In addition, compared to the SLRC, the Opt⁃ NC scheme also has a better performance as ε5 and ε6 are both lower than 0.45. However, the SLRC gives a lower decoding failure rate as the ε5 and ε6 are both in a range of 0.45 to 0.8. Once the erasure probability increases higher than 0.8 (the edges e5 and e6 are almost interrupted), the Opt⁃NC scheme ap⁃ proaches SLRC⁃multicast with a higher efficiency. In a word, Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR ▲Figure 2. All the edges are erasure free, k =100. ▲Figure 3. All the edges have the same erasure probability, k =100, N =360. ▲Figure 4. Edges e5 and e6 are lossy while other edges are lossless, k =100, N =250. 100 10-1 10-2 10-3 10-4 600550500450400350300250200150 Total slots of transmission DFR SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast 100 10-1 10-2 10-3 10-4 0.8 Erasure probability of all edges DFR 0.70.60.50.40.30.20.10 SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast 100 10-1 10-2 10-3 10-4 1.0 ε5 and ε6 DFR 0.2 0.90.80.70.60.50.40.3 SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR
October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 41 the proposed relaying scheme is an adaptive method to compro⁃ mise the decoding performance of XOR and SLRC. 5 Conclusions In this paper, we have studied rateless network coding ap⁃ plied in machine⁃to⁃machine communications for multiple ac⁃ cess applications. A novel dynamic relaying scheme Opt⁃NC was proposed that exploits the forwarding and combining opera⁃ tions to obtain an enhanced decoding performance of the de⁃ coder at the destination nodes. The Opt⁃NC scheme has adap⁃ tive capability of responding to the vary erasure probability of direct edges. The simulation results show that the proposed re⁃ lay scheme performs close to the optimal XOR scheme in loss⁃ less and lossy network, respectively. Furthermore, the Opt⁃NC scheme can be used in the physical layer by incorporating the XOR operation and superposition practical modulations. Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic References [1] M. Shirvanimoghaddam, Y. Li, and M. Dohler,“Probabilistic rateless multiple access for machine⁃to⁃machine communication,”IEEE Transactions on Wireless Communications, vol. 14, no. 6815-6826, Dec. 2015. [2] K. Zheng, S. Ou, J. Alonso⁃Zarate, et al.,“Challenges of massive access in high⁃ ly dense LTE⁃Advanced networks with machine⁃to⁃machine communications,” IEEE Wireless Communications Magazine, vol. 21, no. 3, pp. 12-18, Jun. 2014. [3] G. Durisi, T. Koch, and P. Popovski. (2015).“Towards massive, ultra⁃reliable, and low ⁃ latency wireless communication with short packets,”CoRR [Online]. Available: https://arxiv.org/abs/1504. 06526 [4] B. W. Khoueiry and M. R. Soleymani,“A novel destination cooperation scheme in interference channels”, in Proc. IEEE 80th Vehicular Technology Conference (VTC 2014 ⁃ Fall), Vancouver, Canada, Sept., 2014. doi: 10.1109/VTC⁃ Fall.2014.6965826. [5] M. Shirvanimoghaddam, M. Dohler, and S. J. Johnson. (2016).“Massive multiple access based on superposition raptor codes for M2M communications,”CoRR [Online]. Available: https://arxiv.org/abs/1602.05671 [6] J. Byers, M. Luby, M. Mitzenmacher, and A. Rege,“A digital fountain approach to reliable distribution of bulk data,”in Proc. ACM SIGCOMM’98, Vancouver, Canada, Jan. 1998, pp. 56-67. doi:10.1145/285237.285258. [7] M. Luby,“LT Codes,”in Proc. IEEE Symposium on the Foundations of Comput⁃ er Science (STOC), Vancouver, Canada, 2002, pp. 271-280. [8] A. Shokrollahi,“Raptor Codes,”IEEE Transactions on Information Theory, vol. 52, no. 6, pp. 2551-2567, Jun. 2006. doi: 10.1109/TIT.2006.874390. [9] P. Pakzad, C. Fragouli, and A. Shokrollahi,“Coding schemes for line networks,” in Proc. ISIT, Adelaide, Germany, 2005, pp. 1853-1857. [10] R. Gummadi and R. S. Sreenivas,“Relaying a fountain code across multiple nodes,”in Proc. SIGCOMM’08, Seattle, USA, 2008, pp. 483-484. [11] S. Puducheri, J. Kliewer, and T. E. Fuja,“The design and performance of dis⁃ tributed LT codes,”IEEE Transactions on Information Theory, vol. 53, no. 10, pp. 3740-3754, Oct. 2007. [12] D. Sejdinovic, R. Piechocki, and A. Doufexi,“AND⁃OR tree analysis of distrib⁃ uted LT codes,”in Proc. ITW, Volos, Greece, 2009, pp. 261-265. [13] A. Liau, S. Yousefi, and I. Kim,“Binary soliton⁃like rateless coding for the Y⁃ network,”IEEE Transactions on Communications, vol. 59, no. 12, pp. 3217- 3222, Dec. 2011. [14] R. Abbas, M. Shirvanimoghaddam, Y. Li, and B. Vucetic,“On SINR⁃based ran⁃ dom multiple access using codes on graph,”in IEEE Global Communications Conference (GLOBECOM), San Diego, USA, 2015. doi: 10.1109/GLO⁃ COM.2015.7417013. Manuscritp received: 2016⁃07⁃14 JIAO Jian (jiaojian@hitsz.edu.cn) received his PhD degree in communication engi⁃ neering from Harbin Institute of Technology (HIT) in 2011. He received his BS de⁃ gree in electrical engineering from Harbin Engineering University in 2005, and his MASc degree in information and communication engineering from HIT Shenzhen Graduate School in 2007. He is an assistant research fellow in the Department of Electrical and Information Engineering of HIT Shenzhen Graduate School. His cur⁃ rent interests include deep space communications, networking and channel coding. Rana Abbas (rana.abbas@sydney.edu.au) is currently a PhD student at the Centre of Excellence in Telecommunications, School of Electrical And Information Engi⁃ neering, The University Sydney, Australia, where she is a recipient of the Australian Postgraduate Awards scholarship and the Norman 1 Price scholarship. She received her bachelor’s degree in electrical engineering from The University of Balamand, Lebanon in 2012 and her master’s degree in electrical engineering from The Uni⁃ versity of Sydney, Australia in 2013. Her research interests include error control codes, machine⁃to⁃machine communications, random multiple access, and coopera⁃ tive networks. LI Yonghui (yonghui.li@sydney.edu.au) received his PhD degree in 2002 from Bei⁃ jing University of Aeronautics and Astronautics. From 1999 to 2003 he was affiliat⁃ ed with Linkair Communication Inc., where he held the position of project manager with responsibility for the design of physical layer solutions for LAS⁃CDMA system. Since 2003 he has been with the Centre of Excellence in Telecommunications, the University of Sydney, Australia. He is now an associate professor at the School of Electrical and Information Engineering, University of Sydney. He is the recipient of the Australian Queen Elizabeth II Fellowship in 2008 and the Australian Future Fellowship in 2012. His current research interests are in the area of wireless com⁃ munications, with a particular focus on MIMO, cooperative communications, coding techniques, and wireless sensor networks. He holds a number of patents granted and pending in these fields. He is an executive editor for European Transactions on Tele⁃ communications (ETT). He received best paper awards at the IEEE International Conference on Communications (ICC) 2014 and the IEEE Wireless Days Conferenc⁃ es (WD) 2014. ZHANG Qinyu (zqy@hit.edu.cn) received his bachelor’s degree in communication engineering from Harbin Institute of Technology (HIT) in 1994, and PhD degree in biomedical and electrical engineering from the University of Tokushima, Japan, in 2003. From 1999 to 2003, he was an assistant professor with the University of Tokushima. From 2003 to 2005, he was an associate professor with the Shenzhen Graduate School, HIT, and was the founding director of the Communication Engi⁃ neering Research Center with the School of Electronic and Information Engineering. Since 2005, he has been a full professor, and serves as the dean of the EIE School. He is on the Editorial Board of some academic journals, such as The Journal on Communications, KSII Transactions on Internet and Information Systems, and Sci⁃ ence China: Information Sciences. He was the TPC Co⁃Chair of the IEEE/CIC ICCC’ 15, the Symposium Co⁃Chair of the IEEE VTC’16 Spring, an Associate Chair for Fi⁃ nance of ICMMT’12, and the Symposium Co ⁃ Chair of CHINACOM’11. He has been a TPC Member for INFOCOM, ICC, GLOBECOM, WCNC, and other flagship conferences in communications. He was the Founding Chair of the IEEE Communi⁃ cations Society Shenzhen Chapter. He has received the National Science Fund for Distinguished Young Scholars, the Young and Middle ⁃ Aged Leading Scientist of China, and the Chinese New Century Excellent Talents in University, and obtained three scientific and technological awards from governments. His research interests include aerospace communications and networks, wireless communications and net⁃ works, cognitive radios, signal processing, and biomedical engineering. BiographiesBiographies
Multiple Access Technologies for Cellular MMultiple Access Technologies for Cellular M22MM CommunicationsCommunications Mahyar Shirvanimoghaddam and Sarah J. Johnson (School of Electrical Engineering and Computer Science, The University of Newcastle, NSW 2308, Australia) Abstract This paper reviews the multiple access techniques for machine⁃to⁃machine (M2M) communications in future wireless cellular net⁃ works. M2M communications aims at providing the communication infrastructure for the emerging Internet of Things (IoT), which will revolutionize the way we interact with our surrounding physical environment. We provide an overview of the multiple access strategies and explain their limitations when used for M2M communications. We show the throughput efficiency of different multi⁃ ple access techniques when used in coordinated and uncoordinated scenarios. Non⁃orthogonal multiple access (NOMA) is also shown to support a larger number of devices compared to orthogonal multiple access techniques, especially in uncoordinated sce⁃ narios. We also detail the issues and challenges of different multiple access techniques to be used for M2M applications in cellu⁃ lar networks. Internet of Things (IoT); massive access; machine⁃to⁃machine (M2M) communications; multiple access Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 006 http://www.cnki.net/kcms/detail/34.1294.TN.20161024.1001.002.html, published online October 24, 2016 1 Introduction achine⁃to⁃machine (M2M) communications is expected to become an integral part of cellular networks in the near future. In M2M communi⁃ cations a large number of multi⁃role devices, such as sensors and actuators, wish to communicate with each other and with the underlying data transport infrastructure. To enable such massive communication in wireless networks, ma⁃ jor shifts from current protocols and designs are necessary [1]. Current wireless networks that have been mainly designed and engineered for human⁃based applications, such as voice, vid⁃ eo, and data, cannot be used for M2M communications due to the different nature of their traffic and service requirements [2]. These differences have posed many questions and chal⁃ lenges in the communication society, in both industry and re⁃ search sectors. M2M communications aims at providing the communication infrastructure for emerging Internet of Things (IoT) and in⁃ volves the enabling of seamless information exchange between autonomous devices without any human intervention. M2M de⁃ vices can be either stationary, such as smart meters, or mobile, such as fleet management devices, and they can connect to the network infrastructure using either wired or wireless links. Key challenges of massive M2M communications can be listed as follows [3]: 1) Device cost: For the mass deployment of M2M communica⁃ tions, low cost devices are necessary for most use cases. 2) Battery life: Most M2M devices are battery operated and re⁃ placing batteries is not practical for many applications. 3) Coverage: Deep indoor and regional connectivity is a re⁃ quirement for many applications. 4) Scalability: Network capacity must be easily scaled to han⁃ dle a large number of devices forecasted to arise in the near future. 5) Diversity: Cellular systems must be able to support diverse service requirements for different use cases, ranging from static sensor networks to tracking systems. The wired solutions include cable, xDSL, and optical fiber, and can provide high reliability, high data rate, short delay, and high security. However, they are cost ineffective and do not support mobility and scalability; therefore, not appropriate for M2M applications [3]. On the other hand, Wireless capil⁃ lary (i.e., short range) solutions, such as WLAN and ZigBee, can provide low cost infrastructure and scalability for most M2M applications, but they suffer from small coverage, low da⁃ ta rate, weak security, and severe interference. Wireless cellu⁃ lar, i.e., GSM, GPRS, 3G, LTE⁃A, WiMAX, etc., however of⁃ fers excellent coverage, mobility and scalability support, and good security, and the fact that the infrastructure already exists M Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS42
makes it a promising solution for M2M communications [3]. Therefore, our focus in this paper is on wireless cellular solu⁃ tions for M2M communications. The mobile industry is standardizing several low power tech⁃ nologies, such as extended coverage GSM (EC⁃GSM), LTE for machine⁃type communication (LTE⁃M), and narrow band IoT (NB⁃IoT). Since GSM is still the dominant mobile technology in many markets, it is expected to play a key role in the IoT due to its global coverage and cost advantages. EC⁃GSM en⁃ ables coverage improvements of up to 20 dB with respect to GPRS on the 900 MHz band [4]. A combined capacity of up to 50,000 devices per cell on a single transceiver has been achieved by defining new control and data channels mapped over legacy GSM. LTE⁃M brings new power saving functional⁃ ity suitable for serving a variety of IoT applications, which ex⁃ tend battery life to 10 years or more. NB⁃IoT is a self contained carrier that can be deployed with a system bandwidth of 200 kHz. These initiatives were undertaken in 3GPP Release 13 for M2M specific applications [3]. Despite all these efforts, further improvements are required in the way that devices communicate with the base station to support a large number of devices and not jeopardizing the hu⁃ man⁃based communication quality. The multiple access (MA) techniques have been identified as a key area where improve⁃ ments for M2M communications are needed. The fact that the radio access strategy in LTE is still based on random access mechanisms turns it into a potential bottleneck for the perfor⁃ mance of cellular networks when the number of M2M devices grows [5]. Moreover, radio resources are orthogonally allocated to the users/devices in the current LTE standards, which is not effective for M2M communications when the number of devic⁃ es goes very large, due to the limited number of radio resourc⁃ es [6]. In this paper, we consider several multiple access technolo⁃ gies and show their performance in coordinated and uncoordi⁃ nated scenarios. Overall, coordinated strategies outperform un⁃ coordinated ones as in coordinated strategies the base station can optimally allocate the radio resources between the devices and support a larger number of devices. We also show that the non⁃orthogonal multiple access (NOMA) scheme achieves the highest throughput in both coordinated and uncoordinated strategies, whereas frequency division multiple access (FDMA) has comparable performance in coordinated scenarios. This suggests that FDMA can be effectively used in coordinated sce⁃ narios to achieve maximum throughput (this has been consid⁃ ered by 3GPP for M2M communications in the NB⁃IoT solu⁃ tion), while in uncoordinated scenarios, NOMA strategies must be considered to effectively support a large number of devices and use the available radio resources in an efficient manner. The remainder of the paper is organized as follows. Section 2 represents the unique characteristics of M2M communications and its challenges in cellular networks. In Section 3, we pro⁃ vide an overview on different multiple access technologies. Co⁃ ordinated and uncoordinated MA techniques are represented in Section 4 and 5, respectively, where we characterize their maximum achievable throughput. Practical issues for imple⁃ menting MA techniques for M2M communications are present⁃ ed in Section 6. Finally, Section 7 concludes the paper. 2 M2M Communications: Characteristics and Challenges Until recently, cellular systems have been designed and en⁃ gineered for human based applications, such as voice, video, and data, with a higher demand on downlink. M2M communi⁃ cations however has different traffic characteristics that in⁃ clude small and infrequent data generated from a very large number of devices, which imposes a higher traffic volume on the uplink. In addition, M2M applications have very diverse service requirements. For instance, in alarm signal applica⁃ tions, a small⁃size message must be delivered to the base sta⁃ tion (BS) within 10 ms, while in other applications, such as smart metering, the delay of up to several hours or even a day is tolerable [7]. Due to limited radio resources and the large number of de⁃ vices involved in M2M communications, wireless networks should minimize the time wasted due to collisions or exchang⁃ ing control messages. The throughput must be large enough to support a large number of devices. Control overhead must be minimized as the payload data in many M2M applications is of small size and the control overhead of conventional approaches in current cellular systems results in an inefficient M2M com⁃ munications [8]. In fact, if the control overhead of a protocol is large, the effective throughput is degraded even though the physical data rate may not be affected. It is also required that the effective throughput remain high irrespective of the traffic level [9]. Scalability is another challenge in M2M communications as it is expected that a large number of devices arise in M2M sce⁃ narios. These devices have dynamic behaviour, i.e., entering and leaving the network frequently; thus the network must easi⁃ ly tolerate the changes in the node density with little control in⁃ formation exchange. Energy efficiency is also one of the most important challenges in M2M communications, as devices in many M2M applications are battery operated and long life times are expected for these devices [10]. More specifically, the energy spent on radio access and data transmission in M2M communications must be minimized to improve the ener⁃ gy efficiency in a large scale. For instance, in high load scenar⁃ ios, exchanging control information may consume more than 50% of the total energy in IEEE 802.11 MAC protocol, which shows its ineffectiveness in dense M2M applications [9]. In many M2M applications, the network latency is a critical factor that determines the effectiveness of the service. For in⁃ stance, in intelligent transportation systems and healthcare monitoring, it is highly important to make the communication Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 43
reliable and fast. Channel access delay then needs to be mini⁃ mized to reduce the overall latency in M2M communications. Moreover, in cellular systems, human⁃to⁃human (H2H) devices coexist with M2M devices, and the communication protocol must be designed in such a way to not jeopardize the quality of human⁃based communications. Resource management and allo⁃ cation are challenging tasks in M2M communications which co⁃ exist with H2H applications, as H2H applications have com⁃ pletely different service requirements [11]. These unique characteristics of M2M communications intro⁃ duce a number of networking challenges in cellular networks. The fundamental issue arises from the fact that most M2M ap⁃ plications involve a huge number of devices. The question is then how the available radio resources have to be shared among devices such that their service requirements are simul⁃ taneously met. 3 Overview of Multiple Access Techniques for M2M Communications Multiple access techniques can be divided into two broad categories, depending on how the radio resources are allocated to the devices. These include 1) uncoordinated, where the de⁃ vices transmit data using slotted random access and there is no need to establish dedicated resources, and 2) coordinated, where devices transmit on separate resources pre⁃allocated by the base station. In coordinated MA, the base station knows a priori the set of devices that have data to transmit. The BS can also acquire channel state information (CSI) of these devices based on which it allocates resources to optimize system throughput. CSI to the devices can be obtained by each device sending an upload pilot signal. Multiple access techniques can be also divided into orthogo⁃ nal and non⁃orthogonal approaches. In orthogonal MA (OMA), radio resources are orthogonally divided between devices, where the signals from different devices are not overlapped with each other. Instances of OMA (Fig. 1) are time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFD⁃ MA), and single carrier FDMA (SCFDMA). First and second generation cellular systems are mainly developed using OMA approaches, which avoid intra ⁃ cell interference and simplify air interface design. However, OMA approaches have no abili⁃ ty to combat the inter⁃cell interference; therefore careful cell planning and interference management techniques are re⁃ quired to solve the interference problem [12]. Non⁃orthogonal MA (NOMA) techniques have been adopted in second and third generation cellular systems. NOMA allows overlapping among the signals from different devices by ex⁃ ploiting power domain, code domain, and interleaver pattern. Code division multiple access (CDMA) is the well⁃known exam⁃ ple of NOMA which has been adopted in second and third gen⁃ eration cellular systems. CDMA is robust against inter⁃cell in⁃ terference, but suffers from intra⁃cell interference [12]. CDMA is also not suitable for data services which require high single⁃ user rates. Rather than CDMA which exploits code domain, NOMA in current study in general exploits power domain. NO⁃ MA is also shown to provide better performance than OMA [12]. In NOMA, signals from multiple users are superimposed in the power⁃domain and successive interference cancellation (SIC) is used at the BS to decode the messages. It is also shown that NOMA can achieve the multiuser capacity region both in the uplink and downlink [12]. In this paper, we compare NOMA and OMA strategies in both coordinated and uncoordinated scenarios, and show that NOMA can provide the system with higher capacity to support M2M devices, especially in the uncoordinated scenario. This is achieved by exploiting the power domain, rather than frequency⁃ domain or time⁃domain as in FDMA and TDMA, respectively. CDMA: code division multiple access FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple access Figure 1. ▶ Different multiple access schemes. Time Power TDMA Frequency (a) Time Power FDMA Frequency (b) Time Power CDMA Frequency (c) Time Power NOMA Frequency (d) October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS44 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
For the analysis in this paper, we consider a single cell cen⁃ tered by base station and devices uniformly distributed around it in a circular region with radius R . The uplink load seen by the base station is modeled by a Poisson point process with mean λ arrivals per second. We further assume a time slotted system with a slot duration of τs . We perform our analysis on a typical radio resource with slot duration τs and bandwidth W . Each device packet is assumed to have a payload of L bits. The channel from a device located at distance r from the base station is modelled by g =(r/R) -γ , where γ denotes the path loss exponent and we ignore shadowing and small scale fading [13]. The received signal⁃to⁃noise ratio (SNR) for a de⁃ vice transmitting with power Pt over bandwidth W is then given by [14]: μr = Pt Pmax W Wt μg , (1) where Pmax is the maximum transmit power and μ is the ref⁃ erence SNR, defined as the average received SNR from a de⁃ vice transmitting at maximum power Pmax over bandwidth W located at the cell edge. Without loss of generality, we assume ordered channel gain g1 ≥ g2 ≥ … ≥ gK , where K is the num⁃ ber of devices. 4 Coordinated Multiple Access Strategies In this section, we consider the coordinated multiple access strategies, i.e., TDMA, FDMA, and NOMA, and compare their throughput efficiency. In this section, we assume that the BS has perfect CSI to all the devices. 4.1 Optimal Throughput FDMA Strategy In FDMA, the spectrum is partitioned between the devices and each device will transmit in a portion of the spectrum. Fig. 1b shows the FDMA strategy, where the whole spectrum has been divided between 6 devices, and each device will use its allocated bandwidth for the data transmission. Using Shannon’s capacity formula, the minimum bandwidth required for the transmission of L bits by the ith device over time τs is given by the solution of the following equation [13]: L τsWmini = log2 æ è çç ö ø ÷÷1 + μ W Wmini gi . (2) The maximum load that can be supported in a resource block of duration τs and bandwidth W is given by: Kmax = max ì í î ü ý þ K:∑i = 1 K Wmini ≤ W . (3) 4.2 Optimal Throughput TDMA Strategy In TDMA, the whole spectrum is used by each device in sep⁃ arate time instances. Fig. 1a shows the TDMA scheme, where the same time duration is allocated for 6 devices, and each de⁃ vice will only transmit in its allocated time slot using the whole spectrum. TDMA is an interesting MA strategy due to its sim⁃ plicity, but it is not efficient for M2M applications with a large number of devices. Moreover, with increasing the number of devices, each device’s transmission will be delayed which is not appropriate for delay⁃sensitive M2M applications. Assuming a capacity approaching code and using Shannon’ s capacity equation, the time required for a device located at distance r from the base station to deliver its packet to the destination is given by [13]: τ ≥ L W log2(1 + μr) , (4) and the minimum time required to deliver the message is ob⁃ tained when the device is transmitting with full power Pmax : τmini = L W log2(1 + μgi) . (5) Similar to FDMA, the maximum number of devices which can be supported in a resource block of duration τs and band⁃ width W can then be found as follows: Kmax = max ì í î ü ý þ K:∑i = 1 K τmini ≤ τs . (6) 4.3 Optimal Throughput NOMA Strategy Unlike TDMA and FDMA, devices in the NOMA strategies are assumed to transmit in the same resource block and their transmissions interfere with each other. We assume that the BS perform successive interference cancellation (SIC), where it starts the decoding with the device with the largest channel gain and treats the signals from other devices as additive noise. After decoding the first device, its signal will be removed from the received signal and the BS continues the decoding for the second device and treats the remainder as additive noise. This process is continued until all the devices are successfully de⁃ coded. Under this decoding strategy, the Shannon Capacity for⁃ mula for the ithdevice is given by: L = Wτs log2 æ è ç ç ö ø ÷ ÷ 1 + Pi μgi 1 +∑j = i + 1 K Pi μgj , (7) and the required transmit power can be calculated as follows: Pi μgi = æ è çç ö ø ÷÷2 L Wτs - 1 æ è çç ö ø ÷÷1 + ∑j = i + 1 K Pi μgj . (8) By substituting, i = K , we have: PK = 2 L Wτs - 1 μgK , (9) Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 45
and by going backwards and finding the transmit power for the ith device, we have: Pi = 2 (K - i)L Wτs æ è çç ö ø ÷÷2 L Wτs - 1 μgi . (10) The maximum load that the BS can support in a resource block of bandwidth W and duration τs can be found as fol⁃ lows: Kmax = max{ }K:Pi ≤ Pmax for i = 1,2,…,K . (11) 4.4 Comparison Between Coordinated MA Techniques Fig. 2 shows the maximum throughput versus arrival rates for different coordinated MA techniques. NOMA can achieve very high throughput when the arrival rate is very large. FDMA performs very close to the NOMA strategy and can support all the active devices for the arrival rates up to 14,000 packets per second. The advantage of NOMA comes from the fact that the devices can use the whole spectrum thus achieving a higher throughput compared to FDMA. Only a fraction of the spec⁃ trum is used by each device in FDMA. Also, TDMA cannot support many devices, which shows that it is not an effective MA strategy for M2M communications. It is clear that the time slot duration τi and subchannel bandwidth Wi cannot be arbitrarily small in TDMA and FD⁃ MA, respectively. As can be seen in Fig. 2, if we put some con⁃ straints on the minimum time slot duration or subchannel band⁃ width, the number of devices which can be supported by FD⁃ MA and TDMA would be limited. For example, if the minimum time slot duration for TDMA is set to be 1 ms, the maximum number of devices which can be supported in a time slot of du⁃ ration 1 s is 1000. Similarly, if the minimum subchannel band⁃ width in FDMA is set to be 1 kHz, the maximum number of de⁃ vices which can be supported by the BS will be 1000. This shows that in practical systems where the minimum subchan⁃ nel bandwidth and time slot duration cannot be very small, the maximum throughput of TDMA and FDMA will be limited. In such cases, NOMA can bring more benefits to the system as it can support a larger number of devices without dividing the ra⁃ dio resource into subchannels or time slots. 5 Uncoordinated Multiple Access Strategies In this section, we assume that the base station does not have CSI to the devices, which is particularly the case for M2M communications with a large number of devices, where it is almost impractical for the base station to estimate the chan⁃ nel to every device with random activities. The only informa⁃ tion we assume is available at the BS is the traffic load which can be obtained using different load estimation algorithms. 5.1 Uncoordinated FDMA In this scheme, we assume that the base station chooses a se⁃ lection probability pc and broadcasts this information to the devices. Each device which has data to transmit only switches on its transmitter with probability pc . We refer to these devic⁃ es as active devices. Let Nc denote the number of active de⁃ vice. We further assume that the BS uniformly divides the spectrum into Nw subchannels, and each device randomly chooses a subchannel for its transmission. We also assume that each device only transmits on a selected subchannel if the max⁃ imum transmit power required to deliver its message to the BS is less than Pmax , assuming no collision on the selected sub⁃ channel. More specifically, the ith device is transmitting in a subchannel if the following condition holds: æ è ç ç ö ø ÷ ÷2 LNw Wτs - 1 ≤ Nw μgi . (12) Therefore, the probability that a device is transmitting can be calculated as follows: P æ è ç ç ö ø ÷ ÷ æ è ç ç ö ø ÷ ÷2 LNw Wτs - 1 ≤ Nw μgi = æ è ç çç ç ö ø ÷ ÷÷ ÷ Nw μ 2 LNw Wτs - 1 2 γ , (13) which is due to the fact that the devices are uniformly distribut⁃ ed in the cell and the probability that a device is located at dis⁃ tance r is given by 2r R2 . The average number of active de⁃ vices which can deliver their messages, considering no colli⁃ sion, can be found as follows: Np = Nc æ è ç çç ç ö ø ÷ ÷÷ ÷ Nw μ 2 LNw Wτs - 1 2 γ . (14) As the devices randomly choose a sub ⁃ channel for their FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple acces ▲Figure 2. Average throughput versus arrival rates for different coordi⁃ nated MA techniques. Total available bandwidth is W = 1 MHz, time slot duration is τs = 1 sec, and the packet length is L = 1000 bits. 18􀆯000 16􀆯000 14􀆯000 12􀆯000 10􀆯000 8􀆯000 6000 4000 2000 0 104 103 102 101 Arrival rate (λ) Throughput(packets/s) FDMA NOMA TDMA FDMA, Wmin =1 KHz TDMA, τmin =1 ms October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS46 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
transmission, more than one device can select the same sub⁃ channel, which leads to collision. The base station cannot de⁃ code any of the devices that are simultaneously transmitting on that particular subchannel. The probability of collision can be calculated as follows [14]: Pc = 1 - æ è ç ö ø ÷1 - 1 Nw Np - 1 . (15) The average number of devices which can successfully deliv⁃ er their messages to the BS is given by NP(1 - Pc) . We assume that the BS finds the optimal values for pc and Nw such that the number of devices which can be supported by the BS is maximized. 5.2 Uncoordinated TDMA Similar to FDMA, we assume that the BS assigns an access probability pc to the devices. Let Nc denote the number of ac⁃ tive device. We further assume that the BS uniformly divides the time into Nt time slots, and each device randomly chooses a time slot for its transmission. We also assume that the each device only transmits in a selected time slot if the maximum transmit power required to deliver its message to the BS is less than Pmax , assuming no collision on the selected time slot. More specifically, the ith device is transmitting in a time slot, if the following condition holds: æ è ç ç ö ø ÷ ÷2 LNt Wτs - 1 ≤ μgi . (16) Therefore, the probability that a device is transmitting can be calculated as follows: p æ è ç ç ö ø ÷ ÷ æ è ç ç ö ø ÷ ÷2 LNt Wτs - 1 ≤ μgi = æ è ç çç ç ö ø ÷ ÷÷ ÷ μ 2 LNt Wτs - 1 2 γ , (17) which is due to the fact that the devices are uniformly distribut⁃ ed in the cell and the probability that a device is located at dis⁃ tance r is given by 2r/R2 . The average number of active de⁃ vices which can deliver their messages, considering no colli⁃ sion, can be found as follows: Np = Nc æ è ç çç ç ö ø ÷ ÷÷ ÷ μ 2 LNt Wτs - 1 2 γ . (18) The average number of devices which can successfully deliv⁃ er their messages to the BS is given by NP(1 - Pc) , where Pc is given by (15) by replacing Nw with Nt . We assume that the BS finds the optimal values for Pc and Nt such that the num⁃ ber of devices which can be supported by the BS is maximized. 5.3 Uncoordinated NOMA We consider that each device performs power control such that the received SNR at the BS for each device is γ0 . A de⁃ vice will only transmit if and only if the transmit power re⁃ quired to achieve the SNR γ0 at the base station is less than Pmax . Let Np denote the number of devices which can trans⁃ mit, i.e., their required transmit power is less than Pmax . The achievable rate for the devices considering the successive in⁃ terference cancellation at the BS can be calculated as follows: Rmin = log2 æ è çç ö ø ÷÷1 + γ0 1 +(Np - 1)γ0 . (19) A message of length L can be delivered by Np devices if WτsRmin ≥ L . Using (19), the required SNR γ0 for successfu⁃ lly delivering a message of length L at the BS is derived as fol⁃ lows: γ0 = 1 1 2 L Wτs - 1 - Np . (20) Accordingly, the number of devices which can be supported at the BS is upper bounded as follows: Np ≤ 1 2 L Wτs - 1 . (21) 5.4 Comparison Between Uncoordinated MA Techniques Fig. 3 shows the maximum number of devices which can be supported by the base station versus different arrival rates for uncoordinated MA strategies. The minimum time slot duration FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple acces ▲Figure 3. Average throughput versus arrival rates for different unco⁃ ordinated MA techniques. Total available bandwidth is W = 1 MHz, time slot duration is τs = 1 s, and the packet length is L = 1000 bits. The minimum time slot duration for TDMA is considered to be 1 ms and the minimum subchannel bandwidth in FDMA is considered to be 1 kHz. 1400 104 Arrival rate (λ) Throughput(packets/s) 103 102 101 1200 1000 800 600 400 200 NOMA TDMA FDMA Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 47
for TDMA is considered to be 1 ms, which corresponds to Nt = 1000 , and the minimum subchannel bandwidth in FDMA is considered to be 1 kHz, which corresponds to Nw = 1000 . As shown in this figure, NOMA can support much larger num⁃ ber of devices compared to the FDMA and TDMA strategies. This is due to the high collision probability in uncoordinated FDMA and TDMA in high arrival rates, while in NOMA a large number of devices can simultaneously transmit in the same resource block by exploiting the power domain. This shows the advantage of NOMA in uncoordinated scenarios. Therefore, NOMA can be an excellent choice for M2M applica⁃ tions with a large number of devices and random traffic. More⁃ over, FDMA outperforms TDMA in moderate loads but they perform similarly in low and high arrival rates. It is important to note that in NOMA the constraints on mini⁃ mum time slot duration or subchannel bandwidth do not affect the throughput efficiency. This is due to the fact that in NOMA all the devices are transmitting in the whole bandwidth in all slot duration. One could consider some limitations in the mini⁃ mum power difference between the devices, which mostly de⁃ pends on the hardware capability to distinguish different power levels which is out of scope of this paper. 6 Practical Considerations of Massive NOMA for M2M Communications and Future Directions NOMA can bring many benefits to cellular systems which in⁃ clude, but are not limited to, the following. NOMA can effec⁃ tively use the spectrum and provide higher throughput by ex⁃ ploiting power domain and non⁃orthogonal multiplexing. It also provides robust performance gain in high mobility scenarios. NOMA is also compatible with OFDMA and can be applied on top of OFDMA for downlink and SC⁃FDMA for uplink. It can be also combined with multi ⁃ antenna techniques to improve the system performance. Using NOMA, multiple users can si⁃ multaneously transmit in the same subband without being iden⁃ tified by the destination a priori. The devices can attach their terminal identities to their messages and the base station can identify the devices after decoding their messages. The RA pro⁃ cedure can be eliminated and therefore the access delay and signaling overhead will be significantly reduced [12]. Although NOMA can improve spectrum efficiency and sys⁃ tem capacity, there are many practical challenges for this tech⁃ nology to be potentially used in real wireless systems for M2M communications. Here, we outline the main practical consider⁃ ation of massive NOMA for M2M communications. First, in uncoordinated strategies the base station needs to estimate the arrival rate to effectively detect the devices. In un⁃ coordinated FDMA, the BS needs to know the number of devic⁃ es to find the optimal access probability and the number of sub⁃ bands. In NOMA, the problem is much more complicated as the BS runs the SIC and needs to know the number of devices with different power levels. For simplicity, one could consider that the devices perform power control such that only one pow⁃ er level is received at the BS, but this may have some implica⁃ tions on the actual performance of the system as the overall sys⁃ tem data rate will be dominated by the device with the lowest SINR; and thus will not effectively use the available spectrum. However, even with this simplification and suboptimal power allocations, NOMA outperforms FDMA in uncoordinated sce⁃ narios and can support a large number of devices under high loads. Second, channel estimation at the devices is necessary in un⁃ coordinated strategies employing NOMA techniques. This is due to the fact that the devices are not identified by the BS be⁃ forehand and they are simultaneously transmitting at the same resource block. To enable the BS to detect the devices and de⁃ code their messages, the devices need to perform channel esti⁃ mation and adjust their power so the BS only deals with some known power levels rather than unknown channel gains. On the other hand, to effectively perform SIC, the multipath effect must be carefully taken into consideration as multipath will spread the signal over time, which decreases the effective sig⁃ nal to noise ratio for each device, and makes the BS unable to perform SIC. One can consider several techniques, such as time reversion [15], to eliminate the multipath effect by treat⁃ ing the channel between each device and the BS as the natural match filter. This has been shown an effective way to combat multipath effect for several fixed location M2M applications [16]. Third, NOMA requires synchronization among the devices at the symbol level. This is very challenging as providing time synchronicity between a large number of devices distributed in a large environment is tedious. However, the devices in many M2M applications are deployed in fixed locations, so each de⁃ vice can determine its propagation delay using different dis⁃ tance estimation strategies or using control information periodi⁃ cally sent by the BS. Fourth, as the number of devices transmitting in each re⁃ source block in uncoordinated NOMA is random, the physical data rate cannot be determined beforehand. One could consid⁃ er a very low rate code at each device, but it might be ineffi⁃ cient when used in low⁃to⁃moderate loads. An effective strate⁃ gy is then to use rateless codes to automatically adapt to the traffic condition. Authors in [17] have proposed to use analog fountain codes to enable massive multiple access for M2M communications and achieve very high throughput even in high loads. Moreover, as shown in [18], binary rateless codes can be effectively used to enable NOMA for M2M communica⁃ tions. These coding strategies were mainly proposed to maxi⁃ mize the throughput in M2M communications and for delay sensitive applications with very short messages, more ad⁃ vanced coding techniques should be combined with rateless ideas to enable low latency massive multiple access in M2M communications. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS48 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
Last but not least, NOMA is still in its early stage of its de⁃ velopment and more research work must be done to clearly identify its effectiveness in real scenarios. From an information theoretic point of view, it achieves the capacity region of the multiple access channel and thus is optimal in terms of throughput. But in real M2M applications when NOMA is joint⁃ ly considered with medium access control layer in real world scenarios, it might not be as efficient as OMA techniques, which have been considered as effective multiple access tech⁃ niques for a long time and several issues and challenges have been solved over the years. 7 Conclusions In this paper, we provided an overview of multiple access techniques for emerging machine⁃to⁃machine communications in cellular systems. The unique challenges of M2M communi⁃ cations were represented, where we identified scalability, ener⁃ gy efficiency, and reliability, as the most important features for every potential multiple access technology which is considered for M2M communications. We provided a simple study on the throughput efficiency of multiple access techniques in both co⁃ ordinated and uncoordinated scenarios. NOMA was shown to provide the highest throughput in both coordinated and uncoor⁃ dinated scenarios, whereas FDMA has shown comparable per⁃ formance with NOMA in coordinated scenarios. NOMA is shown to be scalable in uncoordinated scenarios and can sup⁃ port a large number of devices. It can be also combined with different access management schemes to control the load over the base station. We also provided some of the practical issues in NOMA which needed to be considered for the use of NOMA strategies for M2M communications in future cellular systems. References [1] H. Shariatmadari, R. Ratasuk, S. Iraji, et al.,“Machine⁃type communications: current status and future perspectives toward 5G systems,”IEEE Commununica⁃ tions Magazine, vol. 53, no. 9, pp. 10-17, Sept. 2015. doi: 10.1109/MCOM.2015. 7263367. [2] G. Wu, S. Talwar, K. Johnsson, N. Himayat, and K. Johnson,“M2M: From mo⁃ bile to embedded internet,”IEEE Commununications Magazine, vol. 49, no. 4, pp. 36-43, Apr. 2011. doi: 10.1109/ MCOM.2011.5741144. [3] Ericsson. (Jan. 2016). Cellular networks for massive IoT. Tech. Rep. Uen 284 23⁃ 3278 [Online]. Available: https://www.ericsson.com/res/docs/whitepapers/wp iot. pdf [4] Service Requirements for Machine⁃Type Communications (MTC); Stage 1, 3GPP TS 22.368 V.13.0.0, Jun. 2014. [5] A. Laya, L. Alonso, and J. Alonso⁃Zarate,“Is the random access channel of LTE and LTE⁃A suitable for M2M communications? a survey of alternatives,”IEEE Commununications Surveys & Tutorials, vol. 16, no. 1, pp. 4-16, 2014. doi: 10.1109/SURV.2013.111313.00244. [6] G. Naddafzadeh⁃Shirazi, L. Lampe, G. Vos, and S. Bennett,“Coverage enhance⁃ ment techniques for machine ⁃ to ⁃ machine communications over LTE,”IEEE Communications Magazine, vol. 53, no. 7, pp. 192-200, Jul. 2015. doi: 10.1109/ MCOM.2015.7158285. [7] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi,“Internet of things for smart cities,”IEEE Internet of Things Jounral, vol. 1, no. 1, pp. 22-32, Feb. 2014. doi: 10.1109/JIOT.2014.2306328. [8] D. Wiriaatmadja and K. W. Choi,“Hybrid random access and data transmission protocol for machine⁃to⁃machine communications in cellular networks,”IEEE Transactions on Wireless Communications, vol. 14, no. 1, pp. 33-46, Jan. 2015. doi: 10.1109/TWC.2014.2328491. [9] A. Rajandekar and B. Sikdar,“A survey of MAC layer issues and protocols for machine⁃to⁃machine communications,”IEEE Internet of Things Journal, vol. 2, no. 2, pp. 175-186, Apr. 2015. doi: 10.1109/JIOT.2015.2394438. [10] M. Hasan, E. Hossain, and D. Niyato,“Random access for machine⁃to⁃machine communication in LTE ⁃ advanced networks: issues and approaches,”IEEE Communications Magazine, vol. 51, no. 6, pp. 86-93, Jun. 2013. doi: 10.1109/ MCOM.2013.6525600. [11] K. Zheng, S. Ou, J. Alonso⁃Zarate, et al.,“Challenges of massive access in high⁃ ly dense LTE⁃advanced networks with machine⁃to⁃machine communications,” IEEE Wireless Communications Magazine, vol. 21, no. 3, pp. 12-18, Jun. 2014. doi: 10.1109/MWC.2014.6845044. [12] A. Li, Y. Lan, X. Chen, and H. Jiang,“Non⁃orthogonal multiple access (NO⁃ MA) for future downlink radio access of 5G,”China Communications, vol. 12, Supplement, pp. 28-37, Dec. 2015. [13] H. S. Dhillon, H. C. Huang, H. Viswanathan, and R. A. Valenzuela,“Power⁃ef⁃ ficient system design for cellular⁃based machine⁃to⁃machine communications,” IEEE Transactions on Wireless Communications, vol. 12, no. 11, pp. 5740- 5753, Nov. 2013. doi: 10.1109/TWC.2013.100713.130025. [14] H. S. Dhillon, H. C. Huang, H. Viswanathan, and R. A. Valenzuela,“Funda⁃ mentals of throughput maximization with random arrivals for M2M communica⁃ tions,”IEEE Transactions on Communications, vol. 62, no. 11, pp. 4094-4109, Nov. 2014. doi: 10.1109/TCOMM.2014.2359222. [15] B. Wang, Y. Wu, F. Han, Y. H. Yang, and K. J. R. Liu,“Green wireless com⁃ munications: A time ⁃ reversal paradigm,”IEEE Journal on Selected Areas in Communications, vol. 29, no. 8, pp. 1698-1710, Sept. 2011. doi: 10.1109/JSAC. 2011.110918. [16] Y. Chen, F. Han, Y. H. Yang, et al.,“Time ⁃ reversal wireless paradigm for green internet of things: An overview,”IEEE Internet of Things Jounral, vol. 1, no. 1, pp. 81-98, Feb. 2014. doi: 10.1109/JIOT.2014.2308838. [17] M. Shirvanimoghaddam, Y. Li, M. Dohler, B. Vucetic, and S. Feng,“Probabilis⁃ tic rateless multiple access for machine ⁃ to ⁃ machine communication,”IEEE Transactions on Wireless Communications, vol. 14, no. 12, pp. 6815 - 6826, Dec. 2015. doi: 10.1109/TWC.2015.2460254. [18] M. Shirvanimoghaddam, S. J. Johnson, and M. Dohler. (2016). An efficient mas⁃ sive access strategy based on superposition Raptor codes for M2M communica⁃ tions. CoRR [Online]. Available: http://arxiv.org/pdf/1602.05671v1.pdf Manuscript received: 2016⁃06⁃30 Mahyar Shirvanimoghaddam (fmahyar.shirvanimoghaddam@newcastle.edu.au) re⁃ ceived the BSc degree with 1st Class Honours from University of Tehran, Iran, in September 2008, the MSc degree with 1st Class Honours from Sharif University of Technology, Iran, in October 2010, and the PhD degree from The University of Syd⁃ ney, Australia, in January 2015, all in electrical engineering. He then held a re⁃ search assistant position at the Centre of Excellence in Telecommunications, School of Electrical and Information Engineering, The University of Sydney, before coming to the University of Newcastle, Australia, where he is now a postdoctoral research as⁃ sociate at the School of Electrical Engineering and Computer Science. His general research interests include channel coding techniques, cooperative communications, compressed sensing, machine⁃to⁃machine communications, and wireless sensor net⁃ works. Sarah Johnson (sarah.johnsong@newcastle.edu.au) received the BE (Hons) degree in electrical engineering in 2000, and PhD in 2004, both from the University of Newcastle, Australia. She then held a postdoctoral position with the Wireless Signal Processing Program, National ICT Australia before returning to the University of Newcastle where she is now an Australian Research Council Future Fellow. Her re⁃ search interests are in the fields of error correction coding and network information theory. She is the author of a book on iterative error correction published by Cam⁃ bridge University Press. BiographiesBiographies Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 49
Software Defined OpticalSoftware Defined Optical Networks and ItsNetworks and Its Innovation EnvironmentInnovation Environment LI Yajie1 , ZHAO Yongli1 , ZHANG Jie1 , WANG Dajiang2 , and WANG Jiayu2 (1. Beijing University of Posts and Telecommunications, Beijing 100876, China) 2. ZTE Corporation, Shenzhen 518057, China) Software defined optical networks (SDONs) integrate software defined technology with optical communication networks and represent the promising development trend of future optical networks. The key technologies for SDONs include software⁃ defined optical transmission, switching, and networking. The main features include control and transport separation, hard⁃ ware universalization, protocol standardization, controllable optical network, and flexible optical network applications. This paper introduces software defined optical networks and its innovation environment, in terms of network architecture, protocol extension solution, experiment platform and typical applications. Batch testing has been conducted to evaluate the performance of this SDON testbed. The results show that the SDON testbed has good scalability in different sizes. Meanwhile, we notice that controller output bandwidth has great influence on lightpath setup delay. optical networks; software defined networking; innovation en⁃ vironment DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 007 http://www.cnki.net/kcms/detail/34.1294.TN.20160928.1048.002.html, published online September 28, 2016 Abstract Keywords W 1 Introduction ith the emerging of new network services, the interaction of all kinds of information grows day by day. It is an eternal theme for optical networks to satisfy the transmission demands for high speed, wide broadband, large capacity, and long⁃dis⁃ tance transmission. The changes of servics properties brings a new challenge: intelligence of optical networks. For example, high burst services require optical networks to have dynamic adaptability; large⁃scale networking requires optical networks to be scalable; and variable bandwidth provisioning requires optical networks to be flexible. To realize the intelligent opti⁃ cal network, the industry has carried out a long⁃term research and exploration. So far, the intelligent optical network has gone through three important stages of development. 1) Automatic Switching Optical Networks (ASON) An ASON is divided into three planes: the transmission plane, control plane and management plane. With the control plane based on Generalized Multi ⁃ Protocol Label Switching (GMPLS) protocol, ASON adopts distributed signaling and rout⁃ ing to solve the connection control problem and satisfy the function demands of automatic switching [1]- [4]. However, ASON has obvious limitations in many aspects, including large⁃ scale connection control, complex path calculation, network openness, devices interworking, and cost reduction. Besides, the GMPLS standard is very complex, which greatly affects the application and promotion of ASON. 2) Path Computation Element (PCE) Architecture for ASON In order to better adapt to the characteristics of multi⁃layer multi⁃domain large⁃scale optical networks, the Internet Engi⁃ neering Task Force (IETF) separates the path calculation func⁃ tion from the control layer and develops an independent unit, i.e., PCE [5]-[8]. In order to satisfy the function demands of large⁃scale multi⁃layer/domain, PCE adopts the distributed sig⁃ naling and centralized routing to solve the problem of path se⁃ lection and calculation for inter⁃layer and inter⁃domain path. However, with unitary function of path calculation, PCE needs to cooperate with other technology in applications. 3) Software Defined Optical Networks (SDONs) SDONs can offer a unified schedule and control for various kinds of optical layer resources according to the requirements of users or operators. With programmable software and dynam⁃ ic customization, the SDON solves the problem of function ex⁃ tension and therefore realizes rapid response to requests, effi⁃ cient utilization of resources and flexible service provisioning. The SDON well supports service processing, control strategy and programmable transmission device, which achieves pro⁃ grammable tuning of optical network elements [9]. Therefore, the SDON is more suitable for multi⁃layer/domain and multi⁃ constraint optical networks, and it can effectively improve oper⁃ ational efficiency and reduce cost. The article introduces the SDON innovation environment from the perspectives of architecture, protocol extension, exper⁃ imental platform and typical applications. Section 2 describes the hierarchical control architecture and the process of cross⁃ Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS50 This work was supported by ZTE Industry⁃Academia⁃Research Cooperation Funds under Grant No. Surrey⁃Ref⁃9953.
domain connection provisioning in detail. Section 3 depicts the workflow of connection provisioning in multi ⁃ domain optical networks. Section 4 shows the necessary extension work of OpenFlow 1.3 protocol for optical networks. The experimental environment and typical applications of SDON are respectively discussed in section 5 and section 6. Section 7 is performance evaluation of the SDON testbed and the last section summariz⁃ es the paper. 2 Hierarchical Architecture for SDON As shown in Fig. 1, considering the cross⁃layer distribution of multi⁃domain optical network resources, this paper proposes an OpenFlow enabled hierarchical control architecture in or⁃ der to solve the problem of programmable control in optical net⁃ work. With the advantage of software⁃defined networking, the architecture uniformly abstracts optical transmission network resources and content resources, and provides them with the multi ⁃ domain controller through the northbound interface. In this way, the uniform control of cross⁃layer resources is real⁃ ized. The hierarchical architecture consists of three layers: the physical layer, control layer and application layer. 1) The physical layer mainly includes data⁃center and inter⁃ data⁃center optical transmission networks. OpenFlow⁃enabled IP Routers (OF ⁃ Router) and optical transmission equipment with OpenFlow agents (OF⁃ROADM) are deployed in the net⁃ work. 2) The application layer mainly includes various applica⁃ tions such as dynamic migration of virtual machines, virtual network provisioning, and spectrum defragmentation. It is con⁃ nected with the control layer through the Restful API interface. All the service requests are triggered from this layer. 3) The control layer is mainly composed of optical control⁃ lers and multi⁃domain controller. In the optical controller, the protocol analysis module ana⁃ lyzes the underlying optical transmission equipment via the OpenFlow protocol extended for optical transmission devices. It collects the status of OF⁃ROADM in the optical network and abstracts the network topology information. Then the abstract⁃ ed topology information is sent to the network abstraction mod⁃ ule and stored in the optical database (ON ⁃ TED). The OTN manager manage the optical transmission equipment, such as lightpath setup and deletion, and resources allocation. The op⁃ tical network controller packages the network status and topolo⁃ gy information via the protocol encapsulation module and makes a notification to the multi⁃domain controller. The multi⁃domain controller integrates the network informa⁃ tion collected by the optical controller through the southbound Control Virtual Network Interface (CVNI) interface and moni⁃ tors the network status. With the northbound Restful ⁃ API, it parses the application requests sent by the application layer. It consists of nine function modules and one resource integration module. Resource integration is completed by the heteroge⁃ neous network database (Het⁃TED). The application database (App⁃TED) and ON⁃TED in network are set into the same data⁃ base, with the purpose of supplying the network resources in⁃ formation for the corresponding module in the network. The nine function modules are respectively described as follows. 1) Application monitor: It monitors the computing resources in the network and reports the information of computing re⁃ sources to service selection engine. 2) Service selection engine (SSE): According to the status of application resources and network resources requests, it se⁃ lects the most appropriate application resources to meet the vir⁃ tual requests. 3) Application resource manager: It manages the application resources in the network, and keeps real⁃time synchronous up⁃ date with the resources information in the application resourc⁃ es database (App⁃TED). 4) Request resolver: It parses the requests sent by the appli⁃ cation layer and forward them to the corresponding module for processing. 5) Virtual network manager: It manages the virtual network requests sent by the application layer according to the status of application resources and network resources, and selects the API: Application Program Interface APP⁃TED: application traffic engineering database AR: application resources CVNI: Control Virtual Network Interface ODL: OpenDaylight OF: OpenFlow OFP: OpenFlow protocol ON⁃TED: optical traffic engineering database OTN: optical transport network PCE: path computation element ROADM: Reconfigurable Optical Add/Drop Multiplexer SSE: service selection engine TED: traffic engineering database VN: virtual network ▲Figure 1. Hierarchical network architecture. Application 1, Application 2, Application 3, Application 4App layer Multi⁃domain control layer Optical control layer Physical layer Multi⁃domain controller Restful API Rq resolver VN manager AR manager SSE Application monitor Wrapper ODL core Policy generator Policy analyzing engine App⁃TED Domian 1⁃TED Domian 2⁃TED PCE Multi⁃domain TED Req/Rep PCE ON⁃TED Network abstraction OFP analysis ODL core Resolver OTN manager CVNI Optical controller CVNI Resolver Optical controller PCE OTN manager ON⁃TED Network abstraction OFP analysis ODL core OpenFlow agent OpenFlow agent OpenFlow agent OpenFlow agent OF⁃routerOF⁃router OpenFlow+Optical extension OpenFlow+Optical extension Data center Data center OF⁃ROADM Optical networks OF⁃ROADM Optical networks Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 51
most appropriate network links to meet the virtual network re⁃ quests with the corresponding strategy. 6) Policy generator: According to the network requests from the application layer, it generates the corresponding strategy information of resource provisioning for the heterogeneous net⁃ work controller. 7) Policy analyzing engine: It parses the strategy generated by the strategy generator module, and sends it to the corre⁃ sponding module for network resource allocation. 8) PCE: As a core component of the multi⁃domain controller, PCE is used in response to the request of path computation. The calculation is based on the input path information, strate⁃ gy information and request information, etc. It may return two kinds of calculation results, the appropriate path information computed by multi⁃domain or the failure information. 9) Wrapper: It packages the resource allocation information with the CVNI protocol and sends it to the optical controller via the CVNI interface. 3 Workflow of Connection Provisioning Fig. 2 shows the workflow of providing an end⁃to⁃end con⁃ nection in the multi⁃domain optical networks. After a Transmis⁃ sion Control Protocol (TCP) connection is established, the multi ⁃ domain controller completes the handshake with the optical controller by using OpenFlow messages, then periodically sends packages to keep the connection alive. The multi ⁃ do⁃ main controller requests the abstract topology as well as the de⁃ tailed port information. After receiving a request for connec⁃ tion setup the optical controller completes the path computa⁃ tion and resource allocation in the local domain via the domain⁃ specified protocol. Once the process is finished, the optical controller sends a“success”reply to the multi⁃domain control⁃ ler. When the multi ⁃ domain controller collects all the“suc⁃ cess”messages from optical domains, a“success”notification will be sent to the application layer. At this point, a connection or lightpath is considered to be established successfully. 4 Protocol Extensions for SDON Based on OpenFlow 1.3 protocol, CVNI is an interface proto⁃ col between the multi⁃domain controller and optical layer con⁃ troller. Several OpenFlow messages in CVNI have been extend⁃ ed to satisfy the requirements of optical networks. The multi⁃ domain controller sends a GET_CONFIG_REQUEST message to the optical controller to get the location of network nodes and the optical controller replies a GET_CONFIG_REPLY message. The MULTIPART_REQUET messages is used by the multi⁃domain controller to obtain topology resources including ports and links information. The MULTIPART_REPLY mes⁃ sage carries topology information from the optical controller to the multi⁃domain controller. The multi⁃domain controller em⁃ ploys FLOW_MOD messages to complete connection setup and deletion. The match field and action field in an extended FLOW_MOD message respectively represent the input optical port and output optical port. Note that the multi⁃domain con⁃ troller sends a BARRIER_REQUEST message to the single⁃do⁃ main controller in order to verify whether the optical cross con⁃ nection is deployed successfully. The single⁃domain controller then sends a BARRIER_REPLY message to notify the multi⁃ domain controller that the connection is created or deleted suc⁃ cessfully. Due to space limitation, only FLOW_MOD message extension is illustrated in Fig. 3. 5 Experimental Platform for SDON As shown in Fig. 4, an all⁃optical network innovation (AO⁃ NI) experimental platform for SDON is distributed in three ge⁃ ography locations connected by optical fiber links. Two of them are located in Room 342 and Room 423 in the Sci⁃ ence&Research building of Beijing University of Posts and Telecommunications (BUPT ), and the third location is at 21Vi⁃ anet Company in Jiuxianqiao, Beijing. Two data centers are re⁃ spectively deployed in Room 342 and 21Vianet Company, and Room 423 serves as the access network for users, which com⁃ poses a typical network environment with the application of da⁃ ta center. The AONI platform supports three typical network scenarios, i.e., the inter⁃data center network, user access to da⁃ ta center network and intra ⁃ data center network. The AONI platform focuses on how to embody the advantages of optical switching network in these three scenarios. The platform sup⁃ ports both optical burst switching and optical circuit switching, and supports both flexible grid high⁃speed optical transmission and fixed grid transmission. Therefore, the AONI platform not only provides efficient transmission and switching in the future but also remains compatible with traditional networks. The high capacity optical burst switching (OBS) is mainly used for TCP: Transmission Control Protocol ▲Figure 2. Workflow of connection provisioning in multi⁃domain optical networks. Optical controller #1 Multi⁃domain controller Optical controller #N Application TCP three⁃way handshake TCP three⁃way handshake OpenFlow handshake & keep alive OpenFlow handshake & keep alive Topology resource reply Topology resource request Topology resource request Topology resource reply Connect creation request Connection request in local domain Connection request in local domain Connection reply in local domain Connection reply in local domain Connection creation reply Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS52
the adaption to the high burstiness character⁃ istic of intra⁃data center services. The flexible grid high⁃speed optical transmission andopti⁃ cal circuit switching are mainly applied to the inter⁃data center to realize large⁃grained vari⁃ able bandwidth switching. The all⁃optical ac⁃ cess and convergence layer uses fixed grid transmission and switching to achieve flexible access of broadband services. Thus the archi⁃ tecture of AONI includes intra⁃data center all⁃ optical interconnection, all⁃optical access lay⁃ er, all⁃optical convergence layer and all⁃opti⁃ cal core layer. Such a platform can highly sim⁃ ulate real scenarios of all ⁃ optical switching wide area network (WAN) in the future. 6 Typical Applications of SDON The SDON is a promising solution to high intelligence of next generation optical net⁃ work and has broad application prospects. The typical applications include bandwidth on demand (BoD) provisioning, virtual ma⁃ chine (VM) online migration, spectrum defrag⁃ mentation, and virtual optical networks (VON) provisioning. The homepage of AONI applications is shown in Fig. 5. BoD applications and VM migration are im⁃ plemented based on the physical topology shown in Fig. 4. For lack of flexible⁃grid opti⁃ cal devices, a multi ⁃ domain logical topology (Fig. 6) is designed for VON provisioning and spectrum defragmentation. Both the physical topology and the logical topology are under control of the SDN controller. Each domain in⁃ ◀Figure 4. AONI: all optical network innovation environment. BUPT: Beijing University of Posts and Telecommunications BV⁃OXC: bandwidth⁃variable optical cross connect DC: data center OXC: optical cross connect S⁃NE: static⁃network element S⁃ROADM: static⁃reconfigurable optical add/drop multiplexer ▲Figure 3. FLOW_MOD message extension of CVNI protocol. China Unicom Cernet OpenFlow switch Room 423, BUPT Access layer S⁃NE S⁃ROADM BV⁃OXC Port A Port B Port C Cernet OpenFlow switchRoom 342, BUPT DC, BUPT Energy⁃efficient OXC Energy⁃efficient OXC Optical interconnection in DC Aggregation layer Core layer Chaoyang district, Beijing Jiuxianqiao DC 10GE optical interface 40G optical interface 1GE optical interface Flexible all⁃optical networks Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 53
cludes eight standalone OpenFlow⁃Agents (OF⁃AGs). Running on high ⁃ performance Linux servers, each OF ⁃ AG is pro⁃ grammed based on Open⁃vSwitch. 6.1 BoD Applications BoD applications help users have a global understanding of underlying optical networks and accomplish a series of opera⁃ tions in optical networks, including connection setup, connec⁃ tion deletion, connection query, connection modification and so on. A lightpath connection is built and on ⁃ demand band⁃ width is allocated according to users requirements. Besides the instant operation, users are able to make an appointment to car⁃ ry out above operations by setting starting and ending time. Fig. 7 shows a connection named“service 1”is created from node 20.20.20.14 to node 20.20.20.21 with required band⁃ width. The detailed information about this connection is listed in the lower part of Fig. 7b, including routing, current status, creation delay and so on. 6.2 VM Online Migration VM migration plays an important role in data backup and load balance of data centers. A VM migration application en⁃ ables online migration of virtual machines among different da⁃ ta centers. With transmission advantages of optical networks, it just takes a short time to complete the migration process. In ad⁃ dition, the online migration pattern has no impact on users’ac⁃ cess to resources in the migrating virtual machine. In Fig. 8, a VM, 863VM, is migrated from server 10.108.50.40 to server 10.108.51.124 and the migration path is 20.20.20.14 ⁃ 20.20.20.15 ⁃ 20.20.20.12. Meanwhile, users can query re⁃ source utilization information of the selected servers, such as CPU and memory status. 6.3 VON application Optical network virtualization technologies support the dy⁃ namic provisioning of VONs in the same network infrastruc⁃ ture and achieve high⁃efficiency utilization of network resourc⁃ es. Because of its centralized control manner, software⁃defined networking (SDN) is regarded as a promising technology for re⁃ alizing VON provisioning. In the AONI testbed, network opera⁃ tors can provide virtual optical networks for different custom⁃ ers. The topology of VON can either be pre⁃configured by oper⁃ ▲Figure 5. Homepage of AONI applications. ▲Figure 6. Multi⁃domain logical topology. ▲Figure 7. Web view of BoD application: (a) before connection setup; (b) after connection setup. Domain 1 Domain 3 Domain 2 (a) (b) Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS54
ators or be customized by users. In Fig. 9, a triangle VON to⁃ pology is successfully mapped to multi ⁃ domain networks. Meanwhile, 1+1 protection is available for services deployed in the VON. The green path in Fig. 9b stands for the working route of the service while the purple path represents the protec⁃ tion path. 6.4 Spectrum Defragmentation The frequent setup and release of lightpaths in a dynamic network scenario will fragment the optical spectrum into non⁃ aligned, isolated and small ⁃ sized spectrum segments. Spec⁃ trum fragments result in low spectrum utilization and high blocking probability since these fragments could hardly be oc⁃ cupied for new incoming requests. With the application of spectrum defragmentation, users can have a good knowledge of spectrum utilization and trigger spectrum defragmentation if necessary to optimize spectrum resources. In Fig. 10, there are 50 connections or lightpaths deployed in the multi⁃domain net⁃ work shown in Fig. 6. It is obvious that the spectrum utilization is effectively improved with the implementation of spectrum de⁃ fragmentation 7 Performance Evaluation of SDON Testbed Batch testing has been conducted to evaluate the perfor⁃ mance of this SDON testbed. Ten thousands lightpath requests are generated following Poisson distribution, and their source⁃ destination pairs per execution are randomly chosen. The hold⁃ ing time of lightpath requests follows exponential distribution. To verify the scalability of the SDON testbed, we compare the blocking probabilities of different network sizes. As shown in Fig. 11, the number of network nodes ranges from 200 to ▲Figure 8. Web view of VM migration: (a) before the migration; (b) after the migration. ▲Figure 9. Web view of VON provisioning: (a) before the process; (b) after the process. (a) (b) (a) (b) Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 55
1000. For each network size, the blocking probability increas⁃ es with traffic load. With the same traffic load, 1000⁃nodes net⁃ work has the lowest blocking probability since it has the high⁃ est network capacity. In addition, the relationship between the controller output bandwidth and the lightpath setup delay is studied. The con⁃ troller output bandwidth can be adjusted by the VMware Ether⁃ net bandwidth modulator. As shown in Fig. 12, the output bandwidth of controller is set to five different values, including 400 kbps, 500 kbps, 700 kbps, and 1 Mbps. The average delay of lightpath setup is calculated for each case. We can see that the output bandwidth of controller has great influence on light⁃ path setup delay. With the growth of output bandwidth, the av⁃ erage setup delay decreases significantly from 300 ms to 50 ms. 8 Conclusions With the advantage of programmable network elements, the SDON realizes service customization, adaptive modulation for⁃ mat, flexible bandwidth allocation and dynamic provisioning of virtual network resources with centralized control manner. This paper introduces SDON and its innovation environment— AONI in terms of network architecture, protocol extension solu⁃ tion, experiment platform, typical applications and perfor⁃ mance evaluation. The SDON represents the development di⁃ rection of optical networks and has broad application prospects in the future. ▲Figure 10. Web view of spectrum defragmentation: (a) before defragmentation; (b) after DFefragmentation. ▲Figure 11. Blocking probabilities of different network sizes. ▲Figure 12. Lightpath setup delay of different output bandwidth. References [1] Requirements for Generalized Multi⁃Protocol Label Switching (GMPLS) Routing for the Automatically Switched Optical Network (ASON), IETF RFC4258, Nov. 2005. (a) (b) 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 Blockingprobability 12001000800600200 Traffic load (Erlangs) 200 Nodes 400 Nodes 600 Nodes 800 Nodes 1000 Nodes 400 300 Setupdelay(ms) Traffic load (Erlangs) 250 200 150 100 50 0 1000800600400200 infinite 1 Mbps 700 kbps 500 kbps 400 kbps Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS56 0
[2] Requirements for Generalized MPLS (GMPLS) Signaling Usage and Extenstions for Automatically Switched Optical Network (ASON), IETF RFC4139, Jul. 2005. [3] Requirements for GMPLS⁃Based Multi⁃Region and Multi⁃Layer Networks (MRN/ MLN), IETF RFC5212, Jul. 2008. [4] Y. Ji, D. Ren, H. Li, X. Liu, and Z. Wang,“Analysis and experimentation of key technologies in service⁃oriented optical internet,”Science China Information Sci⁃ ences, vol. 54 no. 2, pp. 215-226, Feb. 2011. doi: 10.1007/s11432⁃010⁃4168⁃5. [5] A Path Computation Element (PCE)⁃Based Architecture, IETF RFC4655, Aug. 2006. [6] Path Computation Element (PCE) Communication Protocol Generic Require⁃ ments, IETF RFC4657, Sept. 2006. [7] Requirements for Path Computation Element (PCE) Discovery, IETF RFC4674, Oct. 2006. [8] Path Computation Element Communication Protocol (PCECP) Specific Require⁃ ments for Inter ⁃ Area MPLS and GMPLS Traffic Engineering, IETF RFC4927, Jun. 2007. [9] J. Zhang, H. Yang, Y. Zhao, et al.,“Experimental demonstration of elastic opti⁃ cal networks based on enhanced software defined networking (eSDN) for data center application,”Optics Express, vol. 21, no. 22, pp. 26990- 27002, Nov. 2013. doi:10.1364/OE.21.026990. Manuscript received: 2016⁃03⁃31 LI Yajie (yajieli@bupt.edu.cn) is a PhD candidate in State Key Laboratory of Infor⁃ mation Photonics and Optical Communication, Beijing University of Posts and Tele⁃ communications (BUPT), China. His research interest is software defined optical networks. ZHAO Yongli (yonglizhao@bupt.edu.cn) received his PhD degree from BUPT. He is an associate professor in State Key Laboratory of Information Photonics and Opti⁃ cal Communication, BUPT. His research interest is optical transport networks. ZHANG Jie (lgr24@bupt.edu.cn) received his PhD degree from BUPT. He is a pro⁃ fessor in State Key Laboratory of Information Photonics and Optical Communica⁃ tion, BUPT. His research interest is optical transport networks. WANG Dajiang (wang.dajiang@zte.com.cn) works in wireline product operation of BN product team, ZTE Corporation. His research interest is optical transport net⁃ works. WANG Jiayu (wang.jiayu1@zte.com.cn) received his master degree from BUPT. He is a SDON R&D representative from BN product team, ZTE Corporation. His re⁃ search interest is optical transport networks. BiographiesBiographies Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 57 New Members of ZTE Communications Editorial Board Dr. CHEN Yan is a professor in the Department of Electrical Engineering and Computer Science at Northwestern University, USA. He is also an adjunct professor in the College of Computer Science at Zhejiang University, China. He got his PhD in Computer Science at University of California at Berkeley, USA in 2003. His research interests include network security, measurement and diagnosis for large scale networks and distributed systems. He won the Department of Energy (DoE) Early CAREER award in 2005, the Department of Defense (DoD) Young Investigator Award in 2007, and the Best Paper nomination in ACM SIGCOMM 2010. Based on the Google Scholar, his papers have been cited for over 9000 times and his h⁃index is 40. Dr. SONG Wenzhan is the Georgia Power Mickey A. Brown Professor of Engineering in the University of Georgia, USA. Dr. Song is a distinguished scientist and educator on cyber⁃physical systems informatics and security in energy, environment and health applications, where decentralized sensing, computing, communication and security play a critical role and need a transformative study. He has an outstanding record of leading large multidisciplinary research projects on those issues with multi⁃million grant support from NSF, NASA, USGS, and industry, and his research was featured in MIT Technology Review, Network World, Scientific America, New Scientist, National Geographic, etc. Dr. Song is a recipient of NSF CAREER Award (2010), Outstanding Research Contribution Award (2012) at GSU, Chancellor Research Excellence Award (2010) at WSU. He was also a recipient of 2004 National Outstanding Oversea Student Scholarship by China (only 40 in USA) during PhD study. Dr. Song also has a outstanding publication record and serves many premium conferences and journals as editor, chair or TPC member. He is also an inaugural member of OpenFog consortium involving industry and academic leaders.
Depth EnhancementDepth Enhancement Methods for CentralizedMethods for Centralized Texture⁃Depth PackingTexture⁃Depth Packing FormatsFormats YANG Jar􀆼Ferr, WANG Hung􀆼Ming, and LIAO Wei􀆼Chen (Department of Electrical Engineering, Institute of Computer and Communication Engineering, National Cheng Kung University, 1 University Road, Taiwan 701, China) To deliver three ⁃ dimension (3D) videos through the current two⁃dimension (2D) broadcasting systems, the frame⁃compati⁃ ble packing formats properly including one texture frame and one depth map in various down ⁃ sampling ratios have been proposed to achieve the simplest and most effective solution. To enhance the compatible centralized texture⁃depth packing (CTDP) formats, in this paper, we further introduce two depth enhancement algorithms to further improve the quality of CT⁃ DP formats for delivering 3D video services. To compensate the loss of color YCbCr 444 to 420 conversion of colored ⁃ depth, two efficient depth reconstruction processes based on texture and depth consistency are proposed. Experimental re⁃ sults show that the proposed enhanced CTDP depacking pro⁃ cess outperforms the 2DDP format and the original CTDP de⁃ packing procedure in synthesizing virtual views. With the help of the proposed efficient depth reconstruction processes, more correct reconstructed depth maps and better synthesized quality can be achieved. Before the available 3D broadcasting systems, which adopt truly depth and texture dependent cod⁃ ing procedure, we believe that the proposed CTDP formats with depth enhancement could help to deliver 3D videos in the current 2D broadcasting systems simply and efficiently. 3D videos; frame⁃compatible; 2D⁃plus⁃depth; CTDP Abstract Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 008 http://www.cnki.net/kcms/detail/34.1294.TN.20160802.1732.004.html, published online August 2, 2016 O 1 Introduction ver past decades, more and more three⁃dimension⁃ al (3D) videos have been produced in the formats of stereo or multiple views with their correspond⁃ ing depth maps. People desire to have more truth⁃ ful and exciting experience through the true 3D visualizations. In order to fit the traditional two⁃dimensional (2D) television (TV) programs, we need to modify the 3D videos to accommo⁃ date the certain constraints. Frame⁃packing is one of possible solutions to introduce 3D services in the current cable and ter⁃ restrial 2D TV systems. There are several well⁃known formats for packing the stereo views into 2D frame such as side⁃by⁃ side (SbS), top ⁃ and ⁃ bottom (TaB), and checkerboard frame ⁃ compatible formats [1]- [4]. However, there exist two major problems, which slow down the development of the 3D TV ser⁃ vices, in the existing frame⁃packing methods. The frame⁃com⁃ patible packing 3D videos of the stereo views mean that two texture images are gathered in one frame, which may make se⁃ rious annoying effects on traditional 2D displays. Besides, ste⁃ reo packing formats cannot support multi⁃view naked⁃eye 3D displays unless the stereo videos are further processed by real⁃ time stereo matching methods [5], [6] and depth image⁃based rendering (DIBR) algorithms [7], [8]. To support multiview 3D displays, the 2D⁃plus⁃depth packing (2DDP) frame⁃compatible format, which arranges the texture in the left and the depth in the right, is suggested [9]. Once the color texture and depth ar⁃ ranged in the SbS fashion, the 2DDP format will bring even worse annoying visualization in 2D displays than the stereo packing formats. Recently, MPEG JCT⁃3V team proposed the latest coding standard for 3D video with depth [9]. However, it still needs some time to be deployed in current digital video broadcasting systems, which are with 2D and 3D capabilities. To deal with the above problems, a novel frame—compati⁃ ble centralized texture⁃depth packing (CTDP) formats for deliv⁃ ering 3D video services is proposed [10]. With AVS2 and HEVC video coders, the proposed CTDP formats [10] show bet⁃ ter objective and subjective visual quality in 2D and 3D dis⁃ plays than the 2DDP format. In the CTDP format, the sub⁃pixel is utilized to store the depth information, while the texture in⁃ formation is arranged in the center of the frame to raise the 2D⁃ compatible visual quality. However, the rearrangement will de⁃ grade the quality of the reconstructed depth map, especially when the video format with YCbCr space is 420 format with 4 Y components, one Cb component and one Cr component for each 4 color pixels. To further increase the visual quality, an efficient depth reconstruction process is also proposed in this paper. The frame structure of the CTDP method in cooperation with the current broadcasting system is shown in Fig. 1. With⁃ out any extra hardware, the 2D TV displays can also exhibit an acceptable 2D visual quality. For glasses or naked⁃eye 3D dis⁃ plays, we only need a simple CTDP depacking circuit followed by DIBR kernel to synthesize stereo or multiple views if the view⁃related sub⁃pixel formation of a naked⁃eye 3D display is given. The rest of the paper is organized as follows. The CTDP for⁃ mats are overviewed in Section 2. The proposed depth recon⁃ struction process is described in Section 3. Experimental re⁃ sults to demonstrate the effectiveness of the proposed system are shown in Section 4. Finally, we conclude this paper in Sec⁃ Research Paper October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS58
tion 5. 2 Centralized Texture􀆼Depth Packing Formats To achieve system compatibility, the basic concept of the CTDP method [10] is similar to frame compatible concept to pack texture and depth information together while keeping the same resolution as 2D videos. To solve the 2D visualization is⁃ sue, we can arrange the texture in the center and the depth in two sides of the packed frame. 2.1 Colored􀆼Depth Frame The depth frame is only a gray image with Y components. To pack the depth frame, the colored⁃depth frame is suggested to represent it [10]. Thus, the colored⁃depth frame can be treated as the normal color texture frame, which can be directly encod⁃ ed by any 2D video encoders with three times efficiency. As shown in Fig. 2, three depth horizontal lines are treated as hor⁃ izontal R, G, and B subpixel lines in the RGB colored⁃depth frame. Since the nearby depth values are very close, the RGB colored⁃depth frame will exhibit nearly gray visual sensation. After color subpixels packing in the vertical direction, the ver⁃ tical resolution of RGB colored⁃depth frame becomes one third of the original resolution. In Fig. 2, for example, the nine depth lines have been packed into three RGB colored⁃ depth lines. For the most video coders, the coding and decoding processes are conducted in YCbCr color space. Therefore, we apply the RGB to YCbCr color space conversion as é ë êê ù û úú Y Cb Cr = é ë êê ù û úú 0.2568 0.5041 0.0979 -0.1482 -0.2910 0.4392 0.4392 -0.3678 -0.0714 é ë êê ù û úú R G B + é ë êê ù û úú 16 128 128 (1) to transfer it to the YCbCr colored⁃depth frame [11]. It is noted that the sub⁃pixels in RGB space are with full resolution of (4, 4, 4). If the YCbCr space is with (4, 4, 4) format, the color space transformation will not change the depth results with about +/- 0.5 error due to the round⁃off errors in color space conversions. However, for the most video coders, the sub⁃pix⁃ els in YCbCr space could be in (4, 2, 0) or (4, 2, 2) format, where Cb and Cr components will be further downsampled. Even without coding errors, the YCbCr colored ⁃ depth frame might have slightly translation errors. 2.2 Centralized Texture􀆼Depth Packing Without loss of generality for frame⁃compatible packing, we assume that the vertical CTDP packing formats are desired. Then, we need to reduce the vertical resolutions of texture and depth separately such that the total packed resolution will re⁃ mind the same, where the original horizontal resolution is H. If the reduction factors for texture and depth resolutions are a and b, we should choose reduction factors to satisfy α +(1/3)β = 1 to achieve the frame compatible requirement [10]. For example, the reduction factors (a = 3/4, b = 3/4) , (a = 5/6, b = 1/2), (a = 7/8, b = 3/8), (a = 11/12, b = 1/4), and (a = 15/16, b = 3/16) will satisfy the above frame compatible re⁃ quirement. Fig. 3 shows the flowchart of the computation of generating the texure ⁃ 5/6 CTDP format. First, we downscale the vertical resolution of texture and depth frames into five ⁃ sixths and one⁃second of the original resolution, respectively. By using the colored⁃depth concept, the resized depth frame with 1/2H can be further represented into RGB subpixels as suggested in Section 2.1 to reduce the vertical size to 1/6H. Then, we can split the depth frame evenly into two separated parts with the size of 1/12H. To make better coding efficiency and better 2D visualization, these two split colored ⁃ depth frames should be flipped vertically. The flipped depth frames will have better alignments to the texture frame and better visu⁃ alization for 2D displays with visual shadow sensation. Finally, we obtain the texture⁃5/6 CTDP frame by combining the first ▲Figure 1. The broadcasting architecture by using the proposed enhanced CTDP format. DIBR: depth image⁃based rendering ▲Figure 2. Rearrangement of the depth frame into RGB colored depth frame in vertical direction. Color and depth packing Video encoder Video decoder Multiview DIBR Color and depth depacking with depth enhancement 2DTV 3DTV No supporting hardware needed TV broadcasting network Stereo multiview 3D display formation Original depth frame (9 gray depth lines) 1 2 3 4 5 6 7 8 9 1R 1G 1B Colored depth frame (3 color depth lines) 2R 2G 2B 3R 3G 3B 3 depth pixels to 1 color depth pixel (RGB subpixels arrangement) Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 59
flipped depth part (1/12H), the resized texture frame (5/6H), and the other flipped depth part (1/12H) from top to bottom se⁃ quentially. The ratio of downscaling can also be changed to generate the other CTDP formats [12]-[15]. For example, the reduction ratio of the texture frame could be 7/8 or 15/16. For texture⁃7/ 8 and texture⁃15/16 reduction ratios, the vertical resolutions of depth frames will be respectively downscaled to 3/8 and 3/16 to satisfy (2). Except the resizing factor, the packing proce⁃ dures for texture⁃7/8 and texture⁃15/16 are similar to that of texture⁃5/6. If we want to attain horizontal CTDP formats, all the resizing of texture and depth frame, the color⁃packed depth frame, slipping, and flipping procedures should be per⁃ formed in the horizontal direction. The packed frame can be obtained by com⁃ bining the first flipped depth part, the resized texture frame, and the other flipped depth part from left to right se⁃ quentially. The outlooks of the original texture, depth, and the CTDP frames with different ratios and different orien⁃ tations are shown in Fig. 4. It is noted that in the proposed CTDP format, the width/height of the flipped depth part will be always in the horizontal/vertical CTDP format, which helps avoid the compression artifact in texture and depth boundary. Please refer to [13] for more details of the arrangement. 2.3 Depacking CTDP Formats With respect to the packing proce⁃ dure in Fig. 3, the flow diagram for de⁃ packing the texture ⁃ 5/6 CTDP format is shown in Fig. 5. Once we receive the CTDP format, we should first split the packed frame into three parts: the top flipped depth part, the central texture, and the bot⁃ tom flipped depth part. For two flipped depth parts, we perform another vertical flipping and combined them into the whole texture⁃packed depth frame. The YCrCb colored⁃depth frame might need to upsample Cr and Cb components back to (4, 4, 4) format first. Then, we can convert it to (4, 4, 4) RGB colored ⁃ depth frame by é ë êê ù û úú R G B = é ë êê ù û úú 1.1644 -0.0001 1.5960 1.1644 -0.3917 -0.8130 1.1644 2.0173 -0.0001 æ è ç ç ö ø ÷ ÷ é ë êê ù û úú Y Cb Cr - é ë êê ù û úú 16 128 128 . (2) After the color space conversion, The RGB colored ⁃depth frame (1/6H) can be finally recovered to the re⁃ sized depth frame (1/2H). After 6/5 upscaling texture and 2/1 depth frames in the vertical direction, we finally depack the origi⁃ nal texture and depth frames. Of course, a possible DIBR method should be used to generate all the necessary views. As for the other texture reduction ratios such as 7/8 and 15/16, all the procedures will be the same except the resizing factors of depth will be 3/8 and 3/16, respectively. 3 Depth Enhancement Algorithms From the previous section, it is known that when the YCbCr space is (4, 2, 0) or (4, 2, 2) format, the YCbCr colored⁃depth frame will induce translation errors along the depth edges. To ▲Figure 3. The computation of the proposed frame compatible texture⁃5/6 CTDP format. (a) ▲Figure 4. Schematics of original (a) texture, (b) depth; (c) vertical texture⁃5/6 CTDP; (d) vertical texture⁃7/8 CTDP; (e) vertical texture⁃15/16 CTDP; (f) horizontal texture⁃5/6; (g) CTDP, horizontal texture⁃7/8 CTDP; (h) horizontal texture⁃15/16 CTDP; and (i) 2DDP frame compatible formats. 5/6H resized texture 1H 1H 1/2H 1/6H 1/12H 1/12H 1/12H 5/6H 1/12H 1H Downscaling in vertical direction Downscaling in vertical direction Pixel rearrangement to RGB subpixels channel YCbCr color format conversion Vertical splitting & flipping Combination Y′ Cb′ = Cr′ 0.2568 0.5041 0.0979 -0.1482 -0.2910 0.4392 0.4392 -0.3678 -0.0714 Y Cb Cr 16 128 128 + (b) (c) (d) (e) (f) (g) (h) (i) Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS60
further reduce the depth edge errors, in this paper, we propose two efficient depth enhancement processes. The enhancement processes can be incorporated with the original depacking pro⁃ cess as shown in Fig. 6. The enhancement processes include YCbCr calibration, texture⁃similarity⁃based depth up⁃sampling and pattern⁃based down⁃sampling. Details of the enhancement algorithms are addressed in the following subsections. 3.1 YCbCr Calibration When the YCbCr color space is (4, 4, 4), the color space transformation between RGB color space and YCbCr color space will only contain round⁃off errors in color space conver⁃ sions. However, for the most video coders, the sub⁃ pixels in the YCbCr color space might be (4, 2, 0) or (4, 2, 2) formats, where Cb and Cr components will be further down⁃sampled in order to save the bandwidth in broadcasting systems. At the depack⁃ ing side, we need to calibrate the translation errors between YCbCr (4, 4, 4) and YCbCr (4, 2, 0) and (4, 2, 2). For simplicity, we will illustrate our pro⁃ posed system in YCbCr (4, 2, 0), however, the simi⁃ lar manner can still be applied for YCbCr (4, 2, 2). Before we start to calibrate the YCbCr data, we first define some anchor pixels, which are shown in Fig. 7. The anchor pixels denote the pixels which have the correct Cb and Cr subpixel values. The diagram of missing components in YCbCr (4, 2, 0) for all surrounding pixels is shown in Fig. 8. Each color means a set which Cb and Cr subpixel components are down ⁃ sampled. The black area means the missing Cb and Cr subpixels and they can be given by: Cbcal(a,b)= argCbC min |YC - Y(a,b)| , (3) and Crcal(a,b)= argCrC min |YC - Y(a,b)| , (4) where YC is a vector of the neighbor anchor pixels of the pix⁃ els Y(a, b). 3.2 Texture􀆼Similarity􀆼Based Depth Up􀆼Sampling In order to preserve the continuity of the edge, the direction⁃ al vectors are utilized to calculate the edge direction in the low⁃ resolution (LR) depth and the corresponding high resolution (HR) texture image. The directional vec⁃ tors of LR depth image and HR texture image can be formed as:   VdL =∑Ω exp(- DE(xL,yL)- DΩ σV ) × uΩ , (5) and  Vc =∑Ω exp(- Y(x,y)- YΩ σV ) × uΩ , (6) where   VdL and  Vc denote the directional vectors of the pixels in LR depth image and HR texture image, respectively, σV represents the standard deviation ▲Figure 5. The computation of the proposed texture⁃5/6 CTDP depacking procedure. ▲Figure 6. Flowchart of the depth⁃enhanced CTDP de⁃packing system. ▲Figure 7. Anchor pixels in YCbCr (4, 2, 0). 1H 1/12H 1/12H View and depth splitting Vertical flipping & combination Inverse YCbCr color format conversion Get depth pixel from RGB subpixels channel Upscaling in vertical direction R G = B 1.1644 -0.0001 1.5960 1.1644 -0.3917 -0.8130 1.1644 2.0173 -0.0001 Y Cb Cr 16 128 128 - 1/12H 5/6H 1/12H 1H 1H 1/6H 1/2H Upscaling in vertical direction View and depth splitting Horizontal flipping & combination View and depth splitting Horizontal flipping & combination YCbCr calibration Inverse YCbCr color format conversion Get depth pixel from RGB subpixels channel Upscaling in horizontal direction (bicubic) Depth refinement Upscaling in horizontal direction H W⁃WRD*2 H W⁃WRD*2WRD WRD H W W H Y Cb Cr Y YY Y :Yanc Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 61
of the directional vector function, Ω denotes the 8 neighbor pixels of the target pixel (Fig. 9), DE represents the combined depth, which is obtained from previous step, Y is the bright⁃ ness of the texture image, and uΩ is the unit vector corre⁃ sponding to the neighbor pixels Ω in 8 directions. Before up⁃sampling the depth image, the directional vectors are first transformed from Cartesian coordinate system to Spherical coordinate system. The transform function is given by: r = x∂ 2 + y∂ 2 , (7) and θ = arctan( y∂ x∂ ) , (8) where x∂ and y∂ denote the coordinate of reconstructed depth at high resolution. For example, at vertical texture⁃11/12 CT⁃ DP, x∂ = 4x and y∂ = y . However, the resolution of directional vectors in depth image is smaller than the resolution of direc⁃ tional vectors in texture image. The bilinear interpolation [16] is utilized to scale up the depth directional vector to the resolu⁃ tion of the texture image. After that, The interpolated depth im⁃ age is formed as: where Tup denotes the normalized factor, p is the target pixel which needs to be scale up, q is the neighbor pixels of the tar⁃ get pixel, and Vd(θ) is the value of θ in the scaled VdL(θ) . ψ denotes the Gaussian weight function and can be giv⁃ en as: ψ(n)= exp(- n2 σψ ) . (10) The basic concept of the depth interpolation is to compare the directional vectors of the depth image and the texture image. The weighted summation of the LR depth is utilized to interpolate the HR depth if the directional vectors of the depth image and the texture image are similar. Otherwise, the pixels in HR depth are regarded as holes, which are filled in the step of hole⁃filling. The function of hole⁃filling is given as: Dhole - filling(x,y)= ì í î argDup ξ(min(ΔPc(θ))), if (x,y)∍ holes Dup(x,y), else , (11) where ΔPc(θ) denotes the difference of the degree between Pcθ and 8 neighbor pixels. ξ represents the selection fun⁃ ction of the hole⁃filling and it can be formed as: ξ(m)={Y(m), if||Y - Y(m)|| < THY ξ(m)+ 1, else , (12) where Y denotes the brightness of the target pixel, Y(m) de⁃ notes the brightness of the neighbor pixels in m direction, THY is the threshold to control the selection range, and ξ(m)+ 1 represents the next pixel in m direction. 3.3 Pattern􀆼Based Down􀆼Sampling In order to contain texture image and depth image in one sin⁃ gle frame, both depth image and texture image need to be down ⁃sampled. For the depth image, the bilinear and bi⁃cubic con⁃ volution methods are utilized to down⁃sample the depth image. However, the weighted summation strategy in bilinear and bi⁃ cubic convolution leads to the blur of the down⁃sampled data. Hence, we propose two sampling patterns to down⁃sample the depth image without fusing the data. There are the direct line pattern and slant line pattern. 1) Direct line pattern The sampling strategy of direct line pattern is to grab pixels in the straight line direction. According to the characteristic of the CTDP format, the reduction of the resolution is only in ei⁃ ther horizontal or vertical direction. The function of direct line pattern is given as: Ddown(x,y)= Dorigin(∂hor × x -[∂hor /2],∂ver × y -[∂ver /2]) , (13) where ∂hor and ∂ver are the factors of down⁃sampling ratios in horizontal direction and vertical direction, respectively. For CTDP format usage, either ∂hor or ∂ver is equal to 1, while the other one denotes the down⁃sampling ratio in packing proce⁃ dure. [x] is the floor function, which means the largest integer not greater than x. The direct line pattern in horizontal direc⁃ tion with 2, 4, 8 down⁃sampling ratio is shown in Fig. 10. 2) Slant line pattern ▲Figure 8. Missing Cb and Cr subpixels in YCbCr (4, 2, 0). ◀Figure 9. The diagram of the neighbor pixels, Ω . (9)Dup = ì í î ï ï 1 Tup ∑DE(q)× ψ(DE(p)- DE(q)), if||Vd(θ)- Vc(θ)|| < π 8 or Vc(r) < 1 hole, else , YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY Ω 135° Ω 90° Ω 45° Ω 0° Ω 180° Ω 225° Ω 275° Ω 315° Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS62
The sampling strategy of slant line is to grab pixels in 45 de⁃ gree direction. The function of direct line pattern is given as: Ddown(x,y)= Dorigin(∂hor × x -(∂hor - y),y) , (14) or Ddown(x,y)= Dorigin(x,∂ver × y -(∂ver - x)) . (15) Equ. (14) is utilized to down⁃sample the depth image in hori⁃ zontal direction, while the down⁃sampling of the vertical direc⁃ tion follows (15). The slant line sampling pattern is suitable for down⁃sampling the depth image both in vertical and horizontal direction, which is shown in Fig. 11 with 2, 4, 8 down⁃sam⁃ pling ratios. With the down⁃sampling by the direct line pattern, the up⁃ sampling function in de⁃packing procedure needs to be modi⁃ fied as: Because of the pattern⁃based sampling strategy, the pixels of the up⁃sampled depth are directly copied from the LR depth if there are located at position of the direct line pattern. 4 Experimental Results 4.1 Performance Evaluation of CTDP Format with Respect to 2DDP Format In order to verify the coding performances of the proposed CTDP formats with respect to the 2DDP format, we conducted a set of experiments to evaluate performances of packing meth⁃ ods in cooperation with a specific video coder (AVS2) in terms of the peak signal⁃to⁃noise ratio (PSNR), bitrate qualities of the depacked texture and depacked depth frames, and their synthe⁃ sized virtual views. In the experimental simulations, we use five MPEG 3D video sequences, which are Poznan Hall, Poznan Street, Kendo, Balloons, and Newspaper sequences as shown in Figs. 12a-12e, respectively. The AVS2 coding conditions are followed by the instruction suggested by the AVS workgroup while the QPs are set to 27, 32, 38, and 45 for Intra frames [17]. Under All Intra (ai), Low Delay P (ldp), Random Access (ra) test conditions, Tables 1 and 2 show the average BDPSNR and BDBR [18] performance for different kinds of CTDP formats with respect to the 2DDP format achieved by AVS2. For calculating the PSNR of the 2DDP format, we first separate the texture and depth frames from the 2DDP frame and upsample them to the original image size W × H. By using the recovered texture and depth frames from 2DDP frame and the original uncompressed texture and depth frames, the PSNR can therefore be calculated. Similarly, the PSNR of CTDP format is calculated by using the texture and depth frames recovered from CTDP frame and the original uncompressed texture and depth frames. From Tables 1 and 2, we can see that the proposed texture⁃5/6, 7/8, and 15/16 CTDP formats have much better PSNR and bitrate saving in texture when comparing with the 2DDP format, which means our CT⁃ DP format can achieve better visual quality in 2D displays when only texture frames are viewed. In addition, the depth quality for CTDP formats will become worse while the resizing factors getting bigger. Besides the comparisons of original tex⁃ ture and depth achieved by different packing formats, we also compare the quality of synthesized virtual view with respect to the 2DDP format. It is noted that the reference synthesized vir⁃ ▲Figure 10. The direct line pattern in horizontal direction of (a) down⁃sampling factor 2; (b) down⁃sampling factor 4; and (c) down⁃sampling factor 8. 10 11 (16)Dup = ì í î ïï ïï DE(p), if p ∈ sampled data 1 T ∑DE(q)× ψ(DE(p)- DE(q)), else if ||Vd(θ)- Vc(θ)|| < π 8 or Vc(r) < 1 hole, else . ▲Figure 11. The slant line pattern of (a) down⁃sampling factor 2; (b) down⁃sampling factor 4; (c) down⁃sampling factor 8. 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (a) (c) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (b) Missing pixel Sampled pixel 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (a) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (b) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (c) Missing pixel Sampled pixel Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 63
tual view for calculating the PSNR is also obtained by the origi⁃ nal uncompressed texture and depth frames. The DIBR setting for virtual view synthesis is shown in Table 3. As to the quali⁃ ty of the synthesized virtual view, the texture⁃5/6 and 7/8 CT⁃ DP formats after the DIBR process show better BDPSNR and BDBR performances than 2DDP format. It is noted that all syn⁃ thesized views do not perform any depth enhancement and depth preprocessing, and the hole filling used in the DIBR pro⁃ cess is the simple background extension. In summary, the texture qualities BDPSNR and BDBR in Ta⁃ bles 2 and 3 can be treated as the objective quality indices in 2D displays, while the virtual view qualities can be the objec⁃ tive quality indices in 3D displays. The results show that the proposed texture⁃5/6 and 7/8 CTDP format will be the better choices for the broadcasters. The texture⁃3/4 CTDP format has better 3D performance while texture⁃7/8 CTDP format achieves better 2D performance. 4.2 Performance Evaluation of Depth Enhancement for CTDP Format To verify the proposed depth enhancement mechanism, we first show the reconstructed depth from original and depth⁃en⁃ hanced CTDP formats. The RD curves for different ratios of CTDP formats are shown in Fig. 13. It can be seen that the pro⁃ posed refined CTDP format can always achieve better perfor⁃ mance. The gains between the depth⁃enhanced CTDP and the original CTDP formats are increased while the ratio of texture is increased. For the subjective evaluation, the partial portions of the re⁃ constructed depth for Shark sequence are shown in Fig. 14. It can be seen that the depth can be reconstructed well especial⁃ ly for the edge region by using the depth enhancements. In the following, we will compare the synthesis results. The partial portions of the generated views are shown in Fig. 15. From the results, the proposed CTDP format can successfully preserve the edges well of the synthesis views without the jag⁃ gy noise. 4.3 Comparison with Different Depth Interpolation Methods The comparison results of different depth interpolation meth⁃ ods are shown in Table 4 for Shark sequence at all⁃intra (ai) coding condition with QP=32. The symbols of Bi and BC de⁃ note the bilinear and bi⁃cubic convolution interpolation meth⁃ ods, respectively. The methods of JBU [19] and FEU [20] are the texture⁃similarity based depth interpolation methods. The proposed depth up ⁃ sampling method has better PSNR and SSIM results for reconstructed depth images in vertical⁃11/12 CTDP and vertical⁃23/24 CTDP formats. For the vertical⁃5/6 CTDP format, the proposed depth up⁃sampling method can al⁃ so provide better reconstructed depth images. The comparison results of partial reconstructed depth with different depth interpolation methods are shown in Fig. 16. The reconstructed depth images of bilinear and bi⁃cubic convo⁃ lution interpolation methods have serious jaggy noise among the edges. It can be seen that the proposed depth up⁃sampling method can outperform other methods with better edges. 5 Conclusions In this paper, we proposed depth enhancement processes for ▲Figure 12. Five texture and depth frame compatible packing formats. ▼Table 1. Averaged BDPSNR performances BDPSNR ai ra ldp avg Recovered texture performance 5/6 2.2693 2.5868 2.2079 2.3546 7/8 2.2587 2.71322 2.42228 2.464733 15/16 2.34168 2.86354 2.54206 2.582427 Synthesized virtual view performance 5/6 0.6150 0.7371 0.3920 0.5813 7/8 0.3664 0.7489 0.6005 0.5720 15/16 ⁃0.7732 ⁃0.2754 ⁃0.3928 ⁃0.4805 ▼Table 2. Averaged BDBR performances BDBR ai ra ldp avg Recovered texture performance 5/6 ⁃55.3282 ⁃61.0706 ⁃48.7147 ⁃55.0378 7/8 ⁃46.5149 ⁃59.5852 ⁃54.3501 ⁃53.4834 15/16 ⁃47.9069 ⁃61.6926 ⁃56.2303 ⁃55.2766 Synthesized virtual view performance 5/6 ⁃22.9490 ⁃24.6686 ⁃9.9111 ⁃19.1762 7/8 ⁃7.92186 ⁃24.6019 ⁃19.6368 ⁃17.3869 15/16 56.0625 30.0742 36.5666 40.9011 ▼Table 3. DIBR settings for virtual view synthesis Sequence Poznan hall Poznan street Kendo Balloons Newspaper Resolution 1920*1088 1920*1088 1024*768 1024*768 1024*768 Frames 200 250 300 300 300 Coded view 6 4 3 3 4 Synthesized view 5 3 4 5 6 (a) Poznan hall (b) Poznan street (c) Keno (d) Balloons (e) Newspaper Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS64
CTDP formats [10]. The CTDP formats can be comfortably and directly viewed in 2DTV displays without the need of any extra computation. However, the CTDP formats slightly suffer from the depth discontinuities for high texture ratios. Comparing to ▲Figure 13. RD curves of reconstructed depth from the original CTDP depacking process and the proposed depth⁃enhanced CTDP depacking process for (a) texture⁃5/6; (b) texture⁃11/12; (c) texture⁃23/24 formats. CTDP: compatible centralized texture⁃depth packing PSNR: peak signal⁃to⁃noise ratio ▲Figure 14. Partial portions of reconstructed depth in S10 Shark with the original CTDP depacked: (a) texture⁃5/6; (b) texture⁃11/12; (c) tex⁃ ture⁃23/24 formats and the proposed depth enhanced CTDP depacked; (d) texture⁃5/6; (e) texture⁃11/12; (f) texture⁃23/24 formats. ▲Figure 15. Partial synthesis views of S02 Poznan Street with original CTDP depacking: (a) texture⁃5/6; (b) texture⁃11/12; (c) texture⁃23/24 and the proposed depth⁃enhanced CTDP depacking; (d) texture⁃5/6; (e) texture⁃11/12; (f) texture⁃23/24. ▼Table 4. The PSNR and SSIM comparison of different depth interpolation methods in S10 Shark at All Intra (ai) QP=32 PSNR (dB) Vertical⁃5/6 CTDP Vertical⁃11/12 CTDP Vertical⁃23/24 CTDP SSIM Vertical⁃5/6 CTDP Vertical⁃11/12 CTDP Vertical⁃23/24 CTDP Bi 33.7164 31.7651 29.8525 Bi 0.9361 0.9147 0.8914 BC 33.5138 31.6581 29.7836 BC 0.9334 0.9122 0.8879 JBU 33.6252 32.0857 30.3490 JBU 0.9361 0.9192 0.8959 FEU 33.7135 31.7850 29.8953 FEU 0.9368 0.9158 0.8925 Proposed 33.5239 32.4422 30.5411 Proposed 0.9397 0.9270 0.9068 (a) (b) (c) (d) (e) (f) (a) (b) (c) (d) (e) (f) 40003500300025002000150010005000 41.0 40.5 40.0 39.5 39.0 38.5 38.0 37.5 37.0 36.5 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) (a) 350030002500200015005000 39.5 39.0 38.5 38.0 37.5 37.0 36.5 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) 36.0 35.5 35.0 (b) 4000 38.0 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) (c) 450040003500300025002000150010005000 37.5 37.0 36.5 36.0 35.5 35.0 34.5 34.0 33.5 33.0 Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 65 1000
the 2DDP format, the CTDP formats with the same video cod⁃ ing systems, such as AVS2 (RD 6.0) and HEVC [10], show bet⁃ ter coding performances in texture and depth frames and syn⁃ thesized virtual views. To further increase the visual quality, in this paper, the depth enhancement methods, including YCbCr calibration and texture ⁃ similarity ⁃ based depth up ⁃ sampling, are proposed. Experimental results reveal that the proposed depth enhancement can efficiently help to increase the depack⁃ ing performances of the CTDP formats to achieve better recon⁃ structed depth images and better synthesis views as well. With the aforementioned simulation results, we believe that the pro⁃ posed depth enhanced CTDP depacking methods will be a greatly⁃advanced system for current 2D video coding systems, which can provide 3D video services effectively and simply. ▲Figure 16. The comparison of partial reconstructed depth with differ⁃ ent depth interpolation methods for vertical ⁃ 11/12 CTDP format: (a) ground truth; (b) bilinear; (c) bi ⁃ cubic convolution; (d) JBU [19]; (e) FEU [20]; (f) proposed. References [1] J.⁃F. Yang, H.⁃M. Wang, K.⁃I. Liao, L. Yu, and J.⁃R. Ohm,“Centralized texture⁃ depthpacking formats for effective 3D video transmission over current video broadcasting systems,”IEEE Transactions on Circuits and Systems for Video Technology, submitted for publication. [2] Dolby Laboratories, Inc. (2015). Dolby Open Specification for Frame⁃Compatible 3D Systems [Online]. Available: http://www.dolby.com [3] ITU. (2015). Advanced Video Coding for Generic Audio—Visual Services [Online]. Available: http://www.itu.int [4] G. Sullivan, T. Wiegand, D. Marpe, and A. Luthra,“Text of ISO/IEC 14496⁃10 advanced video coding (third edition),”ISO/IEC JTC 1/SC 29/WG11, Redmond, USA, Doc. N6540, Jul. 2004. [5] G. J. Sullivan, A. M. Tourapis, T. Yamakage, and C. S. Lim,“ISO/IEC 14496⁃10: 200X/FPDAM 1,”ISO/IEC JTC 1/SC 29/WG11, Apr. 2009. [6] T. Kanade and M. Okutomi,“A stereo matching algorithm with an adaptive win⁃ dow: theory and experiment,”IEEE Transactions on Pattern Analysis and Match⁃ ing Intelligence, vol. 16, no. 9, pp.920-932, Sept. 1994. doi:10.1109/34.310690. [7] K. Zhang, J. Lu, and G. Lafruit,“Cross⁃based local stereo matching using orthog⁃ onal integral images,”IEEE Transactions on Circuits and Systems for Video Tech⁃ nology, vol. 19, no. 7, pp.1073- 1079, Jul. 2009. doi: 10.1109/TCS⁃ VT.2009.2020478. [8] S.⁃C. Chan, H.⁃Y. Shum, and K.⁃T. Ng,“Image⁃based rendering and synthesis,” IEEE Signal Processing Magazine, vol. 24, no. 6, pp. 22-33, Nov. 2007. doi: 10.1109/MSP.2007.905702. [9] T.⁃C. Yang, P.⁃C. Kuo, B.⁃D. Liu, and J.⁃F. Yang,“Depth image⁃based rendering with edge⁃oriented hole filling for multiview synthesis,”in Proc. International Conference on Communications, Circuits and Systems, Chengdu, China, Nov. 2013, vol. 1, pp. 50-53. doi: 10.1109/ICCCAS.2013.6765184. [10] Philips 3D Solutions,“3D interface specifications, white paper,”Eindhoven, The Netherlands, Dec. 2006. [11] Studio Encoding Parameters of Digital Television for Standard 4:3 and Wide⁃ Screen 16:9 Aspect Ratios, ITU⁃R BT.601⁃5, 1995. [12] J.⁃F. Yang, K.⁃Y. Liao, H.⁃M. Wang, and Y.⁃H. Hu,“Centralized texture⁃depth packing (CTDP) SEI message syntax,”Joint Collaborative Team on 3D Video Coding Extensions of ITU⁃T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11, Strasbourg, France, Doc. no. JCT3V⁃J0108, Oct. 2014. [13] J.⁃F. Yang, K.⁃Y. Liao, H.⁃M. Wang, and C.⁃Y. Chen,“Centralized texture⁃ depth packing (CTDP) SEI message,”Joint Collaborative Team on 3D Video Coding Extensions of ITU⁃T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11, Geneva, Switzerland, Doc. no. JCT3V⁃K0027, Feb. 2015. [14] J.⁃F. Yang, H.⁃M Wang, Y.⁃A. Chiang, and K. Y. Liao,“2D frame compatible centralized color depthpacking format (translated from Chinese),”AVS 47th Meeting, Beijing, China, AVS⁃M3225, Dec. 2013. [15] J.⁃F. Yang, H.⁃M. Wang, K.⁃Y. Liao, and Y.⁃A. Chiang,“AVS2 syntax message for 2D frame compatible centralized color depth packing formats (translated from Chinese),”AVS 50th Meeting, Nanjing, China, AVS⁃M3472, Oct. 2014. [16] H. C. Andrews and C. L. Patterson,“Digital interpolation of discrete images,” IEEE Transaction on Computers, vol. 25, no. 2, 1976. [17] X.⁃Z. Zheng,“AVS2⁃P2 common test conditions (translated from Chinese),” AVS 46th Meeting, Shenyang, China, AVS⁃N2001, Sep. 2013. [18] G. Bjontegaard,“Calculation of average PSNR differences between RD ⁃ curves,”Austin, USA, Doc. VCEG⁃M33 ITU⁃T Q6/16, Apr. 2001. [19] J. Kopf, M. F. Cohen, D. Lischinski, and M. Uyttendaele,“Joint bilateral upsampling,”ACM Transaction on Graphics, vol. 26, no. 3, Article 96, Jul. 2007. doi:10.1145/1275808.1276497. [20] S.⁃Y. Kim and Y.⁃S. Ho,“Fast edge⁃preserving depth image upsampler,”Jour⁃ nal of Consumer Electronics, vol. 58, no. 3, pp. 971- 977, Aug. 2012. doi: 10.1109/TCE.2012.6311344. Manuscript received: 2015⁃11⁃12 YANG Jar⁃Ferr (jefyang@mail.ncku.edu.tw) received his PhD degree from the Uni⁃ versity of Minnesota, USA in 1988. He joined the National Cheng Kung University (NCKU) started from an associate professor in 1988 and became a full professor and distinguished professor in 1995 and 2007. He was the chairperson of Graduate Insti⁃ tute of Computer and Communication Engineering during 2004-2008 and the direc⁃ tor of the Electrical and Information Technology Center 2006-2008 in NCKU. He was the associate vice president for Research and Development of the NCKU. Cur⁃ rently, he is a distinguished professor and the director of Technologies of Ubiqui⁃ tous Computing and Humanity (TOUCH) Center supported by National Science Council (NSC), Taiwan, China. Furthermore, he is the director of Tomorrow Ubiqui⁃ tous Cloud and Hypermedia (TOUCH) Service Center. During 2004-2005, he was selected as a speaker in the Distinguished Lecturer Program by the IEEE Circuits and Systems Society. He was the secretary, and the chair of IEEE Multimedia Sys⁃ tems and Applications Technical Committee and an associate editor of IEEE Trans⁃ action on Circuits and Systems for Video Technology. In 2008, he received the NSC Excellent Research Award. In 2010, he received the Outstanding Electrical Engi⁃ neering Professor Award of the Chinese Institute of Electrical Engineering, Taiwan, China. He was the chairman of IEEE Tainan Section during 2009-2011. Currently, he is an associate editor of EURASIP Journal of Advances in Signal Processing and an editorial board member of IET Signal Processing. He has published 104 journal and 167 conference papers. He is a fellow of IEEE. WANG Hung ⁃ Ming (ming@video5.ee.ncku.edu.tw) received the BS and PhD de⁃ grees in electrical engineering from National Cheng Kung University (NCKU), Tai⁃ wan, China in 2003 and 2009, respectively. He is currently a senior engineer of No⁃ vatek Microelectronics Corp., Taiwan, China. His major research interests include 2D/3D image processing, video coding and multimedia communication. LIAO Wei⁃Chen (a800812momo@gmail.com) received the BS and MS degrees in electrical engineering from National Cheng Kung University (NCKU), Taiwan, Chi⁃ na in 2013 and 2015, respectively. His major research interests include image pro⁃ cessing, video coding and multimedia communication. BiographiesBiographies (a) (b) (c) (d) (e) (f) Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS66
• Remit of Journal ZTE Communications publishes original theoretical papers, research findings, and surveys on a broad range of communica⁃ tions topics, including communications and information system design, optical fiber and electro-optical engineering, micro⁃ wave technology, radio wave propagation, antenna engineering, electromagnetics, signal and image processing, and power engineering. The journal is designed to be an integrated forum for university academics and industry researchers from around the world. • Manuscript Preparation Manuscripts must be typed in English and submitted electronically in MS Word (or compatible) format. The word length is ap⁃ proximately 3000 to 8000, and no more than 8 figures or tables should be included. Authors are requested to submit mathe⁃ matical material and graphics in an editable format. • Abstract and Keywords Each manuscript must include an abstract of approximately 150 words written as a single paragraph. The abstract should not include mathematics or references and should not be repeated verbatim in the introduction. The abstract should be a self- contained overview of the aims, methods, experimental results, and significance of research outlined in the paper. Five care⁃ fully chosen keywords must be provided with the abstract. • References Manuscripts must be referenced at a level that conforms to international academic standards. All references must be num⁃ bered sequentially in-text and listed in corresponding order at the end of the paper. References that are not cited in-text should not be included in the reference list. References must be complete and formatted according to ZTE Communications Editorial Style. A minimum of 10 references should be provided. Footnotes should be avoided or kept to a minimum. • Copyright and Declaration Authors are responsible for obtaining permission to reproduce any material for which they do not hold copyright. Permission to reproduce any part of this publication for commercial use must be obtained in advance from the editorial office of ZTE Com⁃ munications. Authors agree that a) the manuscript is a product of research conducted by themselves and the stated co-au⁃ thors, b) the manuscript has not been published elsewhere in its submitted form, c) the manuscript is not currently being con⁃ sidered for publication elsewhere. If the paper is an adaptation of a speech or presentation, acknowledgement of this is re⁃ quired within the paper. The number of co-authors should not exceed five. • Content and Structure ZTE Communications seeks to publish original content that may build on existing literature in any field of communications. Au⁃ thors should not dedicate a disproportionate amount of a paper to fundamental background, historical overviews, or chronolo⁃ gies that may be sufficiently dealt with by references. Authors are also requested to avoid the overuse of bullet points when structuring papers. The conclusion should include a commentary on the significance/future implications of the research as well as an overview of the material presented. • Peer Review and Editing All manuscripts will be subject to a two-stage anonymous peer review as well as copyediting, and formatting. Authors may be asked to revise parts of a manuscript prior to publication. • Biographical Information All authors are requested to provide a brief biography (approx. 100 words) that includes email address, educational back⁃ ground, career experience, research interests, awards, and publications. • Acknowledgements and Funding A manuscript based on funded research must clearly state the program name, funding body, and grant number. Individuals who contributed to the manuscript should be acknowledged in a brief statement. • Address for Submission magazine@zte.com.cn 12F Kaixuan Building, 329 Jinzhai Rd, Hefei 230061, P. R. China ZTE Communications Guidelines for Authors
ZTECOMMUNICATIONS •Cambridge Scientific Abstracts (CSA) •China Science and Technology Journal Database •Chinese Journal Fulltext Databases •Inspec •Ulrich’s Periodicals Directory •Wanfang Data—Digital Periodicals ZTE Communications has been indexed in the following databases:

More Related Content

PPTX
RF System design concepts
HedayathBashaShaik1
 
PDF
Millimeter wave as the future of 5g
Saurabh Verma
 
PPTX
Equal Split Wilkinson Power Divider - Project Presentation
Bhanwar Singh Meena
 
PPTX
Chap 5 (small scale fading)
asadkhan1327
 
PPTX
Software Defined Network - SDN
Venkata Naga Ravi
 
PPTX
MIMO.ppt (2) 2
Deepak Kumar Dhal
 
PPT
Prioritizing handoffs
AJAL A J
 
PPT
Wcdma
Sajan Sahu
 
RF System design concepts
HedayathBashaShaik1
 
Millimeter wave as the future of 5g
Saurabh Verma
 
Equal Split Wilkinson Power Divider - Project Presentation
Bhanwar Singh Meena
 
Chap 5 (small scale fading)
asadkhan1327
 
Software Defined Network - SDN
Venkata Naga Ravi
 
MIMO.ppt (2) 2
Deepak Kumar Dhal
 
Prioritizing handoffs
AJAL A J
 
Wcdma
Sajan Sahu
 

What's hot (20)

PPT
Wireless sensor Networks.ppt
MATRUSRI ENGINEERING COLLEGE
 
PDF
Orthogonal Frequency Division Multiplexing (OFDM)
ashishsoni1505
 
PPTX
W cdma
Deepak Kumar Dhal
 
PPT
Chapter 7 multiple access techniques
Kaushal Kabra
 
PPT
Introduction To Wireless Fading Channels
Nitin Jain
 
PPT
Tetra Ppt 6604 Edited
mamach4
 
PDF
Setting off the 5G Advanced evolution with 3GPP Release 18
Qualcomm Research
 
PPT
Multiple access techniques
Bibhu Prasad
 
PPT
D2D - Device to Device Communication
Francisco Bento da Silva Neto
 
PDF
5G physical layer
Ali Nikfal
 
PPT
Channel assignment strategies
AJAL A J
 
PDF
MIMO Channel Capacity
Pei-Che Chang
 
PPT
1g,2g,3g,4g and 5g
Gokulakumar
 
PDF
LTE Basics
Praveen Kumar
 
PPT
Modulation
sristykp
 
PPTX
MIMO Calculation
Abhijeet Kumar
 
PPTX
Orthogonal Frequency Division Multiplexing (OFDM)
Gagan Randhawa
 
DOCX
Improving coverage and capacity in cellular systems
Tarek Nader
 
PPT
Fading in wireless communication.ppt
pravin patil
 
PPTX
Mac protocols
juno susi
 
Wireless sensor Networks.ppt
MATRUSRI ENGINEERING COLLEGE
 
Orthogonal Frequency Division Multiplexing (OFDM)
ashishsoni1505
 
W cdma
Deepak Kumar Dhal
 
Chapter 7 multiple access techniques
Kaushal Kabra
 
Introduction To Wireless Fading Channels
Nitin Jain
 
Tetra Ppt 6604 Edited
mamach4
 
Setting off the 5G Advanced evolution with 3GPP Release 18
Qualcomm Research
 
Multiple access techniques
Bibhu Prasad
 
D2D - Device to Device Communication
Francisco Bento da Silva Neto
 
5G physical layer
Ali Nikfal
 
Channel assignment strategies
AJAL A J
 
MIMO Channel Capacity
Pei-Che Chang
 
1g,2g,3g,4g and 5g
Gokulakumar
 
LTE Basics
Praveen Kumar
 
Modulation
sristykp
 
MIMO Calculation
Abhijeet Kumar
 
Orthogonal Frequency Division Multiplexing (OFDM)
Gagan Randhawa
 
Improving coverage and capacity in cellular systems
Tarek Nader
 
Fading in wireless communication.ppt
pravin patil
 
Mac protocols
juno susi
 
Ad

Viewers also liked (20)

PDF
Optical Wireless Communications
Sitha Sok
 
PDF
5G network energy efficiency
Sitha Sok
 
PDF
APT 700 LTE Whitepaper Huawei
Sitha Sok
 
PDF
ZTE Mobile World: issue No.5
Sitha Sok
 
PDF
GSA-Evolution_to_LTE_report_October_2016
Sitha Sok
 
PDF
splashmedia-ZTE-CES-2017
Jason Kauzlarich
 
PDF
Zte mobile world 2015
Sitha Sok
 
PDF
Mobile World - ZTE
Sitha Sok
 
PDF
ZTE ZXHN H108L Router Configuration Guide
SriLankaTelecom
 
PPTX
Outcome of WRC 15 and CPM 19-1 conferences
CA
 
PDF
Phydyas 09 fFilter Bank Multicarrier (FBMC): An Integrated Solution to Spectr...
Marwan Hammouda
 
PDF
RA 15, WRC-15 and CPM19-1: Suymmary of Outcomes
Aamir Riaz
 
PPTX
ZTE-Product PPT
hemant chauhan
 
PPTX
5G wireless technology and internet of things
Bhadra Gowdra
 
PDF
Zte technology 2015
Sitha Sok
 
PDF
GSA presentation at GTI spectrum workshop at ITU bangkok
Sitha Sok
 
PDF
ZTE COMMUNICATIONS No.2 2016 : Optical wireless communications
Sitha Sok
 
PDF
Quality of Service at the Internet Engineering Task Force
John Loughney
 
PPT
3b multiple access
kavish dani
 
PDF
ZTE TECHNOLOGIES No.3 2016 - Special topic: Big Video
Sitha Sok
 
Optical Wireless Communications
Sitha Sok
 
5G network energy efficiency
Sitha Sok
 
APT 700 LTE Whitepaper Huawei
Sitha Sok
 
ZTE Mobile World: issue No.5
Sitha Sok
 
GSA-Evolution_to_LTE_report_October_2016
Sitha Sok
 
splashmedia-ZTE-CES-2017
Jason Kauzlarich
 
Zte mobile world 2015
Sitha Sok
 
Mobile World - ZTE
Sitha Sok
 
ZTE ZXHN H108L Router Configuration Guide
SriLankaTelecom
 
Outcome of WRC 15 and CPM 19-1 conferences
CA
 
Phydyas 09 fFilter Bank Multicarrier (FBMC): An Integrated Solution to Spectr...
Marwan Hammouda
 
RA 15, WRC-15 and CPM19-1: Suymmary of Outcomes
Aamir Riaz
 
ZTE-Product PPT
hemant chauhan
 
5G wireless technology and internet of things
Bhadra Gowdra
 
Zte technology 2015
Sitha Sok
 
GSA presentation at GTI spectrum workshop at ITU bangkok
Sitha Sok
 
ZTE COMMUNICATIONS No.2 2016 : Optical wireless communications
Sitha Sok
 
Quality of Service at the Internet Engineering Task Force
John Loughney
 
3b multiple access
kavish dani
 
ZTE TECHNOLOGIES No.3 2016 - Special topic: Big Video
Sitha Sok
 
Ad

Similar to Multiple Access Techniques for 5G (20)

PDF
ZTE Communication - March 2015
Sitha Sok
 
PDF
5G New Radio Non-Orthogonal Multiple Access 1st Edition Yifei Yuan
dharensoreyn98
 
PDF
5G New Radio Non-Orthogonal Multiple Access 1st Edition Yifei Yuan
dilysnoemen
 
PDF
Review Intelligent Reflective Surface Based Non Orthogonal Multiple Access (1...
manaswilatthe
 
PDF
Multiple Access Techniques for 5G Wireless Networks and Beyond
Khamis37
 
PDF
Performance Enhancement of 5G Networks: Remodeling Power Domain Scheme Throug...
IJAEMSJORNAL
 
PDF
ZTE Leading 5G Innovations
ssuser220dc6
 
PDF
NOMA.pdf
Ali Mohseni Ahouei
 
PDF
LDS_IEEE_Access2020_Lajos_Final_Draft (1).pdf
duniabt4
 
PDF
Challenges, Issues and Role of the 5G Network to the new developments
IRJET Journal
 
PDF
Channel Models and Measurements for 5G.pdf
umere15
 
PDF
Multiple Access Technology Towards Ubiquitous Networks Overview And Efficient...
tabloshozi
 
PDF
Multiple Access Technology Towards Ubiquitous Networks Overview And Efficient...
tabloshozi
 
PDF
Monet cf ps-5_g networks_final(1)
Fahmi Yasser
 
PDF
Digital Twins For 6g Fundamental Theory Technology And Applications Hamed Ahmadi
ojasteemrija82
 
PDF
Broadband Wireless Mobile 3G and Beyond 1st Edition Willie W. Lu
puddinyiyu
 
PDF
Spectrum Efficiency improvement in 5G Network using NOMA
IJAEMSJORNAL
 
PDF
Microstrip and Printed Antenna Design Telecommunications 3rd Edition Randy B...
yaoubamenalo
 
PDF
On the performance of non-orthogonal multiple access (NOMA) using FPGA
IJECEIAES
 
PDF
5g Nr The Next Generation Wireless Access Technology Erik Dahlman
suingawanaso
 
ZTE Communication - March 2015
Sitha Sok
 
5G New Radio Non-Orthogonal Multiple Access 1st Edition Yifei Yuan
dharensoreyn98
 
5G New Radio Non-Orthogonal Multiple Access 1st Edition Yifei Yuan
dilysnoemen
 
Review Intelligent Reflective Surface Based Non Orthogonal Multiple Access (1...
manaswilatthe
 
Multiple Access Techniques for 5G Wireless Networks and Beyond
Khamis37
 
Performance Enhancement of 5G Networks: Remodeling Power Domain Scheme Throug...
IJAEMSJORNAL
 
ZTE Leading 5G Innovations
ssuser220dc6
 
NOMA.pdf
Ali Mohseni Ahouei
 
LDS_IEEE_Access2020_Lajos_Final_Draft (1).pdf
duniabt4
 
Challenges, Issues and Role of the 5G Network to the new developments
IRJET Journal
 
Channel Models and Measurements for 5G.pdf
umere15
 
Multiple Access Technology Towards Ubiquitous Networks Overview And Efficient...
tabloshozi
 
Multiple Access Technology Towards Ubiquitous Networks Overview And Efficient...
tabloshozi
 
Monet cf ps-5_g networks_final(1)
Fahmi Yasser
 
Digital Twins For 6g Fundamental Theory Technology And Applications Hamed Ahmadi
ojasteemrija82
 
Broadband Wireless Mobile 3G and Beyond 1st Edition Willie W. Lu
puddinyiyu
 
Spectrum Efficiency improvement in 5G Network using NOMA
IJAEMSJORNAL
 
Microstrip and Printed Antenna Design Telecommunications 3rd Edition Randy B...
yaoubamenalo
 
On the performance of non-orthogonal multiple access (NOMA) using FPGA
IJECEIAES
 
5g Nr The Next Generation Wireless Access Technology Erik Dahlman
suingawanaso
 

More from Sitha Sok (20)

PDF
850 MHz & 900 MHz Co-Existence
Sitha Sok
 
PDF
MIGRATION STRATEGY OF GSM TO MOBILE BROADBAND
Sitha Sok
 
PDF
LTE Call Processing and Handover
Sitha Sok
 
PDF
LTE KPI
Sitha Sok
 
PDF
5G Spectrum in EUROPE
Sitha Sok
 
PDF
Mobile World No.2 2016 : Zte spro plus
Sitha Sok
 
PDF
Microwave backhaul gets a boost with multiband
Sitha Sok
 
PDF
NB-IoT: a sustainable technology for connecting billions of devices
Sitha Sok
 
PDF
ICT Insights Issue 17 (03/2016)
Sitha Sok
 
PDF
Emerging Technologies of Future Multimedia Coding, Analysis and Transmission
Sitha Sok
 
PDF
Special Topic on 5G Evolution
Sitha Sok
 
PDF
Mobile World » 2016 » No.1 ZTE Blade S7
Sitha Sok
 
PDF
NGMN 5G WHITE PAPER
Sitha Sok
 
PDF
5G Spectrum Recommendations White Paper
Sitha Sok
 
PDF
Magazine Mobile World: Current Issue No.6
Sitha Sok
 
PDF
ZTE Communications - Vol No.4
Sitha Sok
 
PDF
Special Toptic on Smart Operations
Sitha Sok
 
PDF
DOCOMO 5G White Paper
Sitha Sok
 
PDF
Understanding 5G by GSMA
Sitha Sok
 
PDF
5G vision-brochure-v1
Sitha Sok
 
850 MHz & 900 MHz Co-Existence
Sitha Sok
 
MIGRATION STRATEGY OF GSM TO MOBILE BROADBAND
Sitha Sok
 
LTE Call Processing and Handover
Sitha Sok
 
LTE KPI
Sitha Sok
 
5G Spectrum in EUROPE
Sitha Sok
 
Mobile World No.2 2016 : Zte spro plus
Sitha Sok
 
Microwave backhaul gets a boost with multiband
Sitha Sok
 
NB-IoT: a sustainable technology for connecting billions of devices
Sitha Sok
 
ICT Insights Issue 17 (03/2016)
Sitha Sok
 
Emerging Technologies of Future Multimedia Coding, Analysis and Transmission
Sitha Sok
 
Special Topic on 5G Evolution
Sitha Sok
 
Mobile World » 2016 » No.1 ZTE Blade S7
Sitha Sok
 
NGMN 5G WHITE PAPER
Sitha Sok
 
5G Spectrum Recommendations White Paper
Sitha Sok
 
Magazine Mobile World: Current Issue No.6
Sitha Sok
 
ZTE Communications - Vol No.4
Sitha Sok
 
Special Toptic on Smart Operations
Sitha Sok
 
DOCOMO 5G White Paper
Sitha Sok
 
Understanding 5G by GSMA
Sitha Sok
 
5G vision-brochure-v1
Sitha Sok
 

Recently uploaded (20)

PDF
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
PDF
NewMind AI Weekly Chronicles - August'25 Week I
NewMind AI
 
DOCX
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
PPTX
Detection-First SIEM: Rule Types, Dashboards, and Threat-Informed Strategy
ReZa AdineH
 
PPTX
Digital-Transformation-Roadmap-for-Companies.pptx
David Malick Dieng
 
PDF
Spectral efficient network and resource selection model in 5G networks
IAESIJAI
 
PDF
Building Integrated photovoltaic BIPV_UPV.pdf
SandeepSingh286037
 
PDF
Review of recent advances in non-invasive hemoglobin estimation
IAESIJAI
 
PDF
Advanced methodologies resolving dimensionality complications for autism neur...
IAESIJAI
 
PPTX
Cloud computing and distributed systems.
kkkkkhan
 
PDF
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Akar16
 
PPTX
A Presentation on Artificial Intelligence
memembruh336
 
PPTX
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
ReZa AdineH
 
PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
PDF
Chapter 3 Spatial Domain Image Processing.pdf
Getnet Tigabie Askale -(GM)
 
PPTX
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
muhammadmuneebnaeem7
 
PDF
Modernizing your data center with Dell and AMD
Principled Technologies
 
PPT
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
PPTX
Big Data Technologies - Introduction.pptx
kuthubussaman1
 
PDF
Empathic Computing: Creating Shared Understanding
Mark Billinghurst
 
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
SynapseIndia
 
NewMind AI Weekly Chronicles - August'25 Week I
NewMind AI
 
The AUB Centre for AI in Media Proposal.docx
GAONE9
 
Detection-First SIEM: Rule Types, Dashboards, and Threat-Informed Strategy
ReZa AdineH
 
Digital-Transformation-Roadmap-for-Companies.pptx
David Malick Dieng
 
Spectral efficient network and resource selection model in 5G networks
IAESIJAI
 
Building Integrated photovoltaic BIPV_UPV.pdf
SandeepSingh286037
 
Review of recent advances in non-invasive hemoglobin estimation
IAESIJAI
 
Advanced methodologies resolving dimensionality complications for autism neur...
IAESIJAI
 
Cloud computing and distributed systems.
kkkkkhan
 
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Akar16
 
A Presentation on Artificial Intelligence
memembruh336
 
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
ReZa AdineH
 
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
Chapter 3 Spatial Domain Image Processing.pdf
Getnet Tigabie Askale -(GM)
 
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
muhammadmuneebnaeem7
 
Modernizing your data center with Dell and AMD
Principled Technologies
 
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
Big Data Technologies - Introduction.pptx
kuthubussaman1
 
Empathic Computing: Creating Shared Understanding
Mark Billinghurst
 

Multiple Access Techniques for 5G

  • 1. ISSN 1673-5188 CN 34-1294/ TN CODEN ZCTOAK ZTECOMMUNICATIONSVOLUME14NUMBER4OCTOBER2016 tech.zte.com.cn ZTECOMMUNICATIONS October 2016, Vol. 14 No. 4An International ICT R&D Journal Sponsored by ZTE Corporation SPECIAL TOPIC: Multiple Access Techniques for 5G
  • 2. ZTE Communications Editorial Board Members (in Alphabetical Order): Chairman ZHAO Houlin: International Telecommunication Union (Switzerland) Vice Chairmen SHI Lirong: ZTE Corporation (China) XU Chengzhong: Wayne State University (USA) CAO Jiannong Hong Kong Polytechnic University (Hong Kong, China) CHEN Chang Wen University at Buffalo, The State University of New York (USA) CHEN Jie ZTE Corporation (China) CHEN Shigang University of Florida (USA) CHEN Yan Northwestern University (USA) Connie Chang􀆼Hasnain University of California, Berkeley (USA) CUI Shuguang University of California, Davis (USA) DONG Yingfei University of Hawaii (USA) GAO Wen Peking University (China) HWANG Jenq􀆼Neng University of Washington (USA) LI Guifang University of Central Florida (USA) LUO Fa􀆼Long Element CXI (USA) MA Jianhua Hosei University (Japan) PAN Yi Georgia State University (USA) REN Fuji The University of Tokushima (Japan) SHI Lirong ZTE Corporation (China) SONG Wenzhan University of Georgia (USA) SUN Huifang Mitsubishi Electric Research Laboratories (USA) SUN Zhili University of Surrey (UK) Victor C. M. Leung The University of British Columbia (Canada) WANG Xiaodong Columbia University (USA) WANG Zhengdao Iowa State University (USA) WU Keli The Chinese University of Hong Kong (Hong Kong, China) XU Chengzhong Wayne State University (USA) YANG Kun University of Essex (UK) YUAN Jinhong University of New South Wales (Australia) ZENG Wenjun Microsoft Research Asia (USA) ZHANG Chengqi University of Technology Sydney (Australia) ZHANG Honggang Zhejiang University (China) ZHANG Yueping Nanyang Technological University (Singapore) ZHAO Houlin International Telecommunication Union (Switzerland) ZHOU Wanlei Deakin University (Australia) ZHUANG Weihua University of Waterloo (Canada)
  • 3. CONTENTSCONTENTS Submission of a manuscript implies that the submitted work has not been published before (except as part of a thesis or lecture note or report or in the form of an abstract); that it is not under consideration for publication elsewhere; that its publication has been approved by all co- authors as well as by the authorities at the institute where the work has been carried out; that, if and when the manuscript is accepted for publication, the authors hand over the transferable copyrights of the accepted manuscript to ZTE Communications; and that the manuscript or parts thereof will not be published elsewhere in any language without the consent of the copyright holder. Copyrights include, without spatial or timely limitation, the mechanical, electronic and visual reproduction and distribution; electronic storage and retrieval; and all other forms of electronic publication or any other types of publication including all subsidiary rights. Responsibility for content rests on authors of signed articles and not on the editorial board of ZTE Communications or its sponsors. All rights reserved. Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng 01 Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei 11 Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong 03 Special Topic: Multiple Access Techniques for 5G A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo 17 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I 26 ISSN 1673-5188 CN 34-1294/ TN CODEN ZCTOAK tech.zte.com.cn ZTECOMMUNICATIONS October 2016, Vol. 14 No. 4An International ICT R&D Journal Sponsored by ZTE Corporation SPECIAL TOPIC: Multiple Access Techniques for 5G
  • 4. ZTE COMMUNICATIONS Vol. 14 No. 4 (Issue 53) Quarterly First English Issue Published in 2003 Supervised by: Anhui Science and Technology Department Sponsored by: Anhui Science and Technology Information Research Institute and ZTE Corporation Staff Members: Editor-in-Chief: CHEN Jie Executive Associate Editor-in-Chief: HUANG Xinming Editor-in-Charge: ZHU Li Editors: XU Ye, LU Dan, ZHAO Lu Producer: YU Gang Circulation Executive: WANG Pingping Assistant: WANG Kun Editorial Correspondence: Add: 12F Kaixuan Building, 329 Jinzhai Road, Hefei 230061, P. R. China Tel: +86-551-65533356 Fax: +86-551-65850139 Email: magazine@zte.com.cn Published and Circulated (Home and Abroad) by: Editorial Office of ZTE Communications Printed by: Hefei Tiancai Color Printing Company Publication Date: October 25, 2016 Publication Licenses: Advertising License: 皖合工商广字0058号 Annual Subscription: RMB 80 ISSN 1673-5188 CN 34-1294/ TN CONTENTSCONTENTS Roundup New Members of ZTE Communications Editorial Board 57 Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen 58 Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu 50 Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu 35 Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson 42
  • 5. Multiple Access Techniques forMultiple Access Techniques for 55GG ▶ YUAN Jinhong YUAN Jinhong received his BE and PhD degrees in electronics engineering from Beijing Institute of Technology in 1991 and 1997. From 1997 to 1999, he was a research fellow at the School of Electrical Engineering, University of Sydney, Australia. In 2000, he joined the School of Electrical Engineering and Tele⁃ communications, University of New South Wales, Australia, and is currently a professor of telecommunications there. Dr. Yuan has authored two books, three book chapters, and more than 200 papers for telecom journals and conferences. He has also au⁃ thored 40 industry reports. He is a co⁃inventor of one patent on MIMO systems and two patents on low⁃density parity⁃check (LDPC) codes. He has co⁃ authored three papers that won Best Paper Awards or Best Poster Awards. Dr. Yuan served as the NSW Chair of the joint Communications/Signal Processions/Ocean Engi⁃ neering Chapter of IEEE during 2011-2014. He is an IEEE fellow and an associate edi⁃ tor for IEEE Transactions on Communications. His research interests include error⁃con⁃ trol coding and information theory, communication theory, and wireless communications. ver the past few decades, wireless communications have advanced tremendously and have become an indispensable part of our lives. Wireless networks have become more and more pervasive in order to guarantee global digital connectivity. Wireless devices have quickly evolved into multimedia smartphones running applications that demand high⁃speed and high⁃quality data connections. The upcoming fifth generation (5G) mobile cellular networks are required to provide significant increase in network throughput, cell⁃edge data rates, massive connectivity, superior spectrum efficiency, high energy efficiency and low latency, compared with the currently deployed long⁃term evolution (LTE) and LTE⁃advanced networks. To meet these demanding challenges of 5G networks, innovative technologies on radio air⁃interface and radio access network (RAN) are of great importance in PHY designs. Recently non⁃orthogonal multiple access (NOMA) has at⁃ tracted increasing research interests from both academic and industrial fields as a potential radio access tech⁃ nique. A few examples include multiuser shared access (MUSA), sparse code multiple access (SCMA), resource spread multiple access (RSMA) and pattern division multiple access (PDMA) proposed by ZTE, Huawei, Qual⁃ comm, DTmobile, etc. In the mean time, multicarrier (MC) technologies that divide frequency spectrum into many narrow subchannels, such as filter bank multicarrier (FBMC) and generalized frequency division multiplexing (GFDM), become attractive and new concepts for dynamic access spectrum management and cognitive radio appli⁃ cations. With these new developments, this special issue is dedicated to multiple access transmission technologies and O October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 01 ▶ XIANG Jiying XIANG Jiying, PhD, is the Chief Scientist of ZTE Corporation. His research is focused on 3G, 4G, 5G, and multi⁃mode wireless infrastructure technologies. He led the development of the first commercial SDR base station in the industry in 2007. He pro⁃ posed the first solution that support COMP on non⁃ideal back⁃ haul (also called Cloud Radio) in 2012. In 2014, he proposed the “pre⁃5G”conception, which includes massive MIMO, D⁃MIMO, MUSA, and UDN. Pre⁃5G allows 5G⁃like user experience on lega⁃ cy 4G handsets. Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng Special Topic ▶ DING Zhiguo DING Zhiguo received his BEng in electrical engineering from Beijing University of Posts and Telecommunications, China in 2000, and the PhD degree in electrical engineering from Imperial College London, UK in 2005. From Jul. 2005 to Aug. 2014, he worked in Queen’s University Belfast, Imperial College and New⁃ castle University, UK. Since Sept. 2014, he has been with Lan⁃ caster University, UK as a chair professor. From Oct. 2012 to Sept. 2017, he has also been an academic visitor in Princeton University, USA. His research interests are 5G networks, game theory, cooperative and energy harvesting networks, and statisti⁃ cal signal processing. He is serving as an editor for IEEE Transactions on Communica⁃ tions, IEEE Transactions on Vehicular Technology, IEEE Wireless Communication Let⁃ ters, IEEE Communication Letters, and Journal of Wireless Communications and Mo⁃ bile Computing. He received the best paper award in IET Comm. Conf. on Wireless, Mo⁃ bile and Computing, 2009, IEEE Communication Letter Exemplary Reviewer 2012, and the EU Marie Curie Fellowship 2012-2014. ▶ YUAN Zhifeng YUAN Zhifeng received his MS degree in signal and information processing from Nanjing University of Post and Telecommunica⁃ tions, China in 2005. He has been working at the Wireless Tech⁃ nology Advance Research Department, ZTE Corporation since 2006 and as the leader of the New Multi⁃Access (NMA) for 5G Wireless System Team since 2012. His research interests include wireless communications, MIMO systems, information theory, multiple access, error control coding, adaptive algorithm, and high⁃speed VLSI design.
  • 6. related for 5G cellular mobile communications. The main focus is on the cutting⁃edge research, review and application on non⁃ orthogonal multiple access and related signal processing and coding methods for the air ⁃ interface of 5G enhanced mobile broadband (eMBB), mMTC, and ultra reliable and low latency communication (URLLC). Papers for this issue were invited, and after peer review, six were selected for publication. The se⁃ lected papers cover reviews of various uplink and downlink NOMA schemes, novel designs for MIMO⁃FBMC systems, re⁃ view and new designs on multiple access technologies for cellu⁃ lar M2M communications and IoT applications. This issue is intended to be a timely, high⁃quality forum for scientists and engineers. In“Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems”by Taheri, Ghoraishi, XIAO, CAO and GAO, the authors discuss a candidate waveform design for fu⁃ ture wireless communications based on MIMO⁃FBMC and tack⁃ le the challenging problem of channel estimation facing the waveform design. Specifically, they propose a novel channel es⁃ timation method which employs intrinsic interference cancella⁃ tion at the transmitter side. Their research results demonstrate that the proposed novel technique incurs less pilot ⁃ overhead compared to the well⁃known intrinsic approximation methods (IAM). In addition, it also has a better PAPR, BER and MSE performance. In“Non ⁃ Orthogonal Multiple Access Schemes for 5G,” YAN, YUAN, LI, and YUAN provide a comprehensive review of six potential multiple access schemes for 5G, including MU⁃ SA, RSMA, SCMA, PDMA, interleaver ⁃ division multiple ac⁃ cess (IDMA) and NOMA. The principles, advantages and dis⁃ advantages of these multiple access schemes are discussed. More importantly, this review offers a comprehensive compari⁃ son of these solutions from the perspective of user overload, re⁃ ceiver type, receiver complexity, performance and grant ⁃ free transmission. In“A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks”by WEI, YUAN, Ng, Elkashlan and DING, the authors use a simple downlink model with two users served by a single⁃carrier to illustrate the basic principles of NOMA and its performance. The related questions and designs for a more general model with an arbi⁃ trary number of users and multiple carriers are discussed. In addition, an overview of existing works on performance analy⁃ sis, resource allocation, and multiple ⁃ input multiple ⁃ output NOMA are summarized and discussed. The key features of NO⁃ MA and its potential research challenges in future networks are raised. In“Unified Framework Towards Flexible Multiple Access Schemes for 5G”, SUN, WANG, HAN and I provide a compre⁃ hensive overview for the multiple access schemes proposed for 5G networks. The authors distinguish three types of multiple access techniques in power, code and interleaver based solu⁃ tions, respectively. The key features of these multiple access techniques are highlighted, and the authors also provide com⁃ parison among these multiple access techniques. Another im⁃ portant contribution of this paper is that a unified framework of the aforementioned multiple access techniques is provided. In“Multiple Access Rateless Network Coding for Machine⁃ to ⁃ Machine Communications” by JIAO, Abbas, LI and ZHANG, the authors propose a novel multiple access rateless network coding scheme for machine⁃to⁃machine (M2M) com⁃ munications. The scheme is capable of increasing transmission efficiency by reducing occupied time slots yet with high decod⁃ ing success rates. In addition, in contrast to existing state⁃of⁃ the⁃art coding schemes, the novel rateless network coding is able to dynamically recode, making it suitable for M2M multi⁃ cast networks with heterogeneous erasure features. In“Multiple Access Technologies for Cellular M2M Commu⁃ nications”, Shirvanimoghaddam and Johnson provide a com⁃ prehensive survey of the multiple access techniques for ma⁃ chine ⁃ to ⁃ machine (M2M) communications in future wireless cellular networks. In particular, the overview highlights the multiple access strategies and explains their limitations when used for M2M communications. The throughput efficiency of different multiple access techniques when used in coordinated and uncoordinated scenarios are illustrated. The authors dem⁃ onstrate that in uncoordinated scenarios, NOMA can support a larger number of devices compared to orthogonal multiple ac⁃ cess techniques. We thank all authors for their valuable contributions and all reviewers for their timely and constructive comments on the submitted papers. We hope the content of this issue is informa⁃ tive and helpful to all readers. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS02 Special Topic Guest Editorial YUAN Jinhong, XIANG Jiying, DING Zhiguo, and YUAN Zhifeng
  • 7. Evaluation of Preamble Based Channel Estimation forEvaluation of Preamble Based Channel Estimation for MIMO⁃FBMC SystemsMIMO⁃FBMC Systems Sohail Taheri1 , Mir Ghoraishi1 , XIAO Pei1 , CAO Aijun2 , and GAO Yonghong2 (1. 5G Innovation Centre, Institute for Communication Systems (ICS), University of Surrey, Guildford, Surrey GU2 7XH, United Kingdom; 2. ZTE Wistron Telecom AB, Kista, Stockholm 164 51, Sweden) Abstract Filter⁃bank multicarrier (FBMC) with offset quadrature amplitude modulation (OQAM) is a candidate waveform for future wireless communications due to its advantages over orthogonal frequency division multiplexing (OFDM) systems. However, because of or⁃ thogonality in real field and the presence of imaginary intrinsic interference, channel estimation in FBMC is not as straightforward as OFDM systems especially in multiple antenna scenarios. In this paper, we propose a channel estimation method which employs intrinsic interference cancellation at the transmitter side. The simulation results show that this method has less pilot overhead, less peak to average power ratio (PAPR), better bit error rate (BER), and better mean square error (MSE) performance compared to the well⁃known intrinsic approximation methods (IAM). channel estimation; filter⁃bank multicarrier (FBMC); multiple⁃input multiple⁃output (MIMO); offset quadrature amplitude modula⁃ tion (OQAM); wireless communication Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 001 http://www.cnki.net/kcms/detail/34.1294.TN.20161014.0955.002.html, published online October 14, 2016 Special Topic This work is supported by ZTE Industry⁃Academia⁃Research Cooperation Funds under Grant No. Surrey⁃Ref⁃9953. 1 Introduction rthogonal frequency division multiplexing (OFDM) has been widely used in communication systems in the last decade. This is because of its immunity to multipath fading and simplicity of channel estimation and data recovery with a low complexity single⁃tap equalization, and also suitability for multiple⁃input multiple⁃output (MIMO) systems [1]. However, it suffers from disadvantages such as sensitivity to carrier frequency offset (CFO), significant out⁃of⁃band radiation, and cyclic prefix over⁃ head. In the presence of CFO, there is loss of orthogonality be⁃ tween subcarriers leading to inter carrier interference (ICI). Moreover, to efficiently use the available spectrum, a waveform with very low spectral leakage is needed. Because of the OFDM shortcomings, filter⁃bank multicarrier (FBMC) modulation combined with offset quadrature ampli⁃ tude modulation (OQAM) has drawn attention in the last de⁃ cade [2], [3]. Regardless of the higher complexity compared to OFDM, FBMC (known as OFDM/OQAM and FBMC/OQAM in the literature) provides significantly reduced out⁃of⁃band emis⁃ sions, robustness against CFO [4], and under certain condi⁃ tions, better spectral efficiency as there is no need to use cy⁃ clic prefix (CP) [5]. These advantages come from well localized prototype filters in time and frequency domain for pulse shap⁃ ing. Accordingly, FBMC can be a promising alternative to con⁃ ventional radio access techniques to improve wireless access capacity. On the other hand, as orthogonality in FBMC systems only holds in the real field, received symbols are contaminated with an imaginary intrinsic interference term coming from the neigh⁃ bouring real symbols. The interference becomes a source of problem in channel estimation and equalization processes, es⁃ pecially in MIMO systems. The pilot symbols used for channel estimation should be protected from interference as the receiv⁃ er has no knowledge about their neighbours to estimate the amount of interference. These protections cause overheads when designing a transmission frame. In a preamble⁃based ap⁃ proach, the preamble should be protected from the subsequent data transmission and the previous frame by inserting null sym⁃ bols, which causes longer preamble and thus more overhead compared to OFDM. This is also true for scattered pilots where the neighbouring data symbols contribute to the interference on the pilots [6]. In this scenario, typically one or two time⁃fre⁃ quency points adjacent to the pilots are used to cancel the in⁃ terference on the pilots [7]-[10]. Interference Approximation Method (IAM) for preamble ⁃ O October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 03
  • 8. Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS04 based channel estimation in single⁃input, single⁃output (SISO) systems was first introduced in [11]. The preamble was named IAM⁃R in the literature, where R denotes real⁃valued pilots. Alternatively, IAM⁃I and IAM⁃C were introduced in [12], [13], where I and C stand for imaginary and complex pilots. Those preamble based channel estimation schemes were extended to FBMC⁃MIMO systems in [14]. In IAM⁃I and IAM⁃C, pilots on each subcarrier interfere with their adjacent subcarriers in a constructive way. That is, these methods use the intrinsic inter⁃ ference to enhance amplitude of the pilots. As a result, better performance of channel estimation is achieved. Despite good performance, IAM methods suffer from increased pilot over⁃ head, i.e., a number of zero symbols are required to protect the pilot symbols from the interference of their adjacent symbols. While the number of pilot symbols is equal to the number of antennas, the total number of symbols in the preamble will be more than twice the number of transmit antennas. This paper proposes a channel estimation method with re⁃ duced preamble overhead compared to the IAM family. The idea was first introduced in [15] for MIMO⁃OFDM. Applying this method to MIMO⁃FBMC with spatial multiplexing needs further consideration to cancel intrinsic interference. By using basic idea of zero forcing from single antenna, this method has modest computation complexity, while it can outperform IAM methods in terms of peak to average power ratio (PAPR), bit er⁃ ror rate (BER), and mean square error (MSE) under perfect syn⁃ chronization conditions and in presence of carrier frequency offset. The rest of this paper is organized as follows: Section 2 re⁃ views the MIMO⁃FBMC systems, the effect of intrinsic interfer⁃ ence, and the conventional channel estimation methods. In Sec⁃ tion 3, the new method for channel estimation is proposed and Section 4 shows the results and comparisons with IAM meth⁃ ods. Finally, conclusions are drawn in Section 5. 2 MIMO⁃FBMC System 2.1 System Model FBMC systems are implemented by a prototype filter g( )t and synthesis and analysis filter⁃banks in transmitter and re⁃ ceiver side respectively. The real and imaginary parts of com⁃ plex symbols are separated in two different branches where they are modulated in FBMC modulators as real symbols. Therefore, at a specific time, each subcarrier in this system car⁃ ries a real⁃valued symbol. Denoting T0 as symbol duration and F0 as subcarrier spacing in OFDM systems, duration and sub⁃ carrier spacing in FBMC are either τ0 = T0 2 , ν0 = F0 or τ0 = T0 , ν0 = F0 2 [16]. For the system model in this paper, the former approach is adopted. That is, subcarrier spacing re⁃ mains the same as OFDM, while symbol duration is reduced by half. Assuming a multiple antenna scenario with P transmit anten⁃ nas, Q receive antennas, and M subcarriers, the baseband sig⁃ nal to be transmitted over the p th branch in general form is expressed as s ( )p ( )t = ∑n = -∞ +∞ ∑m = 0 M - 1 a ( )p m,n gm,n( )t , (1) where a ( )p m,n is the real⁃valued symbol, and gm,n( )t is the shift⁃ ed version of the prototype filter on the m th subcarrier and at n th symbol duration: gm,n( )t = jm + n e j2πmν0t g( )t - nτ0 . (2) The prototype filter g( )t is designed to keep its shifted ver⁃ sions are orthogonal only in the real field [17], i.e., R æ è ç ö ø ÷∫gm,n( )t g* m0,n0 ( )t dt = δm,m0 δn,n0 , (3) where R( ). denotes the real⁃part of a complex number. As a consequence, the outputs of the analysis filter⁃bank have a so⁃ called intrinsic interference term which is pure imaginary. The demodulated signal on the q th receive antenna at a particular subcarrier and symbol point ( )m0,n0 is given by y ( )q m0,n0 =∑p = 1 P h q,p m0,n0 a ( )p m0,n0 + jI ( )q m0,n0 + η ( )q m0,n0 , (4) where h q,p m0,n0 is channel frequency response at ( )m0,n0 be⁃ tween qth receive and pth transmit antenna, η ( )q m0,n0 is the noise component at qth receive antenna, and the interference term I ( )q m0,n0 is formed as jI ( )q m0,n0 =∑p = 1 P ∑ ( )m,n ≠ ( )m0,n0 h q,p m,na ( )p m,n g m0,n0 m,n . (5) In (5), g m0,n0 m,n is expressed as g m0,n0 m,n = ∫gm,n( )t g* m0,n0 ( )t dt . (6) Having the prototype filter g( )t well localized in time and frequency, it can be assumed that the intrinsic interference is mostly due to the first ⁃ order neighbouring points. That is, ( )m,n in (5) can take the values of Ω* as follows [6]: Ω* ={ }( )m0,n0 ± 1 ,( )m0 ± 1,n0 ,( )m0 ± 1,n0 ± 1 , (7) which covers the ( )m0,n0 point first⁃order neighbours. By as⁃ suming constant channel frequency response over ( )m0,n0 and Ω* , we can simplify (5) as jI ( )q m0,n0 =∑p = 1 P h p,q m0,n0 ∑ ( )m,n ∈ Ω* a ( )p m,n g m0,n0 m,n . (8)
  • 9. Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 05 Consequently, (4) can be written as y ( )q m0,n0 =∑p = 1 P h p,q m0,n0 æ è ç ç ç çç ç ö ø ÷ ÷ ÷ ÷÷ ÷       a ( )p m0,n0 + ju ( )p m0,n0 c ( )p m0,n0 + η ( )p m0,n0 , (9) where ju ( )p m0,n0 = ∑ ( )m,n ∈ Ω* a ( )p m,n g m0,n0 m,n . (10) Table 1 shows the number of g m0,n0 m,n coefficients on the first⁃ order neighbours of the point ( )m0,n0 . The weights of interfer⁃ ence, β , γ , and δ, depend on the prototype filter and have been derived in [18]. In this work, the isotropic orthogonal transform algorithm (IOTA) [19] filter is employed. It exploits the symmetrical property of Gaussian function in time and fre⁃ quency. Therefore, the amount of interference out of first⁃order neighbouring points is negligible. The weights of interference for this filter are β = 0.2486 , γ = 0.5755 , and δ = 0.1898 (Table 1) . The MIMO⁃FBMC signal model can be represented as æ è ç ç çç ö ø ÷ ÷ ÷÷ y (1) m0,n0 ⋮ y (Q) m0,n0 = æ è ç ç çç ö ø ÷ ÷ ÷÷ h 1,1 m0,n0 ⋯ h 1,P m0,n0 ⋮ ⋱ ⋮ h Q,1 m0,n0 ⋯ h Q,P m0,n0 æ è ç ç çç ö ø ÷ ÷ ÷÷ c (1) m0,n0 ⋮ c (Q) m0,n0 + æ è ç ç çç ö ø ÷ ÷ ÷÷ η (1) m0,n0 ⋮ η (Q) m0,n0 (11) where c ( )p m0,n0 is defined in (9). To retrieve the transmitted sy⁃ mbols from the system above, it is necessary to have an evalua⁃ tion of the channel coefficients, which are used to detect the linearly combined demodulated complex symbols c ( )p m0,n0 at each receiver branch using zero forcing (ZF), minimum mean square error (MMSE), or maximum likelihood (ML). In c ( )p m0,n0 , the imag⁃ inary parts are intrinsic interference terms. By taking R{}. op⁃ eration, the transmitted symbols a ( )p m0,n0 = R{ }c ( )p m0,n0 are recovered. 2.2 Channel Estimation To obtain the channel information over one frame duration on each receive antenna, we need to know the transmitted pilot symbols. The number of these pilot symbols should be equal to P to form a linear equation system with the least square estima⁃ tion method. For simplicity, let us consider a 2⁃by⁃2 antenna scenario. By allocating two pilot symbols at times n = n0 and n = n1 on each antenna, the equation set of the system on su⁃ bcarrier m is given by æ è ç ç ö ø ÷ ÷ y ( )1 m,n0 y ( )1 m,n1 y ( )2 m,n0 y ( )2 m,n1 = æ è ç ç ö ø ÷ ÷ h 1,1 m,n0 h 1,2 m,n0 h 2,1 m,n0 h 2,2 m,n0 æ è ç ç ö ø ÷ ÷ x ( )1 m,n0 x ( )1 m,n1 x ( )2 m,n0 x ( )2 m,n1 + æ è ç ç ö ø ÷ ÷ η ( )1 m,n0 η ( )1 m,n1 η ( )2 m,n0 η ( )2 m,n1 . (12) In (12), x ( )p m,n are pilot symbols. We have assumed that there is no significant variations in the channel between time slots n0 and n1 . Hence, we can drop the time subscript and express (12) as Ym = HmXm + ηm. (13) Thus, channel coefficients can be calculated by the least square estimation method: Ĥ m = Ym( )XH mXm -1 XH m = Hm + ηm( )XH mXm -1 XH m , (14) or in a special case with the equal number of transmit and re⁃ ceive antenna: Ĥ m = YmX-1 m = Hm + ηmX-1 m . (15) The preamble in the IAM methods is composed of 2P + 1 symbols. That is, the length of the preamble grows linearly with P. The symbols with even time indices are pilots, while other symbols are all zeros to protect pilots from intrinsic inter⁃ ference. Based on the values of pilot symbols, i.e. real, imagi⁃ nary, or complex valued pilots, IAM ⁃ R, IAM ⁃ I and IAM ⁃ C were proposed. In these approaches, the channel coefficients can be obtained using (12). For P=2, pilot symbols in (12) are set as x ( )1 m,n0 = x ( )1 m,n1 = x ( )2 m,n0 = -x ( )2 m,n1 = xm . Hence, they form a system based on (12) as Ym = xmHm( )1 1 1 -1 + ηm = xmHmA2 + ηm, (16) where A2 = A-1 2 is an orthogonal matrix if omitting the co⁃ nstant coefficient of the inverse [14]. Finally, the channel coef⁃ ficients are obtained as follows: Ĥ m = 1 xm YmA2 = Hm + 1 xm ηmA2. (17) The length of the preamble in this method is 2P+1=5 with just two pilot symbols. As a result, this approach suffers from significant pilot overhead which reduces the spectral efficien⁃ cy. Furthermore, the periodic nature of the pilots in these pre⁃ ambles results in high PAPR at the output of the synthesis filter⁃ ▼Table 1. Weights of interference on the first⁃order neighbours m0 - 1 m0 m0 + 1 n0 - 1 ( )-1 m0 δ -( )-1 m0 γ ( )-1 m0 δ n0 -β 1 β n0 + 1 ( )-1 m0 δ ( )-1 m0 γ ( )-1 m0 δ
  • 10. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS06 bank [14]. 3 Proposed Method In order to reduce the preamble overhead and accordingly increase the spectral efficiency, a novel channel estimation ap⁃ proach with modest computation complexity is proposed. Since there is no need to have an estimation of the channel on each subcarrier, we can reduce the number of pilot symbols to one. In this way, each subcarrier is allocated to only one branch to transmit pilot. That is, while a branch is transmitting pilot on a subcarrier, the other branches remain silent. Therefore, the channel parameters between the receive branch and the pilot transmitting branch on that specific subcarrier can be ob⁃ tained. This method enables the increase of transmit branches with a constant length of the preamble. To elaborate the system more precisely, we assume a 2x2 MIMO system where preambles for branches 1 and 2 are shown in Fig. 1. It can be seen that the first and third symbols are all zero to protect the preamble from intrinsic interference from data section and previous frame. In the middle symbol for branch 1, complex pilots are placed on odd subcarriers, while the other subcarriers carry zeros. On branch 2, orthogonal pi⁃ lots to branch 1 are sent, i.e., even subcarriers carry complex pilots and the rest are zero valued. On a particular subcarrier m = m0 , the system equations is written as follows: æ è ç ç ö ø ÷ ÷ y ( )1 m0 y ( )2 m0 = æ è ç ç ö ø ÷ ÷ h 1,1 m0 h 1,2 m0 h 2,1 m0 h 2,2 m0 æ è ç ç ö ø ÷ ÷ x ( )1 m0 x ( )2 m0 + æ è ç ç ö ø ÷ ÷ η ( )1 m0 η ( )2 m0 . (18) On odd subcarriers m0 = 2k + 1, we have x ( )1 m0 = Xm0 , while x ( )2 m0 = 0 . Then, the channel coefficients h 1,1 m0 and h 2,1 m0 are o⁃ btained as h 1,1 m0 = y ( )1 m0 Xm0 |​ x ( )2 m0 = 0 h 2,1 m0 = y ( )2 m0 Xm0 |​ x ( )2 m0 = 0. (19) Likewise, on even subcarriers the channel coefficients of h 1,2 m0 and h 2,2 m0 are given by h 1,2 m0 = y ( )1 m0 Xm0 |​ x ( )1 m0 = 0 h 2,2 m0 = y ( )2 m0 Xm0 |​ x ( )1 m0 = 0. (20) Hence, we have calculated the channel parameters between each pair of antennas on alternative subcarriers. Channel Coef⁃ ficients on the rest of subcarriers can be obtained by interpola⁃ tion. Due to short distance between pilots in this system, linear interpolation provides enough accuracy with the advantage of low complexity. The technique works perfectly for MIMO ⁃ OFDM systems [15]. When applying this method to MIMO⁃FBMC, intrinsic in⁃ terference degrades the channel estimation performance, i.e., transmitted pilots from one branch interfere with the received pilots on other branch. Consequently, the conditions in (19) and (20) no longer hold. To tackle this problem, we propose a precoding approach in which the interference is calculated at the transmitter side. Then, the zero points in pilot symbols are replaced by Im,n , so that there are no interference on the corre⁃ sponding points at the receiver side. That is, the pilots are re⁃ ceived without any interference from other branches. Fig. 2 shows the precoded pilots. The value of cancelling in⁃ terference on subcarrier m is calculated by using (10) as Im,n = - ∑ ( )m' ,n' ∈ Ω* a ( )p m,n g m' ,n' m,n . (21) Moreover, the adjacent points of the pilot Xm are filled with pre⁃calculated values to maximize the received signal energy, thereby to enhance the estimation accuracy [18]. Defining Xm = XR m + jXI m , These values would be X' m = - jXI m X″ m = - XR m. (22) Consequently, the amplitude of the real and imaginary parts Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong ▲Figure 1. The basic preamble for two antennas. ▲Figure 2. The preambles for two antennas after interference cancellation of the first and third time symbols that helps the pilots become stronger. 0 0 0 0 0 0 Xm - 3 0 Xm - 1 0 Xm + 1 0 Branch 1 0 0 0 0 0 0 0 0 0 0 0 0 0 Xm - 2 0 Xm 0 X_(m + 2) Branch 2 0 0 0 0 0 0 X' m - 3 0 X' m - 1 0 X' m + 1 0 Xm - 3 Im - 2,1 Xm - 1 Im,1 Xm + 1 Im + 2,1 Branch 1 X″ m - 3 0 X″ m - 1 0 X″ m + 1 0 0 X' m - 2 0 X' m 0 X' m + 2 Im - 3,1 Xm - 2 Im - 1,1 Xm Im + 1,1 Xm + 2 Branch 2 0 X″ m - 2 0 X″ m 0 X″ m + 2
  • 11. of the received pilots becomes ||X̂ R m = ||XR m + γ ||X' m + γ ||XI m ||X̂ I m = ||XI m + γ ||X″ m + γ ||XR m , (23) where γ is the interference weight shown in Table 1. The com⁃ plete design of the preambles is displayed in Fig. 2. The pilots can take arbitrary values. In this work, the maximum ampli⁃ tude of the used QAM modulation is used so that XR m = XI m . In order to avoid high PAPR, the sign of the pilots should be changed alternatively after a number of repetitions. The final value of the received pilots in (23) with XR m = XI m is X̂ m = ( )1 + 2γ Xm. (24) The extension to P⁃branch MIMO system is straightforward. In this case, one subcarrier of every P subcarriers carries a pi⁃ lot (non⁃zero), while each branch’s pilot symbol is orthogonal to other branches. The more transmit branches, the more dis⁃ tance between pilot subcarriers. Consequently, for larger num⁃ ber of branches, the quality of channel estimation degrades. 4 Simulation Results In this section, different preamble⁃based channel estimators for a 2x2 MIMO⁃FBMC system are simulated and compared. The simulations are performed using 7⁃tap EPA⁃5Hz and 9⁃tap ETU⁃70Hz channel models with low spatial correlations. Per⁃ fect synchronization is assumed for BER and MSE comparison, i.e., there is no timing or frequency offset errors. In order to de⁃ tect symbols, MMSE equalizer is used. Table 2 summarizes the simulation parameters. The results are compared with IAM⁃R and IAM⁃C methods introduced in [14]. For fair comparison, the transmission power is kept equal for all methods. In this system, Eb N0 is defined by Eb N0 = Q SNR α × log2( )M , (25) where M = 16 is the modulation order, SNR is signal⁃to⁃noise ratio, and α = Ns - Np Ns with the frame length Ns = 14 and the preamble length Np . The length of preamble Np in the pr⁃ oposed method is three symbols resulting in 40% overhead re⁃ duction compared to IAMs. As a result, a performance gain is expected due to shorter preamble. The extra symbols generated by the synthesis filter⁃banks can be dropped before transmis⁃ sion, but one of them with the most power should be kept to avoid filtering errors after demodulation, i.e., Ns + 1 symbols are transmitted. To consider this extra symbol, α can be changed to α = Ns - Np Ns + 1. 4.1 PAPR Comparison Fig. 3 shows the comparison between the proposed method and IAMs in terms of PAPR. The plots show the squared mag⁃ nitude of the preambles at the output of the synthesis filter ⁃ bank on branch 1. Evidently, from the point of practical imple⁃ mentations, the proposed method is preferable. Whereas in the others, the signal level should be kept very low to avoid A/D saturations. The PAPR levels for the pilot symbols are com⁃ pared in Table 3 for the three methods. 4.2 Channel Estimation Performance Comparison Fig. 4 shows the MSE comparison of the channel estimation methods. To calculate MSE, the channel tap on the second symbol in frame is considered as reference and it is assumed constant during the symbol duration. Then, the MSE is calcu⁃ Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 07 ▼Table 2. Simulation parameters EPA: Extended Pedestrian A model ETU: Extended Typical Urban model FFT: fast Fourier transform QAM: quadrature amplitude modulation Modulation type FFT size Used subcarriers Sampling frequency Symbols per frame Channel M⁃QAM, M =16 256 144 3.84 MHz 14 EPA 5 Hz, ETU 70 Hz ▼Table 3. PAPR comparison for the three methods IAM⁃C: Interference Approximation Method⁃complex pilots IAM⁃R: Interference Approximation Method⁃real valued pilots PAPR: peak to average power ratio PAPR IAM⁃C 17.5 IAM⁃R 9.3 Proposed 7.2 ▲Figure 3. Squared magnitude of the preambles on output of the branch 1. 150010005000 15 10 5 0 The proposed preamble 150010005000 15 10 5 0 IAM⁃R preamble 150010005000 15 10 5 0 IAM⁃C preamble Time (Samples) Transmitsignalcomparison IAM⁃C: Interference Approximation Method⁃complex pilots IAM⁃R: Interference Approximation Method⁃real valued pilots
  • 12. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS08 lated using the estimated channel Ĥ as Eæ è ö ø ( )H - Ĥ H ( )H - Ĥ . It can be seen that the proposed preamble outperforms IAM⁃R and has approximately the same performance as IAM⁃C in both channel models. In the EPA⁃5Hz scenario, the proposed meth⁃ od gradually reaches an error floor. This is due to domination of errors from ISI and interference cancellation residual. How⁃ ever, the performance is still as good as IAM⁃C. In the ETU⁃ 70Hz scenario, because of rapid variation of the channel taps, the assumption of constant channel over Ω* in (8) is invalid. Consequently, the performance of all the methods degrades and reaches an error floor in higher SNRs. This is a general problem in channel estimation for FBMC systems where the re⁃ ceiver should necessarily have an estimation of intrinsic inter⁃ ferences for channel estimation. However, the degradation on IAMs is more significant as the channel is estimated using two symbols with one zero symbol in between. Therefore, as the channel is not constant over the two pilot symbols, degradation is higher than the proposed method with only one symbol for channel estimation. The Cramer⁃Rao lower bound (CRLB) for the proposed method, derived in Appendix A has also been plotted in the figure for benchmark comparison. It can be seen that the proposed scheme achieves closest performance to the theoretical lower bound in comparison to the other schemes. Fig. 5 shows the MSE comparisons in terms of residual CFO. It is assumed that the CFO has been estimated and com⁃ pensated before channel estimation. As the estimated CFO is not perfect, the residual CFO affects the quality of channel esti⁃ mation. Therefore, the methods are compared in presence of re⁃ sidual CFO in the two channel scenarios without added white Gaussian noise. When the CFO is zero, the MSEs show the er⁃ ror floor of the methods in Fig. 4 at very high SNRs. It can be seen that in EPA channel, the error floor of the proposed meth⁃ od is higher than IAM⁃C, while it has the best performance un⁃ der ETU channel. This is also true for the other values of CFO, where the degradation of MSE in the proposed method is lower than the other two in both channels. 4.3 Bit Error Rate Performance Comparison The BER performance comparison with respect to Eb N0 is i⁃ llustrated in Fig. 6. Evidently, the proposed method performs Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong CRLB: Cramer⁃Rao lower bound EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method SNR: signal⁃to⁃noise ratio CFO: carrier frequency offset EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method EPA: Extended Pedestrian A model ETU: Extended Typical Urban model IAM: Interference Approximation Method ▲Figure 4. MSE performance of the channel estimation methods. ▲Figure 5. MSE performance of the channel estimation methods in presence of residual CFO. ▲Figure 6. BER performance of the channel estimation methods. 30 100 SNR (dB) Meansquareerror 2520151050 10-1 10-2 10-3 10-4 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz Proposed⁃CRLB 150 10-1 Residual CFO (Hz) Meansquareerror 10-2 10-3 100500-50-100-150 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz 22 100 Eb /NO (dB) Biterrorrate 10-1 10-2 10-3 2018161412108642 Proposed⁃EPA 5 Hz IAMC⁃EPA 5 Hz IAMR⁃EPA 5 Hz Proposed⁃ETU 70 Hz IAMC⁃ETU 70 Hz IAMR⁃ETU 70 Hz
  • 13. better compared to the others in low mobility EPA⁃5Hz scenar⁃ io. In the high mobility ETU⁃70Hz channel, the performance deteriorates as the channel varies significantly during the frame time. Consequently, the preamble⁃based channel estima⁃ tion is not a proper choice for high mobility applications and there is an error floor for all the curves showing around six per⁃ cent bit error rate. 5 Conclusions In this paper, we proposed a novel channel estimation algo⁃ rithm with much reduced pilot overhead compared to the exist⁃ ing IAM based approaches. Our results show that the proposed method has better PAPR property. The system performance un⁃ der low mobility and high mobility channels, as well as in the presence of CFO, has been simulated and compared. Accord⁃ ing to the results, the proposed method achieves comparable channel estimation performance to the IAM methods, and bet⁃ ter BER performance due to shorter preamble. Appendix A Cramer⁃Rao Lower Bound for the Proposed Channel Es⁃ timation In this section, a lower bound for the proposed channel esti⁃ mator is derived. We simplify the system using equations (13), (18), (19), and (20) as Y = XH + η, (26) where Y =[ ]y1 y2 is the received signal vector, η =[ ]η1 η2 is the noise vector, H =[ ]h1 h2 is the channel vector to be esti⁃ mated, X is the pilot symbol. The subcarrier index has also been dropped for simplicity. The CRLB is a bound on the smallest covariance matrix that can be achieved by an unbiased estimator, Ĥ , of a parameter vector H as J-1 ≤ CĤ = E{ }( )H - Ĥ ( )H - Ĥ * ; J = E ì í î ï ï ü ý þ ï ï æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H * , (27) where ( )∙ * denotes conjuagate transpose operation, J is the Fisher information matrix and ln p( )Y; H is the log⁃likelihood function of the observed vector Y . The vector Y is a complex Gaussian random vector, i.e., Y ∼ CN( )XH,N0I with likeli⁃ hood function and log⁃likelihood function as where K is a constant. Taking the complex gradient [20] of ln p( )Y; H with respect to H yields ∂ ln p( )Y; H ∂H = - 1 N0 [ ]X* XH - X* Y * . (29) The above equality holds since ∂Y2 ∂H = 0; ∂H* X* Y ∂H = 0; ∂Y* XH ∂H = ( )X* Y * ; ∂H* X* XH ∂H = ( )X* XH * . (30) Thus we can derive, ∂ ln p( )Y; H ∂H* = æ è çç ö ø ÷÷ ∂ ln p( )Y; H ∂H * = X* Y - X* XH N0 = X* X N0 { }( )X* X -1 X* Y - H = J( )H [ ]Ĥ - H . (31) This proves that the minimum variance unbiased estimator of H is Ĥ = ( )X* X -1 X* Y = Y X . (32) It is efficient in that it attains the CRLB. The Fisher informa⁃ tion matrix J( )H and covariance matrix CĤ of this unbiased estimator are J( )H = E é ë êê ù û úú X* XI2 N0 = E[ ]X* X I2 N0 = Ex N0 I2 CĤ = J-1 ( )H = N0 Ex I2. (33) In (33), Ex is the pilot energy. The CRLB for each diagonal element of J-1 ( )H is var( )ĥ 1 = var( )ĥ 2 = diag[ ]CĤ i = N0 Ex . (34) As the pilots in this system are amplified exploiting intrinsic interference by the factor of 1 + 2γ , Ex should be replaced by E' x = ( )1 + 2γ 2 Ex . Assuming Ex N0 is approximately equal to SNR and considering (25), (34) becomes var( )ĥ 1 = var( )ĥ 2 = N0 Ex 1 ( )1 + 2γ 2 . (35) Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 09 ( )Y; H = 1 ( )πN0 2 exp é ë êê ù û úú- ( )Y - XH * ( )Y - XH N0 = 1 ( )πN0 2 exp é ë ê ù û ú- Y2 - H* X* Y - Y* XH + H* X* XH N0 ; ln p( )Y; H = K - Y2 - H* X* Y - Y* XH + H* X* XH N0 , (28) References [1] A. Sahin, I. Guvenc, and H. Arslan,“A survey on multicarrier communications: Prototype filters, lattice structures, and implementation aspects,”IEEE Commu⁃ nications Surveys Tutorials, vol. 16, no. 3, pp. 1312-1338, Mar. 2014.
  • 14. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS10 Special Topic Evaluation of Preamble Based Channel Estimation for MIMO⁃FBMC Systems Sohail Taheri, Mir Ghoraishi, XIAO Pei, CAO Aijun, and GAO Yonghong [2] B. Farhang ⁃ Boroujeny,“OFDM versus filter bank multicarrier,”IEEE Signal Processing Magazine, vol. 28, no. 3, pp. 92-112, May 2011. [3] F. Schaich and T. Wild,“Waveform contenders for 5G—OFDM vs. FBMC vs. UFMC,”in 6th International Symposium on Communications, Control and Sig⁃ nal Processing, Athens, Greece, 2014, pp. 457 - 460. doi: 10.1109/ISCC⁃ SP.2014.6877912. [4] Q. Bai and J. Nossek,“On the effects of carrier frequency offset on cyclic prefix based OFDM and filter bank based multicarrier systems,”in IEEE Eleventh In⁃ ternational Workshop on Signal Processing Advances in Wireless Communica⁃ tions, Marrakech, Morocco, Jun. 2010, pp. 1-5. doi: 10.1109/SPAWC.2010. 5670999. [5] M. Sriyananda and N. Rajatheva,“Analysis of self interference in a basic FBMC system,”in IEEE 78th Vehicular Technology Conference, Las Vegas, USA, Sept. 2013, pp. 1-5. doi: 10.1109/VTCFall.2013.6692102. [6] J. Javaudin and Y. Jiang,“Channel estimation in MIMO OFDM/OQAM,”in IEEE 9th Workshop on Signal Processing Advances in Wireless Communications, Recife, Brazil, Jul. 2008, pp. 266-270. doi: 10.1109/SPAWC.2008.4641611. [7] J. Javaudin, D. Lacroix, and A. Rouxel,“Pilot ⁃ aided channel estimation for OFDM/OQAM,”in 57th IEEE Semiannual Vehicular Technology Conference, Je⁃ ju, South Korea, Apr. 2003, pp. 1581-1585. doi: 10.1109/VETECS.2003. 1207088. [8] C. Lele, R. Legouable, and P. Siohan,“Channel estimation with scattered pilots in OFDM/OQAM,”in IEEE 9th Workshop on Signal Processing Advances in Wireless Communications, Recife, Brazil, Jul. 2008, pp. 286-290. doi: 10.1109/ SPAWC.2008.4641615. [9] Z. Zhao, N. Vucic, and M. Schellmann,“A simplified scattered pilot for FBMC/ OQAM in highly frequency selective channels,”in 11th international symposium on Wireless communications systems, Barcelona, Spain, Oct. 2014, pp. 819-823. doi: 10.1109/ISWCS.2014.6933466. [10] J. Bazzi, P. Weitkemper, and K. Kusume,“Power efficient scattered pilot chan⁃ nel estimation for FBMC/OQAM,”in 10th International ITG Conference on Sys⁃ tems, Communications and Coding, Hamburg, Germany, Feb. 2015, pp. 1-6. [11] C. Lélé, J. Javaudin, R. Legouable, A. Skrzypczak, and P. Siohan,“Channel es⁃ timation methods for preamble ⁃ based OFDM/OQAM modulations,”Transac⁃ tions on Emerging Telecommunications Technologies, pp. 741-750, Sept. 2008. doi: 10.1002/ett.1332. [12] C. Lélé, P. Siohan, and R. Legouable,“2 dB better than CP⁃OFDM with OFDM/ OQAM for preamble⁃based channel estimation,”in IEEE International Confer⁃ ence on Communications, Beijing, China, 2008, pp. 1302-1306. doi: 10.1109/ ICC.2008.253. [13] J. Du and S. Signell,“Novel preamble⁃based channel estimation for OFDM/ OQAM systems,”in IEEE International Conference on Communications, Dres⁃ den, Germany, 2009, pp. 1-6. doi: 10.1109/ICC.2009.5199226. [14] E. Kofidis and D. Katselis,“Preamble ⁃ based channel estimation in MIMO ⁃ OFDM/OQAM systems,”in IEEE International Conference on Signal and Im⁃ age Processing Applications, Kuala Lumpur, Malaysia, 2011, pp. 579-584. doi: 10.1109/ICSIPA.2011.6144161. [15] J. Siew, R. Piechocki, A. Nix, and S. Armour. (2002).“A channel estimation method for MIMO⁃OFDM systems,”London Communicaitons Symposium (LCS) [Online]. Available: http://www.ee.ucl.ac.uk/lcs/previous/LCS2002/LCS087.pdf [16] J. Du, P. Xiao, J. Wu, and Q. Chen,“Design of isotropic orthogonal transform algorithm⁃based multicarrier systems with blind channel estimation,”IET com⁃ munications, vol. 6, no. 16, pp. 2695- 2704, Nov. 2012. doi: 10.1049/iet ⁃ com.2012.0029. [17] P. Siohan, C. Siclet, and N. Lacaille,“Analysis and design of OFDM/OQAM systems based on filterbank theory,”IEEE Transactions on Signal Processing, vol. 50, no. 5, pp. 1170-1183, May 2002. [18] E. Kofidis and D. Katselis,“Improved interference approximation method for preamble⁃based channel estimation in FBMC/OQAM,”in 19th European sig⁃ nal processing conference (EUSIPCO⁃2011), Barcelona, Spain, 2011. pp. 1603- 1607. [19] J. Du and S. Signell,“Time frequency localization of pulse shaping filters in OFD/OQAM systems,”in 6th International Conference on Information, Commu⁃ nications Signal Processing, Singapore, 2007, pp. 1-5. [20] S. Kay, Fundamentals of Statistical Signal Processing. Upper Saddle River, USA: Prentice Hall, 1998. Manuscript received: 2016⁃04⁃04 Sohail Taheri (s.taheri@surrey.ac.uk) received his BS degree in electronic engineer⁃ ing and MSc degree in digital electronics from Amirkabir University of Technology, Iran in 2010 and 2012 respectively. He is currently working towards his PhD degree from the Institute for Communication Systems (ICS), University of Surrey, United Kingdom. His current research interests include signal processing for wireless com⁃ munications, waveform design for 5G air interface and physical layer for 5G net⁃ works. Mir Ghoraishi (m.ghoraishi@surrey.ac.uk) is a senior research fellow in the Insti⁃ tute for Communication Systems (ICS), University of Surrey. He joined the Institute in 2012 and is currently leading 5GIC testbed and proof⁃of⁃concept projects. This work area includes several implementation and proof⁃of⁃concept projects, e.g. 5G air⁃ interface proof⁃of⁃concept, distributed massive MIMO implementation, wireless in⁃ band full⁃duplex, millimeter wave hybrid beamforming system, and millimeter wave wireless channel analysis and modelling. He was involved in EU FP7 DUPLO proj⁃ ect as work package leader. He has previously worked in Tokyo Institute of Technol⁃ ogy as assistant professor and senior researcher from 2004 to 2012, after getting his PhD from the same institute. In Tokyo Tech he was involved in several national and small scale projects in planning, performing, implementation, analysis and model⁃ ling different aspect of wireless systems in physical layer, propagation channel and signal processing. He has co⁃authored 100 publications including refereed journals, conference proceedings and three book chapters. XIAO Pei (p.xiao@surrey.ac.uk) received the BEng, MSc and PhD degrees from Huazhong University of Science & Technology, Tampere University of Technology, Chalmers University of Technology, respectively. Prior to joining the University of Surrey in 2011, he worked as a research fellow at Queen’s University Belfast and had held positions at Nokia Networks in Finland. He is a Reader at University of Surrey and also the technical manager of 5G Innovation Centre (5GIC), leading and coordinating research activities in all the work areas in 5GIC. Dr Xiao’s research in⁃ terests and expertise span a wide range of areas in communications theory and sig⁃ nal processing for wireless communications. He has published 160 papers in refer⁃ eed journals and international conferences, and has been awarded research funding from various sources including Royal Society, Royal Academy of Engineering, EU FP7, Engineering and Physical Sciences Research Council as well as industry. CAO Aijun (cao.aijun@zte.com.cn) is a principal architect in ZTE R&D Center, Sweden (ZTE Wistron Telecom AB). He has over 17 years of experience in wireless communications research and development from baseband processing to network ar⁃ chitecture, including design and optimization of commercial UMTS/LTE base⁃sta⁃ tion and handset products, HetNet and small cell enhancement, etc. He has also been involved in standardization works and contributed to several 3GPP technical reports. He is also active in academic and industrial workshops and conferences re⁃ lated to the future wireless networks being as panelists or (co⁃)authors of published papers in refereed journals and international conferences. In addition, he holds more than 50 granted or pending patents. His current focus is 5G technologies relat⁃ ed to the new energy⁃efficient unified air⁃interface and network architecture, e.g., new waveform design, non⁃orthogonal multiple access schemes, random access chal⁃ lenges and innovative signaling architecture for 5G networks. GAO Yonghong (gao.yonghong@zte.com.cn) received his BEng degree in electronic engineering from Tsinghua University, China in 1989, and PhD degree in electronic systems from Royal Institute of Technology, Sweden in 2001. In 1996, he was a visit⁃ ing scientist at Royal Institute of Technology and Ericsson Sweden. In 1999, he joined Ericsson Sweden to develop 3G base stations, baseband algorithms, and base⁃ band ASICs. He joined ZTE European Research Institute (ZTE Wistron Telecom AB, Sweden) in 2002 and has been the CTO of ZTE European Research Institute till now, leading and participating the development of 3G/4G commercial base sta⁃ tions, baseband/RRM algorithms, and baseband ASICs, 3GPP small cell enhance⁃ ment, and from 3 years ago focusing on 5G pre⁃study, 5G standardization, and 5G re⁃ search projects in Europe. He has filed 40+ patents as a main author or co⁃author. His research interests include mobile communication standards/systems, and solu⁃ tions and algorithms for commercial wireless products. BiographiesBiographies
  • 15. Non⁃Orthogonal Multiple Access Schemes forNon⁃Orthogonal Multiple Access Schemes for 55GG YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei (ZTE Corporation, Shengzhen 518057, China) Abstract Multiple access scheme is one of the key techniques in wireless communication systems. Each generation of wireless communica⁃ tion is featured by a new multiple access scheme from 1G to 4G. In this article we review several non⁃orthogonal multiple access schemes for 5G. Their principles, advantages and disadvantages are discussed, and followed by a comprehensive comparison of these solutions from the perspective of user overload, receiver type, receiver complexity and so on. We also discuss the applica⁃ tion challenges of non⁃orthogonal multiple access schemes in 5G. 5G; non⁃orthogonal multiple access; mMTC Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 002 http://www.cnki.net/kcms/detail/34.1294.TN.20161008.1038.002.html, published online October 8, 2016 Special Topic 1 Introduction ultiple access scheme is the key technique of wireless communications. In 3rd generation (3G) code division multiple access is applied. In 4G orthogonal frequency division multiplex⁃ ing access (OFDMA) is employed. In the coming 5G, non⁃or⁃ thogonal multiple access schemes are hot topics because they can achieve high system capacity. Moreover, massive machine type communication (mMTC) is one of the key scenarios for 5G in which massive connection is required. In this paper, we mainly focus on the non⁃orthogonal multiple access schemes supporting mMTC which has the rapidest growing speed and the urgent deploy demand. Several non ⁃ orthogonal multiple access schemes are pro⁃ posed for 5G, which include multi⁃user shared multiple access (MUSA) [1]-[4], resource spread multiple access (RSMA) [5], sparse code multiple access (SCMA) [6]-[8], pattern division multiple access (PDMA) [9]-[11], interleaver⁃division multiple access (IDMA) [12], [13], and non⁃orthogonal multiple access (NOMA) by power domain [14]. In this paper, the principles, merits and demerits of these schemes are discussed to let read⁃ ers have a full overview on that. 2 Features of 5G 5G has three main technical features, including enhanced mobile broadband (eMBB), mMTC and ultra reliable and low latency communication (URLLC). The eMBB is the evolution of MBB targeting for high data rate and can support high mobil⁃ ity The mMTC is characterized by massive connection with low cost terminals. High reliability and ultra ⁃ low latency are the goals of URLLC. With the development of Internet of things, a large number of terminals will have access to the network. Therefore, mMTC needs to support one million of connections per square kilome⁃ ter. The mMTC, which has the fastest growing speed and the most urgent deployment demand, will create new chances in 5G. The non⁃orthogonal multiple access should support at least mMTC where high user overload is the key requirement. In LTE there are several interactive processes between base station and terminal before the data is transmitted from termi⁃ nal to the base station. This makes sense for long time and con⁃ tinuous data transmission because signaling overhead is small by averaging over a long time. In mMTC each terminal only transmits small data and massive terminals would sporadically transmit their data to the base station. When the same access procedure like in LTE ⁃ A is applied, the signaling overhead will be comparably large and the access efficiency is very low, thus grant⁃free for mMTC is needed in which multiple termi⁃ nals can send their data on the same resource block without multi⁃step negotiations with base station. 3 Non⁃Orthogonal Multiple Access Schemes for 5G Several non⁃orthogonal multiple access schemes have been proposed for 5G. Based on their properties, they can be catego⁃ rized to different types. Most non⁃orthogonal multiple access schemes use spreading codes. When such schemes have other M October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 11
  • 16. Special Topic Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei predominant properties, such as SCMA and PDMA use code matrix to illustrate how multiple users share the same resource block, and IDMA uses interleaver for user separation, we cate⁃ gorize them as other kind of schemes. In the following joint de⁃ tection denotes message passing algorithm (MPA) based schemes. 3.1 Non􀆼Orthogonal Multiple Access Schemes Based on Spreading Sequences 3.1.1 MUSA MUSA is a non⁃orthogonal multiple access scheme operat⁃ ing in code domain and power domain. Spreading code with short length is applied in MUSA to support a large number of users that share the same resource block. When the number of users is large and the length of the spreading code is small, it is difficult to design large number of spreading code with low correlation when binary element of the spreading code is as⁃ sumed. For binary spreading code the element of the spreading code belongs to the set {1, ⁃1}. Only two values are employed in the spreading code. To overcome this drawback, non⁃binary and complex⁃value spreading code is proposed in MUSA. Ei⁃ ther the real or the image element of the non⁃binary spreading code belongs to the set {1, 0, ⁃1}, there are nine values for se⁃ lection. This provides much more flexibility of spreading code design. Because the real and image elements of the spreading code are 1, 0 or ⁃1, the multiplication operation can be imple⁃ mented by addition operation which will reduce the implemen⁃ tation complexity. Fig. 1 shows the basic features of MUSA, where multiple users could transmit data on the same resourc⁃ es by using randomly selected non⁃orthogonal complex spread⁃ ing codes with short length (e.g. 4). In this example 12 users share 4 resource blocks, and the user overload is 300%. MU⁃ SA is always modeled by multiple spreading codes superposed on the same resource block. It can also be modeled by a code matrix. The code matrix of MUSA with 300% overload is given by In 5G, mMTC is one important application scenario. In this scenario MUSA is preferred since grant⁃free transmission can be readily supported. A device terminal autonomously access⁃ es the communication system without base station (BS) sched⁃ uling. Blind detection is applied at BS for MUSA in which ac⁃ tive user, user spreading code and user channel would not be known before hand. Because the spreading code length is rela⁃ tive short and its elements have limited values, BS can gener⁃ ate numerous local spreading codes with low correlation. By us⁃ ing these local spreading codes and the received signal, we can closely approximate the optimal performance of MMSE estima⁃ tor. Then the user signal with the highest signal⁃to⁃interference ⁃plus⁃noise ratio (SINR) can be detected and decoded. After that user’s signal is successfully decoded, it can be employed for channel estimation. After interference cancellation, the us⁃ er signal with the second highest SINR is detected and decod⁃ ed. During this process no pilots or preamble are needed for channel estimation, which facilitates MUSA application in mMTC because most other schemes rely on additional over⁃ head for channel estimation. The blind detection for MUSA is verified over flat fading channel and multi⁃path fading channel [3], [15]. The main advantages of MUSA are reflected by high over⁃ loading factor, robust blind detection and true sense of grant⁃ free transmission. Due to frequency ⁃ diversity gain achieved, 700% user overload can be achieved by MUSA over multi⁃path fading channel [15]. User detection can be carried out without the knowledge of the spreading code. User transmitted signal can be applied for enhanced channel estimation once it has been correctly decoded. Users can transmit their signals ac⁃ cording to their demand. The possibility of collision due to the same spreading code applied is small since large number of the spreading codes can be accommodated. Successive interference cancelation (SIC)⁃based receiver is applied for MUSA. It works well when there is SINR difference among the received signals. However, when the difference is small, there would be certain performance loss due to error propagation. While there is inherent SINR different in mMTC due to free power control, it is not a so serious problem for the signal detection of MUSA. The SINR difference is small, so it can be solved by using more advanced receiver, such as joint detection and decoding scheme. 3.1.2 RSMA In RSMA (Fig. 2), a group of users’signals are superposed on the same resource blocks, and each user’s signal is spread over the entire frequency/time resource blocks. Different users’ October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS12 G = é ë ê êê ê ù û ú úú ú 1 + i 1 + i 1 + i 1 1 - i 1 + i i -i -1 + i -1 + i -1 i i -i 1 1 + i -i -i 1 + i 1 - i -1 - i -1 + i 1 i 1 -1 1 + i -1 - i -1 1 1 -1 1 + i -i 0 1 1 -1 + i 0 0 1 - i 1 0 0 0 0 0 1 + i SIC: successive interference cancelation ▲Figure 1. An example of MUSA with 300% user overload [4]. Elements of complex spreading codeR-1 0 1 -1 I Complex spreading code set ··· Each user randomly picks one code for spreading Codeword⁃level SIC receiver C1 C2 C12 S1 + S2 + +··· S12 = 1
  • 17. Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei Special Topic signals within the resource blocks may be not orthogonal. Low code rate channel codes are employed to achieve large coding gain. Relative long spreading codes with good correlation prop⁃ erty are applied to reduce the multi⁃user interference. Scram⁃ blers can be employed with the same purpose as the spreading codes. Interleaver is optional for RSMA according to the sys⁃ tem requirements. Depending on the application scenarios, it includes single carrier RSMA and multi⁃carrier RSMA. For the former it is op⁃ timized for battery power consumption and coverage extension for small data transactions by utilizing single carrier wave⁃ forms, very low peak⁃to⁃average⁃power⁃ratio (PAPR) modula⁃ tions. It allows grant ⁃ less transmission and potentially allow asynchronous access. While for the latter it is optimized for low latency access for radio resource connection (RRC) con⁃ nected users (i.e., timing with eNB already acquired) and al⁃ lows for grant⁃less transmission. The advantage of RSMA is that it supports asynchronous and grant ⁃ less transmission, so the signaling overhead is re⁃ duced. The disadvantage is that its user overload is limited when rake receiver is applied. By using advanced receiver, such as SIC based receiver, the overload can be enhanced. 3.2 Non􀆼Orthogonal Multiple Schemes Based on Structured Coding Matrix 3.2.1 SCMA Sparse codebook is applied at SCMA to reduce the detection complexity. At the same time joint detection is employed for SCMA to achieve excellent performance. The codewords are composed of multi⁃dimensional com⁃ plex symbols, and the codewords in the same codebook have the same sparse pattern. Sparse codeword mapping utilizes low density spreading and could be referred to as sparse spreading. At the receiver, iterative multi⁃user detection based on MPA is used. Fig. 3 shows an exam⁃ ple of SCMA, where the coded bits of a data stream are directly mapped to a codeword with sparse non⁃zero ele⁃ ments from a codebook. With 6 sparse codewords transmitted over 4 orthogonal resources, the user overload is 150% . The coding matrix of Fig. 3 is given by G = é ë ê êê ê ù û ú úú ú 1 1 0 0 1 0 1 0 1 0 0 1 0 1 1 0 0 1 0 1 0 0 1 1 To reduce the multi⁃user interference and the de⁃ tection complexity, sparse signature sequence is ap⁃ plied in SCMA for spreading. User signal is modu⁃ lated by a codebook in which multidimensional modulation maps of the input coded bits to the points in the multiple complex dimensions [6]. By such operation shaping gain is achieved, which is claimed as one major property of SCMA. The main disadvantage of SCMA is its high detection and de⁃ coding complexity even sparse signature sequence is applied. The detection and decoding complexity is even higher when large size constellation and a large number of users are em⁃ ployed. And additional pilots or preambles are needed for multi ⁃user channel estimation, which may reduce system spectral ef⁃ ficiency. Because the size of the codebook is limited, if two us⁃ ers choose the same codeword, collision will happen. Collision is a serious problem for SCMA, which limits its overload capa⁃ bility. For example, with 6 users transmitted over 4 units, the user overload is only 150% . Although the overloading factor can be enhanced by using longer spreading codes, the detec⁃ tion complexity will increase significantly since the size of the codebook and the searching space is enlarged. 3.2.2 PDMA For PDMA, the code in a code matrix is used to define map⁃ ping from data to a group of resources. Each element in the code corresponds to a resource in the resource group. PDMA can be detected by SIC type receiver. It also can be detected by MPA based scheme in the receiver. PDMA is designed for SIC⁃based receiver originally. The different diversity orders of different users by carefully design the code matrix facilitate the multi⁃user signal detection. The user with the largest diver⁃ October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 13 CP: cyclic prefix IFFT: inverse fast fourier transform OFDM: orthogonal frequency division multiplexing PAPR: peak⁃to⁃average⁃power⁃ratio RSMA: resource spread multiple access TDM: time division multiplexing MUD: multiple user detection MPA: message passing algorithm ▲Figure 2. RSMA block diagrams [5]. ▲Figure 3. An example of SCMA with 150% user overload [8]. Variable rate encoder TDM pilot insertion Spreader/ scrambler Low PAPR modulation Optional CP (a) Single carrier RSMA Pilot insertion Spreader/ scrambler Coder Serial to parallel IFFT Parallel to serial Cyclic prefix e j2πfct (b) OFDM RSMA e j2πfct Codebook 1 Codebook 2 Codebook 3 Codebook 4 Codebook 5 Codebook 6 (0,0) (1,0) (0,1) (1,1) (1,1) (0,0) Bit streams are mapped to sparse codewords MUD based on MPA 6 sparse codewords are transmitted over 4 orthogonal resources
  • 18. sity order is detected first, and then the user with the largest di⁃ versity order among the remaining users is detected; in this way, all users’signals will be detected. To further improve the performance of PDMA, joint detec⁃ tion based scheme is proposed. In this case the unbalance weight of each column is interpreted as the irregular code weight. As we know irregular low density parity check (LDPC) code has better performance than that of the regular one. By carefully designing the code matrix with joint detection, even better performance can be obtained by PDMA compared with regular code matrix (for example non⁃orthogonal multiple ac⁃ cess with low density signatures can be regards as regular code). The main disadvantage of PDMA is its low user overload (us⁃ er overload is defined by the number of user over the resource block that all users share). It is difficult to achieve overload of 400% with the 4⁃row code matrix (when the row of the code ma⁃ trix is K, the largest user number it supported is 2K ⁃1 [10]).The complexity is high for high order modulation when joint⁃detec⁃ tion scheme is applied. Additional pilots or preamble are need⁃ ed for channel estimation. Because the number of patterns is limited, there is high probability of collision when users are al⁃ lowed to randomly select the patterns. 3.3 Non􀆼Orthogonal Multiple Schemes Based on Interleaver IDMA was proposed by [12], [13], in which users are sepa⁃ rated by different interleavers. Low ⁃ rate channel decoding is applied and the coded bits are repeated multiple times to in⁃ crease the SINR after accumulating the received signals. After channel coding and repetition, interleaver is employed to make the transmission bits randomly distributed. A block diagram of IDMA is shown in Fig. 4 where C represents channel encod⁃ ing, S denotes repetition and π is the interleaver. The strategy of user separation for IDMA is different from other non⁃orthogo⁃ nal multiple access schemes. Interleaver is used for user sepa⁃ ration and the length of the interleaver is very large (the length of the interleaver equals to the number of the bits after channel coding and repetition), thus this provides good base for a large number of users access by using IDMA. It is reported that 64 users can be supported by IDMA which share the same re⁃ source block [12]. This goal can never be achieved by other non⁃ orthogonal multiple access schemes at present. At the receiver side each user’s signal is detected, demodu⁃ lated and de⁃interleaved according to its own interleaver pat⁃ terns. The soft information of decoded bits is input to elementa⁃ ry signal estimator (ESE) for soft information updating. After soft information updating new soft information is input to the decoder for channel decoding again. Several iterative detec⁃ tions between ESE and channel decoder are needed to achieve the best performance. The detection and decoding complexity does not increase exponentially with the user number and total spectral efficiency. The complexity increases linearly, which is also different from other non ⁃ orthogonal multiple access schemes which use joint detection and decoding scheme. The main advantages of IDMA are its high user overload and excellent performance. And high spectral efficiency can be achieved by IDMA (as high as 8 b/s/Hz). The performance gap between IDMA simulation result and the system capacity bound is almost the same from the spectral efficiency 1 b/s/Hz to 8 b/s/Hz (this means the detection and decoding scheme is very robustness) [12]. These two merits are seldom achieved by other non⁃orthogonal multiple schemes simultaneously. The main disadvantage of IDMA may be its large decoding complexity and decoding latency, especially when a large num⁃ ber of users are supported. The reason is that when large num⁃ ber of iterative detection and decoding are needed with the in⁃ creasing of user number. For example, tens of channel decoder procedures are needed in the signal detection and tens of inter⁃ active actions between channel decoder and ESE detector are required. Thus high convergence algorithm is needed in the signal detection for IDMA in future. To solve the problem of large decoding complexity and decoding latency, interleaver patterns can be pre⁃allocate to small number of users, i.e., the relatively small pool size, so that the complexity of blind decod⁃ ing and channel decoding latency can be maintained below cer⁃ tain level. Another disadvantage is that additional pilots or long preamble is needed to estimate the users’channels. 3.4 Non􀆼Orthogonal Multiple Access (NOMA) Scheme Based on Power􀆼Domain Division Multi⁃user signals can be superposed together in NOMA. In NOMA, capacity or throughput improvement can be expected by sharing the same radio resources among multiple user equipments (UEs) as shown in Fig. 5a and Fig. 5b. A typical application scenario of NOMA is that a cell⁃center user and a cell⁃edge user are serviced by NOMA. Due to small path loss of cell center user, in the signal detection it is detected first and the signal of cell edge user is treated as interference. In the signal detection of cell edge user, the signal of cell center user is detected and decoded first. Then the signal of the cell center user is cancelled from the received signal and signal of cell edge user is detected and decoded. The main advantage of NOMA is that excellent performance can be achieved when a cell center user and cell edge user are scheduled with moderate computational complexity (SIC detec⁃ Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS14 ▲Figure 4. IDMA block diagram [13]. Multiple access channel π1 x1 SC d1 Transmitter for user 1 Transmitter for user K πK xK SC dK … … Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei
  • 19. tor is always applied). And a user overload of 200% is easily achieved. The main disadvantage of NOMA is that there is re⁃ striction on the scheduled users. Usually a cell center user and a cell edge user should be scheduled on the same resource block. When two cell center users or two cell edge users are scheduled and SIC ⁃ type receiver is applied, there is perfor⁃ mance loss because one user always has low SINR due to inter⁃ ference from another user’s signal. The NOMA is designed for eMBB originally. Thus when it is applied for mMTC, the re⁃ ceived SINR would not be high and the number of supported users is very limited (two or three users are supported on the same resource block, which is much smaller than other non⁃or⁃ thogonal multiple access schemes). And additional pilots or long preamble is needed to estimate the users’channels. A summary of these non ⁃ orthogonal multiple schemes are shown on Table 1. They are compared in terms of multiplexing domain, user overload, receiver type, receiver complexity and so on. Among these schemes MUSA achieves a good balance between performance and complexity, such as high user over⁃ load, low implementation complexity and flexible in grant⁃free transmission. 4 Application Challenges of Non⁃Orthogonal Multiple Access Schemes in 5G Followings are the requirements for the non⁃orthogonal mul⁃ tiple access schemes. These factors should be considered when we design the non⁃orthogonal multiple access schemes. 4.1 Coverage Coverage is an important issue for mMTC since terminals may distribute over a large area, thus it is crucial for non⁃or⁃ thogonal multiple access schemes to support terminals with dif⁃ ferent received power due to path loss. And the non⁃orthogonal multiple access schemes should have the ability of robustness to the high interference. To increase the coverage, low code rate channel coding or large spreading factor could be consid⁃ ered. High efficiency power amplifier is appealing for coverage extension, which requires transmit signals with low PAPR. 4.2 PAPR When the non⁃orthogonal multiple access scheme is applied for uplink, PAPR should be considered to increase the trans⁃ mission efficiency and reduce the transmission power thus save the battery life. The battery life is desired to be 10 years for mMTC, so it puts a big challenge on the non ⁃ orthogonal multiple access scheme. The signal of the non⁃orthogonal mul⁃ tiple access schemes which have low PAPR will be preferred in practical implementation. Filtered π/2 ⁃ binary phase shift keying (BPSK) and Gaussian filtered minimum shift keying (GMSK) have good property of low PAPR and are employed for PAPR reduction in RSMA [16]. 4.3 Implementation Complexity The implementation complexity includes two parts: transmit⁃ ter implementation complexity and receiver implementation complexity. Because multi⁃user detection is carried out at re⁃ ceiver side, which has the highest complexity over the entire signal processing chain, the main implementation complexity is at the receiver side. Two types of receivers are always ap⁃ plied for non⁃orthogonal multiple access schemes: SIC⁃based receiver and joint ⁃ detection ⁃ based receiver. The former can achieve a good balance between performance and complexity. As the number of user increases, the complexity only increases linearly. While it suffers performance loss in some cases, such as the path⁃losses among different users are the same. Joint⁃de⁃ tection ⁃ based receiver achieves excellent performance at the Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 15 ▲Figure 5. NOMA block diagram. NOMA: non⁃orthogonal multiple access ▼Table 1. Summary of different non⁃orthogonal multiple access schemes MUSA: multi⁃user shared multiple access RSMA: resource spread multiple access SCMA: sparse code multiple access PDMA: pattern division multiple access IDMA: interleaver⁃division multiple access NOMA: non⁃orthogonal multiple access SIC: successive interference cancelation BS: base station Multiplexing domain User overload Receiver type Receiver complexity Grant⁃free transmission MUSA Spreading High SIC Low Users can randomly pick up spreading sequence RSMA Spreading/ scramble Low Raker or SIC Low Power control needed SCMA Codebooks Middle Joint detection High Codeword for each user is predefined and known at BS. Codeword collision is a problem due to limited number of codewords PDMA Pattern Middle SIC or joint detection Low for SIC High for joint detection Pattern is predefined and known at BS. User collision is a problem due to limited number of patterns IDMA Interleaver High Iterative detection and decoding High* Interleaver patterns are known at BS NOMA Power Low SIC Low Grant⁃ based (a) NOMA transmission (b) Signal strength for NOMA Base station Cell center user Cell edge user Strength of cell edge user signal Strength of cell center user signal Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei * Unlike joint detection scheme whose complexity increases exponentially as the number of the users and spectral efficiency increases, the complexity of IDMA only linear in⁃ creases with the number of users and the spectral efficiency. The high complexity is due to large number of iterative detection and decoding.
  • 20. cost of high computational complexity. Although by some de⁃ signs, such as sparse coding matrix, the decoding complexity is reduced significantly, however, as the constellation size and the number of users increase, the decoding complexity grows exponentially. This bottleneck should be solved before such scheme is employed in practical systems. 4.4 Combination with Multiple􀆼Input Multiple􀆼Output (MIMO) By applying MIMO technique large system capacity or high transmission/receiver reliability can be achieved. It had been proved that MIMO is a very effective technique in wireless communication systems. The non ⁃ orthogonal multiple access schemes should be amiable for MIMO. As the first step, SISO is assumed in the research of the new non⁃orthogonal multiple access schemes. However, compatibility with MIMO should be considered in the next research step. 4.5 Flexibility The non ⁃ orthogonal multiple access schemes should have flexibility. It can change its parameters to support different use scenarios. For example, in some cases high user overload is the system design target, while in other cases coverage is the most important factor. This imposes requirements on the non⁃ orthogonal multiple access scheme design. By changing the pa⁃ rameter of the non⁃orthogonal multiple access schemes, differ⁃ ent targets can be achieved. Another example is that non⁃or⁃ thogonal multiple access schemes should support both multi⁃ carrier system and single⁃carrier systems to facilitate its appli⁃ cation scenarios. 5 Conclusion This article reviews the main non ⁃ orthogonal multiple ac⁃ cess schemes for 5G. Their principles and unique properties are discussed. MUSA can support high user overload with low implementation complexity and is more suitable for grant⁃free transmission. RSMA is suitable for single⁃carrier system and multi ⁃ carrier system. It has good property of large coverage. SCMA can achieve additional shaping gain and PDMA has the flexibility in the patterns design. IDMA can accommodate very high user overload and support high spectral efficiency at the cost of large decoding complexity and decoding latency. NO⁃ MA works well for large SINR difference among the non ⁃ or⁃ thogonal multiple users. At the same time they have their own disadvantages. It is important to integrate the advantages of dif⁃ ferent schemes to make the final designed scheme fulfill the challenging requirements of coming 5G. Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS16 References [1] Discussion on Multiple Access for New Radio Interface, 3GPP R1⁃162226, Apr. 2016. [2] Z. Yuan, G. Yu, W. Li, Y. Yuan, and X. Wang,“Multi⁃user shared access for in⁃ ternet of things,”in IEEE Vehicular Technology Conference, Nanjing, China, May 2016, pp 1-5. doi: 10.1109/VTCSpring.2016.7504361. [3] Receiver Implementation for MUSA, 3GPP R1⁃164270, May 2016. [4] Contention ⁃ Based Non ⁃ Orthogonal Multiple Access for UL mMTC, 3GPP R1 ⁃ 164269, May 2016. [5] Resource Spread Multiple Access, 3GPP R1⁃164688, May 2016. [6] M. Taherzadeh, H. Nikopour, A. Bayesteh, H. Baligh,“SCMA codebook de⁃ sign”, in IEEE Vehicular Technology Conference, Vancouver, Canada, Sept. 2014, pp.1-5, doi: 10.1109/VTCFall.2014.6966170. [7] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE Internation⁃ al Symposium On Personal, Indoor And Mobile Radio Communications, London, UK, Sept. 2013, pp. 332-336. doi: 10.1109/PIMRC.2013.6666156. [8] Future Mobile Communication Forum. (2016, Jul. 7). 5G white paper v2.0, part d—alternative multiple access v1 [Online]. Available: http://www.future ⁃ forum. org/dl/151106/whitepaper.rar [9] Candidate Solution for New Multiple Access, 3GPP R1⁃163383, Apr. 2016. [10] X. Dai, S. Chen, S. Sun, et al.,“Successive interference cancelation amenable multiple access (SAMA) for future wireless communications,”in Proc. IEEE In⁃ ternational Conference on Communication Systems, Macau, China, Nov. 2014, pp. 222-226. doi: 10.1109/ICCS.2014.7024798. [11] X. Dai,“Successive interference cancellation amenable space⁃time codes with good multiplexing⁃diversity tradeoff,”Wireless Personal Communications, vol. 55, no. 4, pp. 645-654, Dec. 2010. doi: 10.1007/s11277⁃009⁃9826⁃9. [12] P. Li, L. Liu, K. Wu, and W. K. Leung,“On interleave⁃division multiple⁃ac⁃ cess,”in IEEE International Conference on Communications, Paris, France, Jun. 2004, pp. 2869-2873. doi: 10.1109/ICC.2004.1313053. [13] P. Li, L. Liu, K. Wu, and W. K. Leung,“Interleave division multiple⁃access,” IEEE Transactions on Wireless Communications, vol. 5, no. 4, pp. 938-947, Apr. 2006. doi: 10.1109/TWC.2006.1618943. [14] Y. Saito, Y. Kishiyama, A. Benjebbour, et al.,“Non⁃orthogonal multiple access (NOMA) for cellular future radio access,”in IEEE Vehicular Technology Con⁃ ference, Dresden, Germany, Jun. 2013, pp. 1-5. doi: 10.1109/VTC Spring.2013. 6692652. [15] Receiver Details and Link Performance for MUSA, 3GPP R1⁃166404, Aug. 2016. [16] Resource Spread Multiple Access, 3GPP R1⁃166359, Aug. 2016. Manuscript received: 2016⁃07⁃07 Non⁃Orthogonal Multiple Access Schemes for 5G YAN Chunlin, YUAN Zhifeng, LI Weimin, and YUAN Yifei YAN Chunlin (yan.chunlin@zte.com.cn) received his PhD degree from University of Electronic Science and Technology of China (UESTC), China in 2004. He worked at DOCOMO Beijing communications lab from 2005 to 2016. Since 2016 he has been with ZTE Corporation. He has published tens of papers in IEEE ICC, Globecom, VTC, PIMRC and other international conferences. His main research interests in⁃ clude synchronization, binary and non⁃binary channel coding, MIMO detection and non⁃orthogonal multiple access technique for 5G. YUAN Zhifeng (yuan.zhifeng@zte.com.cn) received his MS degree in signal and in⁃ formation processing from Nanjing University of Post and Telecommunications (NUPT), China in 2005. He has been worked with the Wireless Technology Ad⁃ vance Research Department of ZTE Corporation since 2006 and the leader of the team for new multi⁃access (NMA) for 5G wireless systems since 2012. His research interests include wireless communication, MIMO systems, information theory, multi⁃ ple access, error control coding, adaptive algorithm, and high⁃speed VLSI design. LI Weimin (li.weimin6@zte.com.cn) received his master degree from NUPT, China. He joined in ZTE Corporation in 2010, and is responsible for technology research of power control and interference control in wireless communications. His current re⁃ search focuses on multiple access technology for 5G system. YUAN Yifei (yuan.yifei@zte.com.cn) received his master degree from Tsinghua Uni⁃ versity, China, and PhD from Carnegie Mellon University, USA. He was with Alca⁃ tel⁃Lucent from 2000 to 2008, working on 3G/4G key technologies. Since 2008, he has been with ZTE as the technical director of standards research on LTE⁃advanced physical layer and 5G new radio. His research interests include MIMO, channel cod⁃ ing, resource scheduling, multiple access, and NB⁃IoT. He was admitted to Thou⁃ sand Talent Plan Program of China in 2010. He has extensive publications, includ⁃ ing two books on LTE⁃Advanced. BiographiesBiographies
  • 21. A Survey of Downlink Non⁃Orthogonal MultipleA Survey of Downlink Non⁃Orthogonal Multiple Access forAccess for 55G Wireless Communication NetworksG Wireless Communication Networks WEI Zhiqiang 1 , YUAN Jinhong 1 , Derrick Wing Kwan Ng 1 , Maged Elkashlan2 , and DING Zhiguo3 (1. The University of New South Wales, Sydney, NSW 2052, Australia; 2. Queen Mary University of London, London E1 4NS, UK; 3. Lancaster University, Lancaster LA1 4YW, UK) Abstract Non⁃orthogonal multiple access (NOMA) has been recognized as a promising multiple access technique for the next generation cel⁃ lular communication networks. In this paper, we first discuss a simple NOMA model with two users served by a single⁃carrier si⁃ multaneously to illustrate its basic principles. Then, a more general model with multicarrier serving an arbitrary number of users on each subcarrier is also discussed. An overview of existing works on performance analysis, resource allocation, and multiple⁃in⁃ put multiple⁃output NOMA are summarized and discussed. Furthermore, we discuss the key features of NOMA and its potential re⁃ search challenges. non⁃orthogonal multiple access (NOMA); successive interference cancellation (SIC); resource allocation; multiple⁃input multiple⁃ output (MIMO) Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 003 http://www.cnki.net/kcms/detail/34.1294.TN.20161019.0829.002.html, published online October 19, 2016 Special Topic 1 Introduction and Background he fifth generation (5G) communication system is on its way. It is widely believed that 5G is not just an incremental version of the fourth generation (4G) communication systems [1] due to the in⁃ creasing demand of data traffic and the expected new services and functionalities, such as internet⁃of⁃things (IoT) and cloud⁃ based architectural applications [2]. These envisioned services pose challenging requirements for 5G wireless communication systems, such as much higher data rates (100-1000 times fast⁃ er than current 4G technology), lower latency (1 ms for a roundtrip latency), massive connectivity and support of diverse quality of service (QoS) (106 devices/km2 with diverse QoS re⁃ quirements) [1]. From a technical perspective, to meet the aforementioned challenges, some potential technologies, such as massive multiple⁃input multiple⁃output (MIMO) [3], [4], mil⁃ limeter wave [5], [6], and ultra densification and offloading [7]- [9], have been discussed extensively. Besides, it is expected to employ a future radio access technology for 5G, which is flexi⁃ ble, reliable [10], and efficient in terms of energy and spec⁃ trum [11], [12]. Radio access technologies for cellular commu⁃ nications are characterized by multiple access schemes, such as frequency⁃division multiple access (FDMA) for the first gen⁃ eration (1G), time⁃division multiple access (TDMA) for the sec⁃ ond generation (2G), code ⁃ division multiple access (CDMA) used by both 2G and the third generation (3G), and orthogonal frequency division multiple access (OFDMA) for 4G. All these conventional multiple access schemes are categorized as or⁃ thogonal multiple access (OMA) technologies, where different users are allocated to orthogonal resources in either time, fre⁃ quency, or code domain in order to mitigate multiple access in⁃ terference (MAI). However, OMA schemes are not sufficient to support the massive connectivity with diverse QoS require⁃ ments. In fact, due to the limited degrees of freedom (DoF), some users with better channel quality have a higher priority to be served while other users with poor channel quality have to wait to access, which leads to high unfairness and large laten⁃ cy. Besides, it is inefficient when allocating DoF to users with poor channel quality. In this survey, we focus on one promising technology, non⁃orthogonal multiple access (NOMA), which in our opinion will contribute to disruptive design changes on ra⁃ dio access and address the aforementioned challenges of 5G. In contrast to conventional OMA, NOMA transmission tech⁃ niques intend to share DoF among users via superposition and consequently need to employ multiple user detection (MUD) to separate interfered users sharing the same DoF, as illustrated in Fig. 1. NOMA is beneficial to enlarge the number of connec⁃ tions by introducing controllable symbol collision in the same DoF. Therefore, NOMA can support high overloading transmis⁃ T October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 17
  • 22. Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS18 sion and further improve the system capacity given limited re⁃ source (spectrum or antennas). In addition, multiple users with different types of traffic request can be multiplexed to transmit concurrently on the same DoF to improve the latency and fair⁃ ness. The comparison of OMA and NOMA is summarized in Table 1. As a result, NOMA has been recognized as a promis⁃ ing multiple access technique for the 5G wireless networks due to its high spectral efficiency, massive connectivity, low la⁃ tency, and high user fairness [13]. For example, multiuser su⁃ perposition transmission (MUST) has been proposed for the third generation partnership project long ⁃ term evolution ad⁃ vanced (3GPP⁃LTEA) networks [14]. Three kinds of non⁃orthogonal transmission schemes have been proposed and studied in the MUST study item. Through the system⁃level performance evaluation, it has been shown that the MUST can increase system capacity as well as improve user experience. Recently, several NOMA schemes have been proposed and received significant attention. According to the domain of mul⁃ tiplexing, the authors in [13] divided the existing NOMA tech⁃ niques into two categories, i.e., code domain multiplexing (CDM) and power domain multiplexing (PDM). The CDM⁃NO⁃ MA techniques, including low⁃density spreading (LDS) [15]- [17], sparse code multiple access (SCMA) [18], pattern divi⁃ sion multiple access (PDMA) [19], etc, introduce redundancy via coding/spreading to facilitate the users separation at the re⁃ ceiver. For instance, LDS ⁃ CDMA [15] intentionally arranges each user to spread its data over a small number of chips and then interleave uniquely, which makes optimal MUD affordable at receiver and exploits the intrinsic interference diversity. LDS⁃ OFDM [16], [17], as shown in Fig. 2, can be interpreted as a system which applies LDS for multiple access and OFDM for multicarrier modulation. Besides, SCMA is a generalization of LDS methods where the modulator and LDS spreader are merged. On the other hand, PDM⁃NOMA exploits the power do⁃ main to serve multiple users in the same DoF, and performs successive interference cancellation (SIC) at users with better channel conditions. In fact, the non⁃orthogonal feature can be introduced either in the power domain only or in the hybrid code and power domain. Although DM⁃NOMA has the poten⁃ tial code gain to improve spectral efficiency, PDM⁃NOMA has a simpler implement since there is almost no big change in the physical layer procedures at the transmitter side compared to current 4G technologies. In addition, PDM⁃NOMA paves the way for flexible resource allocation via relaxing the orthogonali⁃ ty requirement to improve the performance of NOMA, such as spectral efficiency [20], [21], energy efficiency [22], and user fairness [23]. Therefore, this paper will focus on the PDM⁃NO⁃ MA, including its basic concepts, key features, existing works, and future research challenges. 2 Fundamentals of NOMA This section presents the basic model and concepts of single⁃ antenna downlink NOMA. The first subsection presents a sim⁃ ple downlink single⁃carrier NOMA (SC⁃NOMA) system serving two users simultaneously, while the second subsection pres⁃ ents a more general multi⁃carrier NOMA (MC⁃NOMA) model for serving an arbitrary number of users in each subcarrier. 2.1 Two􀆼User SC􀆼NOMA1 Benjebbour, Saito et al. [24], [25] proposed the system mod⁃ el of downlink NOMA with superposition transmission at the base station (BS) and successive interference cancellation DoF: degrees of freedom ▲Figure 1. From OMA to NOMA via power domain multiplexing. AWGN: additive white Gaussian noise FEC: forward error correction LDS: low⁃density spreading OFDM: orthogonal frequency⁃division multiplexing 1 In this paper, a two⁃user NOMA system means that two users are multi⁃ plexed on each subcarrier simultaneously. Similarly, a multiuser NOMA system means that an arbitrary number of users are multiplexed on each subcarrier simultaneously. ▼Table 1. Comparison of OMA and NOMA NOMA: non⁃orthogonal multiple access OMA: orthogonal multiple access QoS: quality of service OMA NOMA Advantages Simpler receiver detection •Higher spectral efficiency •Higher connection density •Enhanced user fairness •Lower latency •Supporting diverse QoS Disadvantages •Lower spectral efficiency •Limited number of users •Unfairness for users •Increased complexity of receivers •Higher sensitivity to channel uncertainty ▲Figure 2. Block diagram of an uplink LDS⁃OFDM system. Power Orthogonality between users Power Superposition & power allocation DoF DoF FEC encoder b1 Symbol mapper LDS spreader S1 OFDM modulator OFDM channel FEC encoder bM Symbol mapper LDS spreader SM OFDM modulator OFDM channel … AWGN FEC decoder b1 Symbol demapper S1 FEC decoder bM Symbol demapper SM LDS detector … … … y1 yK OFDM demodulator …
  • 23. A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 19 (SIC) at the user terminals, which is illustrated in Fig. 3 in case of one BS and two users. The BS transmits the messages of both user 1 and user 2, i.e., s1 and s2 , with different trans⁃ mit powers p1 and p2 , on the same subcarrier, respectively. The corresponding transmitted signal is represented by x = p1 s1 + p2 s2 , (1) where transmit power is constrained by p1 + p2 = 1. The re⁃ ceived signal at user i is given by yi = hi x + vi , (2) where hi denotes the complex channel coefficient including the joint effect of large scale fading and small scale fading. Variable vi denotes the additive white Gaussian noise (AW⁃ GN), and vi ∼ CN(0,σ2 i ) , where CN(0,σ2 i ) denotes the circula⁃ rly symmetric complex Gaussian distribution with mean zero and variance σ2 i . We assume that user 1 is the cell⁃center u⁃ ser with a better channel quality (strong user), while user 2 is the cell⁃edge user with a worse channel quality (weak user), i.e., ( ||h1 2 σ2 1)≥( ||h2 2 σ2 2) . According to the NOMA protocol [26], the BS will allocate more power to the weak user to pro⁃ vide fairness and facilitate the SIC process, i.e., p1 ≤ p2 . In downlink SC⁃NOMA, the SIC process is implemented at the receiver side. The optimal SIC decoding order is in the de⁃ scending order of channel gains normalized by noise. It means that user 1 will decode s2 first and remove the inter⁃user inter⁃ ference of user 2 by subtracting s2 from the received signal yi before decoding its own message s1 . On the other hand, u⁃ ser 2 does not perform interference cancellation and directly decodes its own message s2 with interference from user 1. For⁃ tunately, the power allocated to user 2 is larger than that of us⁃ er 1 in the aggregate received signal y2, which will not intro⁃ duce much performance degradation compared to allocating us⁃ er 2 on this subcarrier exclusively. The rate region of SC⁃NO⁃ MA is illustrated in Fig. 4 in comparison with that of OMA, where it has been proved that NOMA schemes are very likely to outperform OMA schemes in [27]. It is noted that the rate re⁃ gion of NOMA only covers a part of the capacity region of broadcast channel with SIC receiver [28] due to the power con⁃ straint p1 ≤ p2 . 2.2 Multiuser MC􀆼NOMA For a downlink MC⁃NOMA system with one BS serving an arbitrary number of users, such as N, the available bandwidth is divided into a set of K subcarriers, where N> K, i.e., an over⁃ loading scenario that OFDMA cannot afford. The channel be⁃ tween user n and the BS on subcarrier k is denoted by hk,n , and is assumed to be perfectly known at both the transmitter and receiver side. The BS schedules all users across all subcar⁃ riers by ξk and ζn , where ξk denotes a user set allocated on subcarrier k and ζn denotes a subcarrier set occupied by user n. Without loss of generality, the channel gains of all users allo⁃ cated on subcarrier k are sorted as ||hk,b(1) 2 ≥ ||hk,b(2) 2 ≥ … ≥ | | | |hk,b( ||ξk ) 2 , where ||ξk denotes the card⁃ inality of the user set ξk and b( )∙ indicates the mapping b⁃ etween the sorted channel gain order and the original one. For instance, for subcarrier k occupied by three users ξk ={ }1,2,3 and ||hk,2 2 ≥ ||hk,3 2 ≥ ||hk,1 2 , we will have b(1)= 2 , b(2)= 3 , and b(3)= 1, respectively. It is noted that the mapping func⁃ tions are various on different subcarriers due to users’differ⁃ ent frequency selective fading patterns. According to NOMA protocol [26], all users in ξk share sub⁃ carrier k by different transmission power pk,b(l) based on the gi⁃ ven channel gain, where l = 1,2,…, ||ξk and pk,b(1) ≤ pk,b(2) ≤ … ≤ pk,b( ||ξk ) . The sharing strategy saves the su⁃ bcarriers those might be wasted by only transmitting the mes⁃ sages of the weak users and accommodates more users with di⁃ SIC: successive interference cancellation ▲Figure 3. A downlink NOMA model with one base station and two users. ▲Figure 4. The rate region of two⁃user SC⁃NOMA in comparison with that of OMA. User 1 is a strong user with ( ||h1 2 σ2 1 )= 100 , while user 2 is a weak user with ( ||h2 2 σ2 2)= 1 . NOMA: non⁃orthogonal multiple access OMA: orthogonal multiple access Power … … Frequency User 1 User 2 Base station SIC of user 2’s signal User 2’s signal decoding User 1’s signal decoding 7 Rateofuser1(bit/s/Hz) 1.0 Rate of user 2 (bit/s/Hz) 6 5 4 3 2 1 0 0.90.80.70.60.50.40.30.20.10 Capacity gain NOMA OMA
  • 24. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS20 Special Topic verse QoS requirements, which is favorable to massive connec⁃ tivity and IoT in 5G networks. All messages of users in ξk are superimposed on subcarrier k, where the transmitted signal is given by xk =∑l = 1 ||ξk pk,b(l) Sk,b(l), (3) where Sk,b(l) and pk,b(l) denote the message and allocated pow⁃ er of user n(l) on subcarrier k, respectively. Assuming the independent and identically distributed (IID) AWGN over all subcarriers and all users for simplicity, the us⁃ er scheduling, power allocation, and the SIC decoding order on⁃ ly depend on the channel gain order. At the receiver side, the received signal of user b(l) on subcarrier k can be represented by yk,b(l) = hk,b(l) xk + v (4) = hk,b(l)∑l′ = 1 ||ξk pk,b(l′) Sk,b(l′) + v, ∀l ∈{ }1,2,…, ||ξk , (5) where v denotes the AWGN, i.e., vi ∼ CN(0,σ2 ) , and σ2 de⁃ notes the noise power. On subcarrier k, the scheduled users in ξk perform SIC to eliminate inter⁃user interference. Similar to the case of two⁃us⁃ er NOMA, the optimal SIC decoding order is in the descending channel gain order, i.e., { }b(1),b(2),…,b( )||ξk . It means that the user b(l) first decodes and subtracts the message sk,b(l′) , ∀l′ > l , in descending order from ||ξk to l + 1, and then de⁃ codes its own message sk,n(l) by treating sk,n(l′) , ∀l′ > l , as in⁃ terference. 3 Performance and Key Features of NOMA In this section, we present the performance characteristics of NOMA in existing works, and then discuss the pros and cons of NOMA schemes. 3.1 Performance of NOMA It has been shown that NOMA offers considerable perfor⁃ mance gain over OMA in terms of spectral efficiency and out⁃ age probability [25]-[27], [29]-[31]. Initially, the performance of NOMA was evaluated through simulations given perfect CSI by utilizing the proportional fairness scheduler [25], [29], frac⁃ tional transmission power allocation (FTPA) [25], and tree ⁃ search based transmission power allocation (TTPA) [30]. These works showed that the overall cell throughput, cell⁃edge user throughput, and the degrees of proportional fairness achieved by NOMA are all superior to those of OMA. In [27], the author analyzed a two ⁃ user SC ⁃ NOMA system under statistical CSI from an information theoretic perspective, where it was proved that NOMA outperforms native TDMA with high probability in terms of both the sum rate and individual rates. In [26], for a fixed power allocation, the performance of a multiuser SC⁃NO⁃ MA system in terms of outage probability and ergodic sum rates under statistical CSI was investigated in a cellular down⁃ link scenario with randomly deployed users. With the proposed asymptotic analysis, it showed that user n experiences a diver⁃ sity gain of n and NOMA is asymptotically equivalent to the op⁃ portunistic multiple access technique. Furthermore, the au⁃ thors in [32] analyzed the performance degradation of a mul⁃ tiuser SC ⁃ NOMA system on outage probability and average sum rates due to partial CSI. It showed that NOMA based on second order statistical CSI always achieves a better perfor⁃ mance than that of NOMA based on imperfect CSI, while it can achieve similar performance to the NOMA with perfect CSI in the low SNR region. In summary, most of the existing works on performance anal⁃ ysis of NOMA focused on a SC⁃NOMA system since the user scheduling in MC⁃NOMA complicates the analysis due to its combinatorial nature. A remarkable work in [31] characterized the impact of user pairing on the performance of a two⁃user SC⁃ NOMA system with fixed power allocation and cognitive radio inspired power allocation, respectively. The authors proved that, for fixed power allocation, the performance gain of NOMA over OMA increases when the difference in channel gains be⁃ tween the paired users becomes larger. However, further explo⁃ ration on performance analysis of MC⁃NOMA system should be carried out in the future since user scheduling is critical for performance of NOMA. 3.2 Pros 1) Higher spectral efficiency By exploiting the power domain for user multiplexing, NO⁃ MA systems are able to accommodate more users to cope with system overload. In contrast to allocate a subcarrier exclusive⁃ ly to a single user in OMA scheme, NOMA can utilize the spec⁃ trum more efficiently by admitting strong users into the subcar⁃ riers occupied by weak users without compromising much their performance via utilizing appropriate power allocation and SIC techniques. 2) Better utilization of heterogeneity of channel conditions As we mentioned before, NOMA schemes intentionally mul⁃ tiplex strong users with weak users to exploit the heterogeneity of channel condition. Therefore, the performance gain of NO⁃ MA over OMA is larger when channel gains of the multiplexed users become more distinctive [31]. 3) Enhanced user fairness By relaxing the orthogonal constraint of OMA, NOMA en⁃ ables a more flexible management of radio resources and offers an efficient way to enhance user fairness via appropriate re⁃ source allocation [23]. 4) Applicability to diverse QoS requirements NOMA is able to accommodate more users with different types of QoS requests on the same subcarrier. Therefore, NO⁃ MA is a good candidate to support IoT which connects a great A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo
  • 25. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 21 Special Topic number of devices and sensors requiring distinctive targeted rates. 3.3 Cons 1) The BS needs to know the perfect channel state informa⁃ tion (CSI) to arrange the SIC decoding order, which increases the CSI feedback overhead. 2) The SIC process introduces a higher computational com⁃ plexity and delay at the receiver side, especially for multicarri⁃ er and multiuser systems. 3) The strong users have to know the power allocation of the weaker users in order to perform SIC, which also increases the system signalling overhead. 4) Allocating more power to the weak users, who are general⁃ ly in the cell⁃edge, will introduce more inter⁃cell interferences into the whole system. 4 Design of NOMA Schemes Due to the remarkable performance gain of NOMA over con⁃ ventional OMA, a lot of works on design of NOMA schemes have been proposed in literatures. In this section, we present the existing works on resource allocation of NOMA and MIMO⁃ NOMA, and then briefly introduce other works associated with NOMA. 4.1 Resource Allocation Resource allocation has received significant attention since it is critical to improve the performance of NOMA. However, optimal resource allocation is very challenging for MC⁃NOMA systems, since user scheduling and power allocation couple with each other severely. Some initial works on resource alloca⁃ tion in [25], [29], [30] have been reported, but they are far from optimal. In [33], [34], the authors studied a two⁃user MC⁃NO⁃ MA system by minimizing the number of subcarriers assigned under the constraints of maximum allowed transmit power and requested data rates, and further introduced a hybrid orthogo⁃ nal ⁃ nonorthogonal scheme. Furthermore, the authors in [21] studied a joint power and subcarrier allocation problem for a two ⁃ user MC ⁃ NOMA system. They proposed an optimal scheme and a suboptimal scheme with close⁃to⁃optimal perfor⁃ mance based on monotonic optimization and difference of con⁃ vex function programming, respectively. Besides, there are also several works on resource allocation for multiuser MC⁃NOMA systems. In [35], the authors formulat⁃ ed the resource allocation problem to maximize the sum rate, which is a non⁃convex optimization problem due to the binary constraint and the existence of the interference term in the ob⁃ jective function. Interestingly, they proposed a suboptimal solu⁃ tion by employing matching theory and water⁃filling power allo⁃ cation. In [36], the authors presented a systematic approach for NOMA resource allocation from a mathematical optimization point of view. They formulated the joint power and channel al⁃ location problem of a downlink multiuser MC⁃NOMA system, and proved its NP⁃hardness based on [37] via defining a spe⁃ cial user. Furthermore, they proposed a competitive subopti⁃ mal algorithm based on Lagrangian duality and dynamic pro⁃ gramming, which significantly outperforms OFDMA as well as NOMA with FTPA. Most of works aforementioned focus on the optimal resource allocation for maximizing the sum rate. However, fairness is an⁃ other objective to optimize for resource allocation of NOMA. Proportional fairness (PF) has been adopted as a metric to bal⁃ ance the transmission efficiency and user fairness in many works [38], [39]. In [40], the authors proposed a user pairing and power allocation scheme for downlink two⁃user MC⁃NO⁃ MA based on the PF objective. A prerequisite for user pairing was given and a closed⁃form optimal solution for power alloca⁃ tion was derived. Apart from PF, max⁃min or min⁃max methods are usually adopted to achieve user fairness. Given a preset user group, the authors in [23] studied the power allocation problem from a fairness standpoint by maxi⁃ mizing the minimum achievable user rate with instantaneous CSI and minimizing the maximum outage probability with aver⁃ age CSI. Although the resulting problems are non⁃convex, sim⁃ ple low⁃complexity algorithms were developed to provide close⁃ to⁃optimal solutions. Similarly, another paper [41] studied the outage balancing problem of a downlink multiuser MC⁃NOMA system to maximize the minimum weighted success probability with and without user grouping. Joint power allocation and de⁃ coding order selection solutions were given, and the inter ⁃ group power and resource allocation solutions were also provid⁃ ed in the paper. In summary, many existing works focus on the resource allo⁃ cation for NOMA systems under perfect CSI at the transmitter side. However, there are only few works on the joint user scheduling and power allocation problem for MC⁃NOMA sys⁃ tems under imperfect CSI, not to mention the SIC decoding or⁃ der selection problem. In fact, under imperfect CSI, the SIC de⁃ coding order cannot be determined by channel gain order, and some other metrics, such as distance, priority, and target rates, are potential criteria to decide the SIC decoding order. 4.2 MIMO􀆼NOMA The application of MIMO techniques to NOMA systems is important for enhancing the performance gains of NOMA. Therefore, MIMO⁃NOMA is another hot topic that has been re⁃ searched, where the BS and users are equipped with multiple antennas, and multiple users in the same beam are multi⁃ plexed on power domain. Fig. 5 illustrates a simple MIMO⁃NO⁃ MA system with one base station and four users. Initially, the concept of MIMO ⁃ NOMA was proposed in [30], [42], [43], which demonstrated that MIMO ⁃ NOMA outperforms conven⁃ tional MIMO OMA. The authors in [44] proposed a two⁃user MIMO⁃NOMA scheme with a clustering and power allocation algorithm, where the correlation and channel gain difference A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo
  • 26. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS22 Special Topic were taken into consideration to reduce intra ⁃ beam interfer⁃ ence and inter⁃beam interference simultaneously. In [45], the authors proposed a minimum power multicast beamforming scheme and applied to two⁃user NOMA systems for multi⁃reso⁃ lution broadcasting. The proposed two ⁃ stage beamforming method outperforms the zero ⁃ forcing beamforming scheme in [44]. The design of precoding and detection algorithms also re⁃ ceived considerable attention since they are the key to elimi⁃ nate or reduce inter⁃cluster interference. The authors in [46] studied the ergodic sum capacity maximization problem of a two⁃user MIMO⁃NOMA system under statistical CSI with the total power constraint and minimum rate constraint for the weak user. This paper derived the optimal input covariance ma⁃ trix, and proposed the optimal power allocation scheme as well as a low complexity suboptimal solution. Furthermore, in [47], the authors studied the sum rate optimization problem of two⁃ user MIMO ⁃ NOMA under perfect CSI with the same con⁃ straints, while different precoders were assigned to different us⁃ ers. The optimal precode covariance matrix was derived by uti⁃ lizing the duality between uplink and downlink, and a low com⁃ plexity suboptimal solution based on singular value decomposi⁃ tion (SVD) was also provided. In [48], the authors proposed a new design of precoding and detection matrices for a downlink multiuser MIMO⁃NOMA system, then analyzed the impact of user pairing as well as power allocation on the sum rate and outage probability of MIMO ⁃ NOMA system. Furthermore, in [49], a transmission framework based on signal alignment was proposed for downlink and uplink two⁃user MIMO⁃NOMA sys⁃ tems. The authors in [20] studied the sum rate maximization problem of a downlink multiuser multiple⁃input single⁃output (MISO) NOMA system. The MISO NOMA transmission outper⁃ forms conventional OMA schemes, particularly when the trans⁃ mit SNR is low, and the number of users is greater than the number of BS antennas. Recently, a multiuser MIMO⁃NOMA scheme based on limited feedback was proposed and analyzed in [50]. In summary, most of the existing works on MIMO⁃NOMA fo⁃ cused on design of precoding and detection algorithms, and their performance analyses. However, user scheduling and power allocation were rarely discussed in the spatial domain, which play important roles in improving the spatial efficiency of MIMO⁃NOMA. 4.3 Other Works on NOMA In addition to the above two aspects, there are many other works associated NOMA. We will not discuss further in detail due to the limited space. Compared to downlink NOMA, up⁃ link NOMA was also studied in several works [51]-[57]. More⁃ over, asynchronous NOMA has also been investigated in up⁃ link scenarios [58], [59]. Cooperative NOMA, where strong us⁃ ers serve as relays for weak users, was studied in [60], [61]. In addition, several works on NOMA combined with other tech⁃ niques were also reported, such as energy harvesting [62], [63], cognitive radio networks [64], visible light communication [65], and physical layer security [66]. 5 Research Challenges As discussed above, NOMA can be employed to improve the spectral efficiency, user fairness, as well as to support massive connections with diverse QoS requirements. Based on our over⁃ view of existing works on NOMA and its potential applications in practical systems, we present the research challenges of NO⁃ MA in the following three aspects. 5.1 Resource Allocation under Imperfect CSI Most of existing works on resource allocation of NOMA are based on the assumption of perfect CSI at the transmitter side, which is difficult to obtain in practice due to either the estima⁃ tion error or the feedback delay. Therefore, it is nature to inves⁃ tigate how CSI error affects the performance of NOMA and to consider robust resource allocation under imperfect CSI. Since NOMA is expected to offer lower latency in order to support de⁃ lay⁃sensitive applications in 5G, one promising solution is the outage⁃based robust approach for designing the resource allo⁃ cation of NOMA. In this direction, the SIC decoding order un⁃ der imperfect CSI is still an open problem. Furthermore, it is important to study the joint optimization of power allocation, user scheduling and SIC decoding order selection of NOMA under imperfect CSI. 5.2 Cooperative NOMA A key feature of NOMA is that the strong users have prior in⁃ formation of the weak users, which has not been fully exploited in existing works. In cooperative NOMA, the strong users can serve as relays for the weak users, which has the potential to utilize the spatial DoF even for users with a single antenna. Some preliminary works showed that cooperative NOMA can achieve the maximum diversity gain for all the users [60], [61]. It is important to study the optimal resource allocation for coop⁃ A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo SIC: successive interference cancellation ▲Figure 5. A downlink MIMO⁃NOMA model with one base station and four users. Base station User 3 User 1 User 2 SIC of user 2’s signal User 1’s signal decoding User 2’s signal decoding SIC of user 3’s signal User 3’s signal decoding User 4’s signal decoding User 4 Power Power … … Frequency Frequency … …
  • 27. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 23 Special Topic erative NOMA. Besides, distributed beamforming can be em⁃ ployed in cooperative NOMA to harvest the spatial DoF with⁃ out much signalling overhead. Considering that cooperative NOMA will introduce more complexity and extra delay into sys⁃ tems, it is important to investigate the tradeoffs among the sys⁃ tem performance, complexity, and delay. 5.3 QoS􀆼Based NOMA As we mentioned before, NOMA has great potential to sup⁃ port diverse QoS requirements. The heterogeneity of QoS re⁃ quirements might in turn facilitate the power allocation and us⁃ er scheduling of NOMA, which is also an interesting topic to explore in the future. For example, users in NOMA systems can be categorized according to their QoS requirements, in⁃ stead of their channel conditions, which offers two following benefits. One is that the SIC decoding order, power allocation, and user scheduling can be designed more appropriately to meet the users QoS requests. The other is to make NOMA com⁃ munications more general, e.g., applicable to scenarios in which users channel conditions are the same. 6 Conclusions In this article, a promising multiple access technology for 5G networks, NOMA, is discussed. A two ⁃ user SC ⁃ NOMA scheme and a multiuser MC ⁃ NOMA scheme were presented and discussed to illustrate the basic concepts and principles of NOMA. A literature review about performance analyses of NO⁃ MA, resource allocation for NOMA, and MIMO⁃NOMA was dis⁃ cussed. Furthermore, we presented the key features and poten⁃ tial research challenges of NOMA. A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo References [1] J. Andrews, S. Buzzi, W. Choi, et al.,“What will 5G be?”IEEE Journal on Se⁃ lected Areas in Communications, vol. 32, no. 6, pp. 1065-1082, Jun. 2014. doi: 10.1109/JSAC.2014.2328098. [2] G. Wunder, P. Jung, M. Kasparick, et al.,“5GNOW: non⁃orthogonal, asynchro⁃ nous waveforms for future mobile applications,”IEEE Communications Maga⁃ zine, vol. 52, no. 2, pp. 97-105, Feb. 2014. doi: 10.1109/MCOM.2014.6736749. [3] T. Marzetta,“Noncooperative cellular wireless with unlimited numbers of base station antennas,”IEEE Transactions on Wireless Communications, vol. 9, no. 11, pp. 3590-3600, Nov. 2010. doi: 10.1109/TWC.2010.092810.091092. [4] J. Zhu, D. W. K. Ng, N. Wang, R. Schober, and V. K. Bhargava,“Analysis and design of secure massive MIMO systems in the presence of hardware impair⁃ ments,”arXiv preprint arXiv:1602.08534, 2016. [5] Z. Pi and F. Khan,“An introduction to millimeter⁃wave mobile broadband sys⁃ tems,”IEEE Communications Magazine, vol. 49, no. 6, pp. 101-107, Jun. 2011. doi: 10.1109/MCOM.2011.5783993. [6] T. Rappaport, S. Sun, R. Mayzus, et al.,“Millimeter wave mobile communica⁃ tions for 5G cellular: It will work!”IEEE Access, vol. 1, pp. 335-349, May 2013. doi: 10.1109/ACCESS.2013.2260813. [7] J. Andrews, H. Claussen, M. Dohler, S. Rangan, and M. Reed,“Femtocells: past, present, and future,”IEEE Journal on Selected Areas in Communications, vol. 30, no. 3, pp. 497-508, Apr. 2012. doi: 10.1109/JSAC.2012.120401. [8] J. Andrews,“Seven ways that HetNets are a cellular paradigm shift,”IEEE Com⁃ munications Magazine, vol. 51, no. 3, pp. 136-144, Mar. 2013. doi: 10.1109/ MCOM.2013.6476878. [9] D. Ramasamy, R. Ganti, and U. Madhow,“On the capacity of picocellular net⁃ works,”in IEEE International Symposium on Information Theory, Istanbul, Tur⁃ key, Jul. 2013, pp. 241-245. doi: 10.1109/ISIT.2013.6620224. [10] D. W. K. Ng, E. S. Lo, and R. Schober,“Robust beamforming for secure com⁃ munication in systems with wireless information and power transfer,”IEEE Transactions on Wireless Communications, vol. 13, no. 8, pp. 4599-4615, Aug. 2014. doi: 10.1109/TWC.2014.2314654. [11] D. W. K. Ng, E. S. Lo, and R. Schober,“Energy⁃efficient resource allocation in OFDMA systems with large numbers of base station antennas,”IEEE Transac⁃ tions on Wireless Communications, vol. 11, no. 9, pp. 3292-3304, Sept. 2012. doi: 10.1109/TWC.2012.072512.111850. [12] D. W. K. Ng, E. S. Lo, and R. Schober,“Wireless information and power trans⁃ fer: Energy efficiency optimization in OFDMA systems,”IEEE Transactions on Wireless Communications, vol. 12, no. 12, pp. 6352- 6370, Dec. 2013. doi: 10.1109/TWC.2013.103113.130470. [13] L. Dai, B. Wang, Y. Yuan, S. Han, I. Chih⁃Lin, and Z. Wang,“Non⁃orthogonal multiple access for 5G: solutions, challenges, opportunities, and future re⁃ search trends,”IEEE Communications Magazine, vol. 53, no. 9, pp. 74- 81, Sept. 2015. doi: 10.1109/MCOM.2015.7263349. [14] J. M. Meredith,“Study on downlink multiuser superposition transmission (MUST) for LTE (Release 13),”3GPP Tech. Rep. TR 36.859, Dec. 2015. [15] R. Hoshyar, F. P. Wathan, and R. Tafazolli,“Novel low⁃density signature for synchronous CDMA systems over AWGN channel,”IEEE Transactions on Sig⁃ nal Processing, vol. 56, no. 4, pp. 1616- 1626, Apr. 2008. doi: 10.1109/ TSP.2007.909320. [16] R. Hoshyar, R. Razavi, and M. Al⁃Imari,“LDS⁃OFDM an efficient multiple ac⁃ cess technique,”in IEEE Vehicular Technology Conference, Taiwan, China, May 2010, pp. 1-5. doi: 10.1109/VETECS.2010.5493941. [17] R. Razavi, M. Al⁃Imari, M. A. Imran, R. Hoshyar, and D. Chen,“On receiver design for uplink low density signature OFDM (LDS⁃OFDM),”IEEE Transac⁃ tions on Communications, vol. 60, no. 11, pp. 3499- 3508, Nov. 2012. doi: 10.1109/TCOMM.2012.082812.110284. [18] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE Personal, Indoor and Mobile Radio Communications Symposium, London, United King⁃ dom, Sept. 2013, pp. 332-336. doi: 10.1109/PIMRC.2013.6666156. [19] X. Dai, S. Chen, S. Sun, et al.,“Successive interference cancelation amenable multiple access (SAMA) for future wireless communications,”in IEEE Interna⁃ tional Conference on Communication Systems, Macau, China, Nov. 2014, pp. 222-226. doi: 10.1109/PIMRC.2013.6666156. [20] M. F. Hanif, Z. Ding, T. Ratnarajah, and G. K. Karagiannidis,“A minorization⁃ maximization method for optimizing sum rate in the downlink of non⁃orthogonal multiple access systems,”IEEE Transactions on Signal Processing, vol. 64, no. 1, pp. 76-88, Jan. 2016. doi: 10.1109/TSP.2015.2480042. [21] Y. Sun, D. W. K. Ng, Z. Ding, and R. Schober,“Optimal joint power and sub⁃ carrier allocation for MC⁃NOMA systems,”arXiv preprint, arXiv:1603.08132, 2016. [22] Q. Sun, S. Han, C.⁃L. I, and Z. Pan,“Energy efficiency optimization for fading MIMO non⁃orthogonal multiple access systems,”in IEEE International Confer⁃ ence on Communications, London, United Kingdom, Jun. 2015, pp. 2668- 2673. doi: 10.1109/ICC.2015.7248728. [23] S. Timotheou and I. Krikidis,“Fairness for non⁃orthogonal multiple access in 5G systems,”IEEE Signal Processing Letters, vol. 22, no. 10, pp. 1647-1651, Oct. 2015. doi: 10.1109/LSP.2015.2417119. [24] A. Benjebbour, A. Li, Y. Saito, Y. Kishiyama, A. Harada, and T. Nakamura, “System⁃level performance of downlink NOMA for future LTE enhancements,” in IEEE Global Communications Conference, Atlanta, USA, Dec. 2013, pp. 66- 70. doi: 10.1109/GLOCOMW.2013.6824963. [25] Y. Saito, A. Benjebbour, Y. Kishiyama, and T. Nakamura,“System⁃level perfor⁃ mance evaluation of downlink non ⁃ orthogonal multiple access (NOMA),”in IEEE Personal, Indoor and Mobile Radio Communications Symposium, London, United Kingdom, Sept. 2013, pp. 611- 615. doi: 10.1109/GLO⁃ COMW.2013.6824963. [26] Z. Ding, Z. Yang, P. Fan, and H. Poor,“On the performance of non⁃orthogonal multiple access in 5G systems with randomly deployed users,”IEEE Signal Processing Letters, vol. 21, no. 12, pp. 1501-1505, Dec. 2014. doi: 10.1109/ LSP.2014.2343971. [27] P. Xu, Z. Ding, X. Dai, and H. V. Poor,“A new evaluation criterion for non⁃or⁃ thogonal multiple access in 5G software defined networks,”IEEE Access, vol. 3, pp. 1633-1639, Sept. 2015. doi: 10.1109/ACCESS.2015.2480117. [28] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Cam⁃
  • 28. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS24 Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo bridge, United Kingdom: Cambridge University Press, 2005. [29] N. Otao, Y. Kishiyama, and K. Higuchi,“Performance of non⁃orthogonal access with SIC in cellular downlink using proportional fair ⁃ based resource alloca⁃ tion,”in IEEE International Symposium on Wireless Communication Systems, Paris, France, Aug. 2012, pp. 476-480. doi: 10.1109/ISWCS.2012.6328413. [30] Y. Saito, Y. Kishiyama, A. Benjebbour, et al.,“Non⁃orthogonal multiple access (NOMA) for cellular future radio access,”in IEEE Vehicular Technology Con⁃ ference, Dresden, Germany, Jun. 2013, pp. 1- 5. doi: 10.1109/VTC⁃ Spring.2013.6692652. [31] Z. Ding, P. Fan, and V. Poor,“Impact of user pairing on 5G non⁃orthogonal multiple access downlink transmissions,”IEEE Transactions on Vehicular Technology, vol. 65, no. 8, pp. 6010- 6023, 2016. doi: 10.1109/ TVT.2015.2480766. [32] Z. Yang, Z. Ding, P. Fan, and G. K. Karagiannidis,“On the performance of non⁃ orthogonal multiple access systems with partial channel information,”IEEE Transactions on Communications, vol. 64, no. 2, pp. 654-667, Feb. 2016. doi: 10.1109/TCOMM.2015.2511078. [33] M. R. Hojeij, J. Farah, C. Nour, and C. Douillard,“Resource allocation in downlink non⁃orthogonal multiple access (NOMA) for future radio access,”in IEEE Vehicular Technology Conference, Glasgow, Scotland, May 2015, pp. 1- 6. doi: 10.1109/VTCSpring.2015.7146056. [34] M. R. Hojeij, J. Farah, C. Nour, and C. Douillard,“New optimal and subopti⁃ mal resource allocation techniques for downlink non⁃orthogonal multiple ac⁃ cess,”Wireless Personal Communications, vol. 87, no. 3, pp. 837- 867, May 2015. doi:10.1007/s11277⁃015⁃2629⁃2. [35] B. Di, S. Bayat, L. Song, and Y. Li,“Radio resource allocation for downlink non ⁃orthogonal multiple access (NOMA)networks using matching theory,”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1-6. doi: 10.1109/GLOCOM.2015.7417643. [36] L. Lei, D. Yuan, C. K. Ho, and S. Sun,“Joint optimization of power and chan⁃ nel allocation with non⁃orthogonal multiple access for 5G cellular systems,”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1- 6. doi: 10.1109/GLOCOM.2015.7417761. [37] Y.⁃F. Liu and Y.⁃H. Dai,“On the complexity of joint subcarrier and power allo⁃ cation for multi⁃user OFDMA systems,”IEEE Transactions on Signal Process⁃ ing, vol. 62, no. 3, pp. 583-596, Feb. 2014. doi: 10.1109/TSP.2013.2293130. [38] H. Kim, K. Kim, Y. Han, and S. Yun,“A proportional fair scheduling for multi⁃ carrier transmission systems,”in IEEE Vehicular Technology Conference, Los Angeles, USA, Sept 2004, vol. 1, pp. 409- 413. doi: 10.1109/ VETECF.2004.1400034. [39] C. Wengerter, J. Ohlhorst, and A. G. E. von Elbwart,“Fairness and throughput analysis for generalized proportional fair frequency scheduling in OFDMA,”in IEEE Vehicular Technology Conference, Stockholm, Sweden, May 2005, vol. 3, pp. 1903-1907. doi: 10.1109/VETECS.2005.1543653. [40] F. Liu, P. Mahonen, and M. Petrova,“Proportional fairness⁃based user pairing and power allocation for non⁃orthogonal multiple access,”in IEEE Personal, In⁃ door and Mobile Radio Communications Symposium, Hong Kong, China, Aug. 2015, pp. 1127-1131. doi: 10.1109/PIMRC.2015.7343467. [41] S. Shi, L. Yang, and H. Zhu,“Outage balancing in downlink non⁃orthogonal multiple access with statistical channel state information,”IEEE Transactions on Wireless Communications, vol. 15, no. 7, pp. 4718-4731, Jul. 2016. doi: 10.1109/TWC.2016.2544922. [42] Y. Lan, A. Benjebboiu, X. Chen, A. Li, and H. Jiang,“Considerations on down⁃ link non⁃orthogonal multiple access (NOMA) combined with closed⁃loop SU⁃ MIMO,”in IEEE International Conference on Signal Processing and Communi⁃ cation System, Gold Coast, Australia, Dec. 2014, pp. 1- 5. doi: 10.1109/IC⁃ SPCS.2014.7021086. [43] X. Chen, A. Benjebbour, Y. Lan, A. Li, and H. Jiang,“Impact of rank optimiza⁃ tion on downlink non⁃orthogonal multiple access (NOMA) with SU⁃MIMO,”in IEEE International Conference on Communication Systems, Macau, China, Nov. 2014, pp. 233-237. doi: 10.1109/ICCS.2014.7024800. [44] B. Kim, S. Lim, H. Kim, et al.,“Non⁃orthogonal multiple access in a downlink multiuser beamforming system,”in IEEE Military Communications Conference, San Diego, USA, Nov. 2013, pp. 1278-1283. doi: 10.1109/MILCOM.2013.218. [45] J. Choi,“Minimum power multicast beamforming with superposition coding for multiresolution broadcast and application to NOMA systems,”IEEE Transac⁃ tions on Communications, vol. 63, no. 3, pp. 791-800, Mar. 2015. doi: 10.1109/ TCOMM.2015.2394393. [46] Q. Sun, S. Han, C.⁃L. I, and Z. Pan,“On the ergodic capacity of MIMO NOMA systems,”IEEE Wireless Communications Letters, vol. 4, no. 4, pp. 405-408, Aug. 2015. doi: 10.1109/LWC.2015.2426709. [47] Q. Sun, S. Han, Z. Xu, S. Wang, C.⁃L. I, and Z. Pan,“Sum rate optimization for MIMO non⁃orthogonal multiple access systems,”in IEEE Wireless Communica⁃ tions and Networking Conference, New Orleans, USA, Mar. 2015, pp. 747-752. doi: 10.1109/WCNC.2015.7127563. [48] Z. Ding, F. Adachi, and H. V. Poor,“The application of MIMO to non⁃orthogo⁃ nal multiple access,”IEEE Transactions on Wireless Communications, vol. 15, no. 1, pp. 537-552, Jan. 2016. doi: 10.1109/TWC.2015.2475746. [49] Z. Ding, R. Schober, and H. V. Poor,“A general MIMO framework for NOMA downlink and uplink transmission based on signal alignment,”IEEE Transac⁃ tions on Wireless Communications, vol. 15, no. 6, pp. 4438-4454, Jun. 2016. doi: 10.1109/TWC.2016.2542066. [50] Z. Ding and H. V. Poor,“Design of massive⁃MIMO⁃NOMA with limited feed⁃ back,”IEEE Signal Processing Letters, vol. 23, no. 5, pp. 629-633, May 2016. doi: 10.1109/LSP.2016.2543025. [51] T. Takeda and K. Higuchi,“Enhanced user fairness using non⁃orthogonal ac⁃ cess with SIC in cellular uplink,”in IEEE Vehicular Technology Conference, San Francisco, USA, Sept. 2011, pp. 1- 5. doi: 10.1109/ VETECF.2011.6093272. [52] M. Al⁃Imari, P. Xiao, M. A. Imran, and R. Tafazolli,“Uplink non⁃orthogonal multiple access for 5G wireless networks,”in IEEE International Symposium on Wireless Communications Systems, Barcelona, Spain, Aug. 2014, pp. 781- 785. doi: 10.1109/ISWCS.2014.6933459. [53] S. Chen, K. Peng, and H. Jin,“A suboptimal scheme for uplink NOMA in 5G systems,”in IEEE International Wireless Communications and Mobile Comput⁃ ing Conference, Dubrovnik, Croatia, Aug. 2015, pp. 1429-1434. doi: 10.1109/ IWCMC.2015.7289292. [54] Y. Chen, J. Schaepperle, and T. Wild,“Comparing IDMA and NOMA with su⁃ perimposed pilots based channel estimation in uplink,”in IEEE Personal, In⁃ door and Mobile Radio Communications Symposium, Aug. 2015, pp. 89- 94. doi: 10.1109/PIMRC.2015.7343274. [55] M. Al⁃Imari, P. Xiao, and M. A. Imran,“Receiver and resource allocation opti⁃ mization for uplink NOMA in 5G wireless networks,”in IEEE International Symposium on Wireless Communication Systems, Brussels, Belgium, Aug. 2015. doi: 10.1109/ISWCS.2015.7454317. [56] B. Kim, W. Chung, S. Lim, et al.,“Uplink NOMA with multi ⁃ antenna,”in IEEE Vehicular Technology Conference, Glasgow, Scotland, May 2015, pp. 1- 5. doi: 10.1109/VTCSpring.2015.7146149. [57] N. Zhang, J. Wang, G. Kang, and Y. Liu,“Uplink non⁃orthogonal multiple ac⁃ cess in 5G systems,”IEEE Communications Letters, vol. 20, no. 3, pp. 458- 461, Mar. 2016. doi: 10.1109/LCOMM.2016.2521374. [58] H. Haci, H. Zhu, and J. Wang,“A novel interference cancellation technique for non ⁃ orthogonal multiple access (NOMA),”in IEEE Global Communications Conference, San Diego, USA, Dec. 2015, pp. 1- 6. doi: 10.1109/GLO⁃ COM.2015.7417128. [59] H. Haci,“Non⁃orthogonal multiple access (NOMA) with asynchronous interfer⁃ ence cancellation,”Ph.D. dissertation, Dept. Electron. Eng., Univ. of Kent, Kent, England, Mar. 2015. [60] Z. Ding, M. Peng, and H. Poor,“Cooperative non⁃orthogonal multiple access in 5G systems,”IEEE Communications Letters, vol. 19, no. 8, pp. 1462- 1465, Aug. 2015. doi: 10.1109/LCOMM.2015.2441064. [61] Z. Ding, H. Dai, and H. V. Poor,“Relay selection for cooperative NOMA,” IEEE Wireless Communications Letters, vol. 5, no. 4, pp. 416-419, Jun. 2016. doi: 10.1109/LWC.2016.2574709. [62] Y. Liu, Z. Ding, M. Eikashlan, and H. Poor,“Cooperative non⁃orthogonal multi⁃ ple access in 5G systems with SWIPT,”in European Signal Processing Confer⁃ ence, Nice, France, Aug. 2015, pp. 1999- 2003. doi: 10.1109/EUSIP⁃ CO.2015.7362734. [63] P. D. Diamantoulakis, K. N. Pappi, Z. Ding, and G. K. Karagiannidis,“Optimal design of non⁃orthogonal multiple access with wireless power transfer,”arXiv preprint arXiv:1511.01291, 2015. [64] Y. Liu, Z. Ding, M. Elkashlan, and J. Yuan,“Non⁃orthogonal multiple access in large⁃scale underlay cognitive radio networks,”IEEE Transactions on Vehicu⁃ lar Technology, vol. PP, no. 99, pp. 1- 1, Feb. 2016. doi: 10.1109/ TVT.2016.2524694. [65] H. Marshoud, V. M. Kapinas, G. K. Karagiannidis, and S. Muhaidat,“Non⁃or⁃ thogonal multiple access for visible light communications,”IEEE Photonics Technology Letters, vol. 28, no. 1, pp. 51- 54, Jan. 2016. doi: 10.1109/ LPT.2015.2479600. [66] Y. Zhang, H. M. Wang, Q. Yang, and Z. Ding,“Secrecy sum rate maximization in non⁃orthogonal multiple access,”IEEE Communications Letters, vol. 20, no. 5, pp. 930-933, Mar. 2016. doi: 10.1109/LCOMM.2016.2539162. Manuscript received: 2016⁃08⁃15
  • 29. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 25 Special Topic A Survey of Downlink Non⁃Orthogonal Multiple Access for 5G Wireless Communication Networks WEI Zhiqiang, YUAN Jinhong, Derrick Wing Kwan Ng, Maged Elkashlan, and DING Zhiguo WEI Zhiqiang (zhiqiang.wei@unsw.edu.au) received the BE degree from Northwes⁃ tern Polytechnical University, China in 2012. He is currently pursuing the PhD de⁃ gree in Wireless Communications Laboratory, University of New South Wales, Aus⁃ tralia. His research interests include non⁃orthogonal multiple access and resource allocation. YUAN Jinhong (j.yuan@unsw.edu.au) received the BE and PhD degrees in electro⁃ nics engineering from Beijing Institute of Technology, China in 1991 and 1997, re⁃ spectively. From 1997 to 1999, he was a reasearch fellow with the School of Electri⁃ cal Engineering, University of Sydney, Australia. In 2000, he joined the School of Electrical Engineering and Telecommunications, University of New South Wales, Australia, where he is currently a professor of telecommunications. He has authored two books, three book chapters, more than 200 papers in telecommunications jour⁃ nals and conference proceedings, and 40 industrial reports. His research interests include error control coding and information theory, communication theory, and wireless communications. He is a co⁃inventor of one patent on MIMO systems and two patents on low⁃density parity⁃check codes. He is currently serving as an associ⁃ ate editor for the IEEE TCOM. He served as the IEEE NSW Chair of Joint Commu⁃ nications/Signal Processions/Ocean Engineering Chapter from 2011 to 2014. He was the co⁃recipient of three best paper awards and one best poster award, includ⁃ ing the Best Paper Award from the IEEE Wireless Communications and Networking Conference, Cancun, Mexico in 2011, and the Best Paper Award from the IEEE In⁃ ternational Symposium on Wireless Communications Systems, Trondheim, Norway in 2007. Derrick Wing Kwan Ng (w.k.ng@unsw.edu.au) received the bachelor degree with first class honors and the Master of Philosophy (M.Phil.) degree in electronic engi⁃ neering from the Hong Kong University of Science and Technology (HKUST) in 2006 and 2008, respectively. He received his PhD degree from the University of British Columbia (UBC) in 2012. He was a senior postdoctoral fellow at the Institute for Digital Communications, University of Erlangen ⁃ Nuremberg, Germany. He is now working as a lecturer at the University of New South Wales, Australia. Dr. Ng has published more than 80 journal and conference papers and his publications have been cited over 2000 times in Google Scholar with an h⁃index of 20. Dr. Ng is currently an editor of IEEE Communications Letters and IEEE Transactions on Green Communications and Networking. He served as a Co⁃Chair for the Wireless Access Track of 2014 IEEE 80th Vehicular Technology Conference and 2016 IEEE GlobeCom Workshop on Wireless Energy Harvesting. He was also a co⁃organizer and guest editor of the special issue on Energy Harvesting Wireless Communica⁃ tions in EURASIP Journal on Wireless Communications and Networking in 2014. Maged Elkashlan (maged.elkashlan@qmul.ac.uk) received the PhD degree in electr⁃ ical engineering from the University of British Columbia, Canada in 2006. From 2007 to 2011, he was with the Wireless and Networking Technologies Laboratory, Commonwealth Scientific and Industrial Research Organization, Australia. During this time, he held an adjunct appointment with the University of Technology Syd⁃ ney, Australia. In 2011, he joined the School of Electronic Engineering and Comput⁃ er Science, Queen Mary University of London, U.K. He currently holds visiting fac⁃ ulty appointments with the University of New South Wales, Australia, and the Bei⁃ jing University of Posts and Telecommunications, China. His research interests fall into the broad areas of communication theory, wireless communications, and statisti⁃ cal signal processing for distributed data processing, heterogeneous networks, and massive MIMO. Dr. Elkashlan received the best paper award at the IEEE Interna⁃ tional Conference on Communications in 2014, the International Conference on Communications and Networking in China in 2014, and the IEEE Vehicular Tech⁃ nology Conference in 2013. He also received the Exemplary Reviewer Certificate of the IEEE CL in 2012. He serves as an editor of IEEE TWC, IEEE TVT, and IEEE CL. He also serves as a lead guest editor of the Special Issue on Green Media: The Future of Wireless Multimedia Networks of the IEEE Wireless Communications Mag⁃ azine and the Special Issue on Millimeter Wave Communications for 5G of the IEEE Communications Magazine, and a guest editor of the Special Issue on Energy Har⁃ vesting Communications of the IEEE Communications Magazine and the Special Is⁃ sue on Location Awareness for Radios and Networks of the IEEE JSAC. DING Zhiguo (z.ding@lancaster.ac.uk) received his BEng from the Beijing Univers⁃ ity of Posts and Telecommunications, China in 2000, and the PhD degree from Im⁃ perial College London, U.K. in 2005. From Jul. 2005 to Aug. 2014, he was working in Queen’s University Belfast, Imperial College and Newcastle University. Since Sept. 2014, he has been with Lancaster University as a Chair Professor in Signal Processing. From Sept. 2012 to Sept. 2017, he has also been an academic visitor in Princeton University. Dr Ding’s research interests are 5G networks, game theory, cooperative and energy harvesting networks and statistical signal processing. He is serving as an editor for IEEE TCOM, IEEE TVT, IEEE WCL, and IEEE CL. He was the TPC Co⁃Chair for ICWMMN2015, and Symposium Chair for ICNC 2016 and WOCC 2015. He received the best paper award in ICWOC 2009 and WCSP 2015, IEEE CL Exemplary Reviewer 2012, and the EU Marie Curie Fellowship 2012 ⁃ 2014. BiographiesBiographies
  • 30. Unified Framework Towards Flexible MultipleUnified Framework Towards Flexible Multiple Access Schemes forAccess Schemes for 55GG SUN Qi, WANG Sen, HAN Shuangfeng, and Chih􀆼Lin I (China Mobile Research Institute, Beijing 100032, China) Abstract Non⁃orthogonal multiple access (NOMA) schemes have achieved great attention recently and been considered as a crucial compo⁃ nent for 5G wireless networks since they can efficiently enhance the spectrum efficiency, support massive connections and poten⁃ tially reduce access latency via grant free access. In this paper, we introduce the candidate NOMA solutions in 5G networks, com⁃ paring the principles, key features, application scenarios, transmitters and receivers, etc. In addition, a unified framework of these multiple access schemes are proposed to improve resource utilization, reduce the cost and support the flexible adaptation of multi⁃ ple access schemes. Further, flexible multiple access schemes in 5G systems are discussed. They can support diverse deployment scenarios and traffic requirements in 5G. Challenges and future research directions are also highlighted to shed some lights for the standardization in 5G. 5G; non⁃orthogonal multiple access; unified framework; flexible multiple access Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 004 http://www.cnki.net/kcms/detail/34.1294.TN.20161018.1014.002.html, published online October 18, 2016 1 Introduction orldwide initiatives on the 5th generation (5G) wireless communication have been extensive⁃ ly carried out, starting with an investigation on user demands, scenarios, key performance indicators (KPIs) and enabling technologies. A global consen⁃ sus is first forming that 5G network will be able to sustainably support 1000⁃fold mobile data traffic growth, improve energy efficiency (EE) and cost efficiency by over 100 times, provide fiber link access data rates and“zero”latency user experi⁃ ence, and be capable of connecting 100 billion devices and ca⁃ pable of delivering a consistent experience across a variety of scenarios including the cases of ultra⁃high traffic volume densi⁃ ty, ultra⁃high connection density and ultra⁃high mobility [1]. Three typical usage scenarios of 5G are also identified: en⁃ hanced mobile broadband (eMBB), massive machine type com⁃ munication (mMTC) and ultra ⁃ reliable low latency machine type communication (URLLC), targeting different 5G capabili⁃ ties. Beyond that, the standardization organizations, e.g. 3GPP has started the new research on 5G, studying the new access technology to meet a broad range of use cases. Multiple access schemes, the most fundamental aspect of the physical layer, to a large extent, are considered as the de⁃ fining technical feature of each wireless communication gener⁃ ation and have continually evolved in each cellular generation from frequency division multiple access (FDMA), time division multiple access (TDMA) in 1G and 2G to code division multi⁃ ple access (CDMA) in 3G and orthogonal frequency ⁃ division multiple access/single⁃carrier FDMA (OFDMA/SC⁃FDMA) for 4G. Facing the stringent demands of diverse scenarios in 5G, e. g., 1000x higher data rates, massive uplink connectivity and low access latency, the traditional pure orthogonal multiple ac⁃ cess is not a good option. Some alternative non⁃orthogonal mul⁃ tiple access schemes have attracted considerable attention and been identified as a crucial technology component in 5G since they can serve multiple users in the same frequency and time resources via code domain multiplexing and/or power domain multiplexing to enhance system access performance. The non⁃ orthogonal multiple access schemes are potentially able to sup⁃ port massive connections, improve spectrum efficiency and al⁃ so reduce access latency via the grant free access. Currently, some potential alternative multiple access schemes are being actively studied in 3GPP for 5G, including superposition cod⁃ ing based non ⁃ orthogonal multiple access (SPC ⁃ NOMA) [2], multi user shared access (MUSA) [3], sparse code multiple ac⁃ cess (SCMA) [4], pattern division multiple access (PDMA) [5], resource spread multiple access (RSMA) [6], non ⁃ orthogonal coded multiple access (NCMA) [7], and interleave⁃grid multi⁃ ple access (IGMA) [8]. In this paper, the principles, advantages and application sce⁃ narios of different multiple access techniques are discussed W Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS26
  • 31. and compared. In addition, we introduce a unified framework that can merge a wide range of multiple access techniques, which helps to minimize the hardware functional module. Based on the unified framework, some initial work on flexible multiple access schemes is also introduced. Finally, the chal⁃ lenges and future directions are discussed. 2 Candidate Non⁃Orthogonal Multiple Access Solutions In this section, we introduce the typical candidate NOMA so⁃ lutions for 5G, which can be basically divided into three cate⁃ gories, i.e., the power domain based, code domain based and interleaver based. Their principles and key features are dis⁃ cussed. At last, we provide their comparison in terms of appli⁃ cation scenarios, system performance, receivers, etc. 2.1 Power Domain Based Solutions 2.1.1 SPC⁃NOMA NOMA based on superposition coding utilizes power domain for user multiplexing and can be applied for both downlink and uplink. Established by network information theory, non⁃orthog⁃ onal access with successive interference cancellation (SIC)/ dirty paper coding (DPC) can achieve the multiuser capacity region both in uplink and downlink. NOMA superposes multi⁃ ple users in power ⁃ domain and exploits channel gain differ⁃ ence between the multiplexed users with the aid of advanced receiver, e.g. the SIC receiver, for user separation. Fig. 1 shows signal transmission and receiving in downlink NOMA system with two users. Currently the NOMA technique is being discussed in the 3GPP under the study item of“study on down⁃ link multiuser superposition transmission (MUST)”for release 13 [9]. For the study in 3GPP, the study scope of NOMA is very limited, e.g. only about downlink transmission, only for the intra⁃cell usage and only for data channels. For 5G system, there are more application scenarios of NO⁃ MA technique, such as uplink and control channel, and more advanced NOMA techniques, such as combination with sophis⁃ ticated multiple⁃input multiple⁃output (MIMO) techniques and inter ⁃ cell techniques. In [10]- [12], MIMO NOMA schemes have been studied. Network NOMA which considering the multi⁃cell scenarios are also studied from EE⁃SE co⁃design per⁃ spective in [13]. 2.2 Code Domain Based Solutions 2.2.1 MUSA MUSA is a non⁃orthogonal multiple access scheme operat⁃ ing in code domain. Conceptually, each user’s modulated data symbols are spread firstly by a specially designed sequence which facilitates robust SIC implementation compared to the sequences employed by traditional direct⁃sequence CDMA (DS ⁃CDMA ). Then, each user’s spread symbols are transmitted concurrently on the same radio resource by means of“Shared Access”, which is essentially a superposition process. Finally, decoding of each user’s data from superimposed signal can be performed at the base⁃station side using SIC technology. The major processing blocks of MUSA transmitter and re⁃ ceiver are illustrated in Fig. 2. Symbols of each user are spread by a spreading sequence. Multiple spreading sequences constitute a pool from which each user can randomly pick one. Note that for the same user, different spreading sequences may also be used to different symbols. This may further improve the performance via interference averaging. Then, all spreading symbols are transmitted over the same time⁃frequency resourc⁃ es. The spreading sequences should have low cross⁃correlation and can be non⁃binary. At the receiver, codeword level SIC is used to separate data from different users. The complexity of codeword level SIC is less of an issue in the uplink as the re⁃ ceiver anyway needs to decode the data for all users. The only noticeable impact on the receiver implementation would be that the pipeline of processing may be changed in order to per⁃ form SIC operation. MUSA relies on a special family of complex spread sequenc⁃ es that can enjoy relatively low cross ⁃ correlation even when they are very short, say, 8 or even 4. The real and imaginary parts of the complex spread sequence can be drawn from an M⁃ ary real value set. For example, for a 3⁃value set {⁃1, 0, 1}, ev⁃ SIC: successive interference cancellation ▲Figure 1. Illustration of SPC ⁃NOMA transmission. ▲Figure 2. An example of MUSA with four resources shared by multiple users [1]. SIC: successive interference cancellation Power User 1 User 2 Time/frequency /spatial resources Base station User 2 User 2 signal detection User 1 signal detection User 1 SIC of user 1 signal Data of user1 SIC Using SIC receiver to decode each user’s data Each user’s spread symbols can be transmitted simultaneously User1 … C1 C2 Cn Each user’s modulated data symbols are spread by a specially designed sequence Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 27 User2 User n Data of user2 Data of usern
  • 32. ery bit of the complex sequence is drawn from the constellation depicted in Fig. 3 with equal probability. It should be pointed out that the spread sequences used in MUSA are different from the spreading codes, in the sense that MUSA spreading does not have the low density property. Equipped with the well ⁃ optimized spreading sequence and state⁃of⁃the⁃art SIC technology, MUSA is capable of decou⁃ pling the multiuser mingled data even if those users are con⁃ tending to access the system. Potentially a large number of de⁃ vices are allowed to transmit data at their will, by randomly picking spread sequences, spread the data and send them. In other words, MUSA is suitable for the scenario where the up⁃ link transmissions are not tightly scheduled, and the grants for transmission are not signaled per user basis, and with a high overloading. The relaxed UL synchronization requirement for MUSA allows simple derivation of UL time from a DL synchro⁃ nization process, which can greatly cut down the battery con⁃ sumption. Lastly, the code domain superposition nature of MU⁃ SA can turn the near⁃far problem into a near⁃far advantage. The disparity in the received signal to noise ratio (SNR) across the simultaneously transmitting users can be exploited in MU⁃ SA to facilitate SIC. Tight transmit power control is no longer needed, which can further lower the device cost and its power consumption. 2.2.2 SCMA SCMA is a novel non ⁃ orthogonal multiple access scheme with sparse codebooks. The main idea of SCMA is to accommo⁃ date more users with limited resources and increase the total network throughput, without scarifying user experience, which can be overloaded to enable massive connectivity and support grant⁃free access. There are multiple layers in SCMA, which can be used for user multiplexing. Each layer has a predefined codebook, which consists of multiple codewords. The code⁃ words are composed of multi ⁃ dimensional complex symbols, and the codewords in the same codebook have the same sparse pattern. For each layer, the coded bits are directly mapped to codewords, which are selected from layer⁃specific SCMA code⁃ books. The codewords of different layers are overlaid in code and power domains and carried over shared time⁃frequency re⁃ sources. Typically, the layer multiplexing may become over⁃ loaded if the number of layers is more than the length of the codewords. Fig. 4 shows an example of bits to codewords mapping in a SCMA system. The codebook design of SCMA has been stud⁃ ied in [14]; it has been shown that with multi⁃dimensional con⁃ stellation, shaping and coding gain can be achieved. At the re⁃ ceiver, joint multiuser detection algorithms are needed. Due to the sparsity of the SCMA codeword structure, message passing algorithm (MPA) on factory graph with much lower complexity can be adopted to achieve a suboptimal performance. Some simplified algorithms are proposed [15]-[19] to further reduce the detection complexity. Besides the codebook and the receiver design, some other challenging issues of SCMA, e.g., the energy efficiency optimi⁃ zation, uplink grant free access, downlink multiuser transmis⁃ sion, and the multi⁃cell transmission based on SCMA have also been studied. The energy efficiency performance and optimiza⁃ tion of SCMA are investigated in [20] and [21]. In [22] and [23], uplink contention based grant⁃free access based on SC⁃ MA has been proposed for 5G radio access. [24] and [25] focus on the downlink multiuser SCMA (MU ⁃ SCMA) network. [24] theoretically derives the capacity for downlink Massive MIMO MU⁃SCMA systems. In [25], a weighted sum rate based user pairing and power sharing algorithm are introduced to the MU⁃ SCMA network. It shows that SCMA can significantly increase the downlink spectral efficiency of 5G wireless cellular net⁃ works. Further, SCMA has also been introduced into multi⁃cell transmission. SCMA based uplink inter⁃cell interference can⁃ cellation technique and open loop joint coordinated multiple point transmission are studied in [26] and [27], respectively. There are still many challenging issues for SCMA, which need to be solved in the future work. For example, the layer multiplexing in SCMA provides new degree⁃of⁃freedom for us⁃ er scheduling. The algorithms for user grouping and power allo⁃ cation need to be optimized. In addition, the combination of SCMA and MIMO can be further enhanced. 2.2.3 PDMA PDMA introduces reasonable diversity between multiple us⁃ ers to promote the capacity, which can obtain higher multiuser ◀Figure 3. The elements of the complex spreading sequence [3]. ▲Figure 4. Illustration of SCMA codebooks and the process of bit mapping [1]. R1-1 -1 1 0 I Layer 1 Layer 2 Layer 3 Layer 4 Layer 5 Layer 6 (1,1)(b1,b2) (1,0) (0,1) (0,0) (0,1) (1,1) Sparsity pattern G = 0 1 0 1 1 0 1 0 1 1 0 0 0 0 1 1 1 0 0 1 0 1 1 0 October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS28 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 33. multiplexing and diversity gain. It considers the joint design of the transmitter and the receiver based on the optimization point of view for multiuser communication system. At the trans⁃ mitter side, the non⁃orthogonal characteristic pattern is used to distinguish users based on the multiple signals domain (includ⁃ ing time, frequency and the space domain). At the receiver side, sub⁃optimal multiuser detection by General SIC based on the features of the user pattern is utilized. To alleviate the error propagation problem of the SIC receiv⁃ er, the pattern used in PDMA is generally designed to ensure unequal transmission diversity for each user. In this way, the identical diversity order can be achieved after detection. In⁃ spired by the idea of unequal transmission diversity and sparse coding, an example of pattern and the related resource map⁃ ping has been proposed (Fig. 5). In the example, a code can also be seen as a pattern, which is used to define sparse mapping from data to a group of re⁃ sources. The code could be represented by a binary vector. The dimension of the vector equals to the number of resources in a group. Each element in the vector corresponds to a re⁃ source in a resource group. A“1”means that data shall be mapped to the corresponding resource. Actually, the number of “1”in the code is defined as its transmission diversity order. A code matrix is constructed by all codes sharing on the same resource group. Assuming six users multiplexing on four re⁃ source elements (REs). The data for User 1 are mapped to all the four resources in the group, and the data for User 2 are mapped to the first three resources, etc. The order of transmis⁃ sion diversity of the six users is 4, 3, 2, 2, 1, and 1, which is ob⁃ viously quite different from the SCMA scheme where all the us⁃ ers bear the same transmission diversity. Generally, if N is the size of resource group (the row num⁃ ber of code matrix), there are 2N - 1 possible binary vectors for a code matrix. Assuming K is the column number deter⁃ mined based on overload factor, we can thus choose K pat⁃ terns out from 2N - 1 candidates to construct code matrix. Se⁃ lection of codes also gives impacts on performance and com⁃ plexity. 2.2.4 RSMA RSMA combines the low rate channel code and the scram⁃ bling code (and optionally different interleavers) with good cor⁃ relation properties to separate different transmitters. In RSMA system, all users use the same frequency and time resources to transmit messages to the base station, regardless of the number of concurrent users. In other words, each user’s transmission power can be spread over all the available time and frequency resources. RSMA can be coupled with various waveforms/modulation schemes depending on the design target. Generally, it includes the single carrier RSMA and the multi⁃carrier RSMA. The sin⁃ gle carrier RSMA is optimized for battery power consumption and link budget extension by using single carrier waveforms. It allows grant⁃less transmission and potentially allows asynchro⁃ nous access. The grant⁃less transmission using RSMA reduces the signaling overhead, while the single carrier waveform fur⁃ ther reduces peak⁃to⁃average power ratio (PAPR) and achieves higher power amplifier efficiency. The pulse shaping block can further enhance the PAPR (e.g. potentially leading to constant envelope waveform), reducing out⁃of⁃band emission simultane⁃ ously. The multi⁃carrier RSMA is optimized for low latency ac⁃ cess, where reducing access delay is the design priority. It is suitable for the scenario where a connected state device is al⁃ ready synchronized to the base station and not link budget lim⁃ ited (e.g., close to the base station). Such a device can use RS⁃ MA with OFDM⁃based multi⁃carrier waveform for grant⁃less transmission to reduce overall access delay. 2.2.5 NCMA NCMA is a multiple access scheme based on the resource spreading by using non⁃orthogonal codewords, which is com⁃ posed of the codewords obtained by Grassmannian line pack⁃ ing problem [28]. To minimize the MUI theoretically, the spreading codes are designed with the minimum correlation. The non ⁃ orthogonal codebook is defined by C =[ ]c1 ⋯cK = é ë ê êê ê ù û ú úú ú c1 1 ⋯ cK 1 ⋮ ⋱ ⋮ c1 N ⋯ cK N ,C ∈ ℂN × K , where N is the sprea⁃ ding factor and K is the superposition factor. Then, the code⁃ book design problem can be posed in terms of maximizing the minimum chordal distance between codeword pairs minC æ è ç ö ø ÷max1 ≤ k ≤ j ≤ K 1 - |(ck )*.c j | where (ck )* is the conjugate cod⁃ eword of ck . NCMA can provide the additional throughput or improved connectivity with a small loss of block error rate (BLER) in spe⁃ cific environments, by exploiting additional layers through the superposed symbol, while satisfying QoS constraints. Since the receiver of NCMA system is available for parallel interference (b) resource mapping RE: resource element ▲Figure 5. Users sharing on four resource elements [5]. 1 G [4,6] Code = 1 1 1 1 1 1 0 1 0 1 0 0 1 0 1 0 0 1 0 0 0 0 1 (a) code matrix RE 1 RE 2 RE 3 RE 4 = + + + + + User 4User 3User 2User 1 User 5 User 6 Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 29 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 34. cancellation (PIC), the multiuser detection can be implement⁃ ed with low complexity. In addition, the MUI level between codeword pairs is always similar due to the correlation charac⁃ teristics mentioned above. Consequently, NCMA provides the potentials in terms of throughput or connectivity under special scenarios, e.g., huge connections with small packet in mMTC scenarios without changing the transmission block size, or for reducing the collision probability in contention based multiple access. 2.3 Interleaver Based Solution IGMA is an interleaver⁃based MA scheme, The typical trans⁃ mitter system structure using IGMA is shown in Fig. 6. Basi⁃ cally, the IGMA scheme distinguishes different users based on different bit⁃level interleavers, different grid mapping patterns and different combinations of bit ⁃ level interleaver and grid mapping pattern. Compared to the need of well⁃designed codewords or code sequences, the sufficient source of bit⁃level interleavers and/or grid mapping patterns are able to provide enough scalability for different connection densities, and also provide flexibility to achieve good balance between channel coding gain and ben⁃ efit from sparse resource mapping. By proper selection, the low correlated bit⁃level interleavers is achieved. In the grid map⁃ ping process, sparse mapping based on zero padding and sym⁃ bol⁃level interleaving is introduced, which provides another di⁃ mension for user multiplexing. Moreover, the density ρ of the grid mapping pattern is defined as the occupied RE numbers Nused dividing the total assigned RE numbers Nall , i.e. ρ = Nused /Nall . Different densities could be flexibly configured. It should be noted that the symbol sequence order is random⁃ ized after the grid mapping process due to symbol⁃level inter⁃ leaving, which may further bring benefit in terms of combating frequency selective fading and inter ⁃ cell interference, com⁃ pared to resource mapping using direct code ⁃ matrices/code⁃ books. At the receiver side, the low complexity multiuser detector (MUD) and the elementary signal estimator (ESE) that takes ad⁃ vantage of the special property of interleaving can be utilized with a simple de⁃mapping operation on the top. It should be noted that lower density of the grid mapping pattern further re⁃ duces detection complexity of ESE for IGMA. In addition, MAP and MPA detectors are also applicable for IGMA, which can improve the detection performance a lot comparing to ESE at the cost of complexity. The complexity of MAP/MPA for IG⁃ MA probably can be alleviated when spare grid mapping is used, due to the similar property of LDS. Fig. 7 shows an example of the grid mapping process of IG⁃ MA. The sparse symbol⁃to⁃RE mapping is performed based on an assigned grid mapping pattern. An exemplary operation can be mathematically formulated as a process by permutation ma⁃ trix αGM . According to the symbol⁃level interleave θk,2 assoc⁃ iated with the grid mapping pattern βk with density ρk(0 < ρk ≤ 1) , the corresponding permutation matrix αGM ∈ ℂN × L can be obtained. Thus, the kth user’s symbol s⁃ equence sk after zero padding and interleaving can be denoted by s' k = sk × αGM =[s' k,1,s' k,2,⋯,s' k,L ] , where L = N/ρk and ρk d⁃ ecides the number of zeros padded. 2.4 Summary of Multiple Access Techniques The pros and cons of the multiple access tech⁃ niques introduced above are summarized here in Table 1. It’s worth mentioning that some of these non⁃or⁃ thogonal schemes, such as SCMA MUSA and PD⁃ MA, can be implemented within a unified frame⁃ work, and each of them corresponds to a different codebook mapping module. In this way, the air interface can handover between different multiple access schemes in a flexi⁃ ble way, and all the other modules can be reused. This helps to improve the resource utilization and reduce the cost. In the fol⁃ lowing section, we will provide a unified framework for the mul⁃ tiple access schemes. 3 Unified Framework of Multiple Access Schemes Fig. 8 shows a unified framework of multiple access RF: Radio Frequency FEC: forward error correction ▲Figure 6. The IGMA transmitter [8]. βk = ì í î θk,2 ={4,0,2,0,0,3,0,1} ρk = 0.5 → αGM = é ë ê êê ê ù û ú úú ú 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 → s' k = sk × αGM = sk × αGM =[sk,4 0 sk,2 0 0 sk,3 0 sk,1] UE: user equipment ▲Figure 7. Example of the grid mapping process of IGMA when N = 4, ρk = 0.5 and L = 8 [8]. User data Channel coding FEC Repetition Modulation Bit⁃level interleaving Grid mapping Carrier modulation Baseband to RF Zero padding Sybol⁃level interleaving Zero⁃padding Symbol⁃level interleaving UE1 UE2 UE3 UE4 UEk f t + + + + … + = October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS30 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 35. schemes. The differences among these multiple access schemes lie in the different realization of interleaver, constella⁃ tion optimization, factor graph and multiplexing domain. The detailed explanations are listed in Table 2. 4 Flexible Multiple Access in 5G The above discussed advanced multiple access schemes as well as the traditional orthogonal multiple access scheme, e.g. OFDMA are all identified as potential candidates for 5G. There is no individual scheme can fulfill the requirements of all applications and scenarios in 5G system. A flexible adapta⁃ tion of these multiple access schemes is needed to support the diverse deployment scenarios and traffic requirements. For ex⁃ ample, in the case of massive connections, how to accommo⁃ date more users with limited resources has become a critical problem for next generation access network. With non⁃orthogo⁃ nal multiple access schemes, e.g., SCMA, MUSA, PDMA and RSMA, the same resources are shared and reused by multiple users, thus the number of connections increases. To support the traffic with low latency requirement, non⁃orthogonal multi⁃ ple access schemes help to realize grant⁃free multiple access, with which the latency is much lower, and the power consump⁃ tion of the devices can be reduced. In other scenarios, such as downlink machine type traffic, the simple orthogonal multiple access schemes are better due to the device cost and imple⁃ mentation complexity. OFDMA can be utilized for the cell⁃cen⁃ ter user with high data rate transmission applications. SIC: successive interference cancellation ▼Table 1. Summary of multiple access techniques DL: downlink eMBB: enhanced mobile broadband ESE: elementary signal estimator IGMA: interleave⁃grid multiple access mMTC: massive machine type communication MPA: message passing algorithm MUD: low complexity multiuser detector MUSA: multi user shared access NCMA: non⁃orthogonal coded multiple access PDMA: pattern division multiple access PIC: parallel interference cancellatio RSMA: resource spread multiple access SCMA: sparse code multiple access SIC: successive interference cancellation SPC⁃NOMA: superposition coding based non⁃orthogonal multiple access UL: uplink URLLC: ultra⁃reliable low latency machine type communication Category Scheme Scenario Multiplexing domain Transmitter Overloading Transmitter Spreading Transmitter multi ⁃ dimension constellation Receiver Power domain based SPC ⁃NOMA DL: eMBB Power Medium No No SIC Code domain based MUSA UL: mMTC, URLLC DL: eMBB Code/Power High Yes No SIC SCMA UL: mMTC, URLLC DL: eMBB Code/Power High Yes Yes MPA/SIC PDMA UL: mMTC, URLLC DL: eMBB Code/Power/Spatial High Yes No SIC/MPA RSMA UL: mMTC, URLLC Code/Power High Yes No SIC NCMA UL: eMBB, mMTC, URLLC code High Yes No PIC Interleaver based IGMA UL: eMBB, mMTC, URLLC Interleaver High Yes No MAP/MPA ESE MUD ▲Figure 8. Unified framework of multiple access schemes. Channel encoding, rate matching & scrambing Channel encoding, rate matching & scrambing Channel encoding, rate matching & scrambing … … …… Multiple access encoder (bit level) Interleaver Multiple access encoder (symbol level) Layer mapping & spatial precoding …… ……… Advanced receiver (IRC/SIC/R⁃ML) Advanced receiver (IRC/SIC/R⁃ML) … … Constellation optimization Factor graph Resource mapping+ + Example: u1 u2 u3 u4 u5 u6 r 1 r 2 r 3 r 4 Resource can be the time/frequency/space /code/power Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 31 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 36. Besides, the multiple access scheme should also be properly selected, taking the tradeoff of multiple conflicting objectives into account, e.g., complexity vs. performance, energy efficien⁃ cy (EE) vs. spectral efficiency (SE) and coverage. In addition, because the channel conditions and service load may also dy⁃ namically vary, the multiple access schemes and their related parameters such as the number of codewords, length of code⁃ word, spreading factor, max number of layers, need to be opti⁃ mized based on the instant services and the link conditions. In the following, we provide two potential adaptive multiple ac⁃ cess schemes in 5G. In [21], the adaptive multiple access scheme is studied from EE⁃SE co⁃design perspective, taking the detection complexity into consideration. The SCMA and OFDMA schemes are taken as the candidate uplink multiple access schemes in the study. The problem is formulated to choose the optimal multiple ac⁃ cess scheme and the related parameters simultaneously to max⁃ imize the EE under the total transmit power constraint, the quality of service (QoS) constraints and other specific require⁃ ments. The considered power consumption includes the trans⁃ mit power consumption, the static circuit power consumption, and the SCMA decoding power consumption related which is proportional to the SCMA decoding complexity order O(M df ) , where M is the constellation size, df = N K J , K and N denotes the codeword length and non⁃zero entries in each codeword in SCMA, respectively, and J = æ è ö ø K N is the maximum number of a⁃ ccess users. Fig. 9 shows the EE performance comparison of SCMA, OFDMA and the proposed link adaptation schemes with various cell radiuses. When the cell radius is small, the SCMA scheme has better EE performance; when the cell radi⁃ us is large, the OFDMA scheme performs better than SCMA scheme. The reason is that the SCMA can access more users than OFDMA, and the increment of the number of access users per resource can improve the system EE when the cell radius is small since the user transmit power efficiency is large when the path loss is small. When the cell radius increases, the user transmit power efficiency decreases and the increment of the number of access users per resource will decrease the system EE. The adaptation scheme can obtain the overall good EE per⁃ formance for all the cell radiuses (Fig. 9). Another example of the adaptive multiple access is between the spatial NOMA (also known as MIMO NOMA) scheme and orthogonal the multiuser MIMO (MU ⁃ MIMO). Fig. 10 shows ▼Table 2. Configuration methods of different multiple access schemes based on a unified framework IGMA: interleave⁃grid multiple access MUSA: multi user shared access MUST: Downlink Multiuser Superposition Transmission NCMA: non⁃orthogonal coded multiple access OMA: orthogonal multiple acce PDMA: pattern division multiple access RSMA: resource spread multiple access SCMA: sparse code multiple access SPC⁃NOMA: superposition coding based non⁃orthogonal multiple access OMA SPC ⁃NOMA MUSA (uplink) SCMA PDMA RSMA NCMA IGMA MUST Cat 1 [9] MUST Cat 2 [9] MUST Cat 3 [9] Interleaver Identity matrix Identity matrix Constraint permutation matrix Permutation matrix Identity matrix Identity matrix Identity matrix Optional Identity matrix Permutation matrix Constellation mapping Gray⁃mapped legacy constellation non⁃Gray⁃mapped superposed constellation Gray⁃mapped superposed constellation Gray⁃mapped legacy constellation Legacy constellation Joint optimization (multi ⁃dimensional modulation + Sparse matrix3 ) Legacy modulation Legacy modulation legacy modulation legacy modulation Factor graph Identity matrix Identity matrix Identity matrix Identity matrix Matrix composed of low cross⁃correlation and non⁃binary spreading sequence Sparse matrix Sparse matrix with unequal diversity order Matrix composed of scrambling code with good correlation properties Matrix obtained by Grassmannian line packing problem Sparse matrix Resource mapping (multiplexing domain) Time/frequency/code/space Power Power/bit Bit Code Code/power Code/power/space Code code code EE: energy efficiency OFDMA: orthogonal frequency ⁃division multiple access SCMA: sparse code multiple access ▲Figure 9. Average EE v.s. Cell Radiuses. 6.4 ×105 6.2 6.0 5.8 5.6 5.4 5.2 5.0 4.8 4.6 4.4 AverageEE(bits/Joule) 1.00.90.80.70.60.50.40.30.20.1 Cell radium (km) OFDMA Adaptation SCMA, N=2, K=4 SCMA, N=2, K=5 SCMA, N=2, K=6 October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS32 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 37. the concept of the orthogonal MU⁃MIMO and the spatial NO⁃ MA. In spatial NOMA, each two users can be served via one beam, and the interference between these two users will be large but can be cancelled via SIC decoding at the stronger us⁃ er. This is unlike the orthogonal MU⁃MIMO precoding (e.g. zero ⁃forcing MU⁃MIMO), in which there is no interference between users or it is usually small. Owing to this feature, when the channel has large correlations, spatial NOMA will have signifi⁃ cantly higher throughput over orthogonal MU⁃MIMO since the MU ⁃ MIMO precoding needs to reduce the interference be⁃ tween users and will suffer large gain loss in that case. In addi⁃ tion, when the user channel gain difference is large, the spatial NOMA will also have better performance compared to orthogo⁃ nal MU⁃MIMO due to the near⁃far effects. While in the high SNR regimes with low transmit correlation, the orthogonal MU⁃ MIMO is preferred since it can approach the capacity bound of MIMO broadcast channel in high SNR regimes. Considering the time varying fading characteristics of MIMO channels and the random distribution feature of active users, in a multiuser scenario, the adaptation between orthogonal MU ⁃ MIMO and spatial NOMA is desired for both the cell average and cell edge throughput enhancement. Fig. 11 shows the gain of the adaptation between orthogonal MU⁃MIMO based on zero⁃forcing scheme and spatial NOMA. When the transmit antenna correlation or the cell radius grows large, the spatial NOMA will have better performance than zero⁃ forcing based MU ⁃ MIMO, and the adaptation gain will in⁃ crease. The reason is that the higher transmit antenna correla⁃ tion will lead to the higher probability that the user channels have high correlation, and the larger cell radium will lead to the higher probability that the user channels have large gain difference. 5 Conclusions All the typical candidate NOMA solutions for 5G have differ⁃ ent strength points and weakness points. None of them can sur⁃ pass other schemes on all aspects. To fully exploit the advan⁃ tages of these candidate technologies and traditional orthogo⁃ nal multiple access solutions, a unified framework and a flexi⁃ ble multiple access schemes are required. Flexible switch among different NORMA schemes and the orthogonal multiple access technologies is expected to efficiently enhance the data rate and accommodate the necessary scalability for massive IoT connectivity and drastic reduction in access latency, and then to fully meet the diversified needs of 5G wireless commu⁃ nication systems. Some challenging problems need to be solved before NOMA schemes are put into use in 5G. In future, the impact of these candidate schemes on the existing systems, e.g. the grant free access procedure, reference signal, channel estimation and network assisted signaling, need to be carefully designed. What’s more, the performance tradeoff of the code mapping manners in these schemes and their implementation complexity may need further evaluated. The adaptive mecha⁃ nisms for part of these candidate schemes are also worth fur⁃ ther study to meet the diversified requirements of 5G. MU⁃MIMO: multiuser multiple⁃input multiple⁃output NOMA: non⁃orthogonal multiple access ▲Figure 11. Average cell throughput comparisons of various multiuser MIMO schemes (32 antenna base station at 6 GHz with four transceiver chains and four single antenna users) MU⁃MIMO: multiuser multiple⁃input multiple⁃output NOMA: non⁃orthogonal multiple access ▲Figure 10. Orthogonal MU⁃MIMO and Spatial NOMA. References [1] FuTURE Mobile Communication Forum 5G SIG. (Nov. 2014). Rethink mobile communications for 2020+ [Online]. Available: http://www.future ⁃forum.org/dl/ 141106/whitepaper.zip [2] A. Benjebbovu, Y. Saito, Y. Kishiyama, et al.,“Concept and practical consider⁃ ations of non ⁃ orthogonal multiple access for future radio access,”in IEEE ISPACS 2013, Naha, Japan, Nov. 2014, pp. 770- 774. doi: 10.1109/ ISPACS.2013.6704653. [3] 3GPP,“Discussion on multiple access for new radio interface,”TSG ⁃RAN WG1 #84bis, Busan Korea, R1 ⁃162226, Apr. 11-15, 2016. [4] H. Nikopour and H. Baligh,“Sparse code multiple access,”in IEEE PIMRC 2013, London, England, Sept. 2013, pp. 332- 336. doi: 10.1109/PIM⁃ RC.2013.6666156. User 1 User 2 Orthogonal beams User 2 User 1 (b) Spatial NOMA(a) Orthogonal MU⁃MIMO 20 Averagecellthroughput(bps/Hz) 0.8 Transmit antenna correlation 0.70.60.50.40.30.20.10 18 16 14 12 10 8 6 4 cell radius=500 m cell radius=1000 m MU⁃MIMO Spatial NOMA Adaptation Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 33 Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 38. [5] 3GPP,“Candidate Solution for New Multiple Access”, TSG ⁃RAN WG1 #84bis, Busan, Korea, R1 ⁃163383, Apr. 11-15, 2016. [6] 3GPP,“Candidate NR multiple access schemes”, TSG ⁃ RAN WG1 #84bis, Busan Korea, R1 ⁃163510, Apr. 11-15, 2016. [7] 3GPP,“Considerations on DL/UL multiple access for NR,”TSG ⁃RAN WG1 # 84bis, Busan Korea, R1 ⁃162517, Apr. 11-15, 2016. [8] 3GPP,“Considerations on DL/UL multiple access for NR,”TSG ⁃RAN WG1 # 84bis, Busan Korea, R1 ⁃163992, Apr. 11-15, 2016. [9] Study on Downlink Multiuser Superposition Transmission (MUST) for LTE, 3GPP TR36.859 v13.0.0, Jan. 2016. [10] Z. Ding, F. Adachi, and H. V. Poor,“The application of MIMO to non ⁃orthogo⁃ nal multiple access,”IEEE Transactions on Wireless Communications., to be published. [11] Q. Sun, S. Han, C. I, and Z. Pan,“On the ergodic capacity of MIMO NOMA sys⁃ tems,”IEEE Wireless Communications Letters, vol. 4, pp. 405-408, Aug 2015, doi: 10.1109/LWC.2015.2426709 [12] Q. Sun, Chih ⁃Lin I, S. Han, et al.,“Sum rate optimization for MIMO non ⁃or⁃ thogonal multiple access systems,”accepted by IEEE WCNC, 2015. [13] S. Han, Chin ⁃Lin I, Z. Xu, et al.,“Energy efficiency and spectrum efficiency co ⁃design: from NOMA to network NOMA,”IEEE MMTC E ⁃Letter (Special Is⁃ sue on 5G), vol. 9, no. 5, pp. 21-24, Sept. 2014. [14] M. Taherzadeh, H. Nikopour, A. Bayesteh, et al.,“SCMA codebook design,”in IEEE VTC Fall 2014, Vancouver, Canada, Sept. 2013, pp. 1-5. doi: 10.1109/ VTCFall.2014.6966170. [15] K. Xiao, B. Xiao, S. Zhang, et al.,“Simplified multiuser detection for SCMA with sum ⁃ product algorithm,”in IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5. doi: 10.1109/WCSP.2015.7341328. [16] Y. Wu, S. Zhang, Y. Chen, et al.,“Iterative multiuser receiver in sparse code multiple access systems,”in Proc. IEEE ICC 2015, London, England, Jun. 2015, pp. 2918-2923. doi: 10.1109/ICC.2015.7248770. [17] A. Bayesteh, H. Nikopour, M. Taherzadeh, et al.,“Low complexity techniques for SCMA detection,”in Proc. IEEE Globecom Workshops 2015, San Diego, USA, Dec. 2015, pp. 1-6. doi: 10.1109/GLOCOMW.2015.7414184. [18] H. Mu, Z. Ma, M. Alhaji, et al.,“A fixed low complexity message pass algo⁃ rithm detector for uplink SCMA system,”IEEE Wireless Communications Let⁃ ters, vol. 4. no. 6, pp. 585-588, 2015. doi: 10.1109/LWC.2015.2469668. [19] Y. Du, B. Dong, Z. Chen, et al.,“A fast convergence multiuser detection scheme for uplink SCMA systems,”IEEE Wireless Communications Letters, to be published. [20] S. Zhang, X. Xu, L. Lu, et al.,“Sparse code multiple access: an energy efficient uplink approach for 5G wireless systems,”in Proc. IEEE Globecom 2014, Aus⁃ tin, USA, Dec. 2014, pp. 4782-4787. doi: 10.1109/GLOCOM.2014.7037563. [21] Q. Sun, Chih ⁃Lin I, S. Han, et al.,“Software defined air interface: a framework of 5G air interface,”in Proc. IEEE WCNC Workshop 2015, New Orleans, USA, Mar. 2015, pp. 6-11. doi: 10.1109/WCNCW.2015.7122520. [22] K. Au, L. Zhang, H. Nikopour, et al.,“Uplink contention based SCMA for 5G radio access,”in Proc. IEEE Globecom Workshop 2014, Austin, USA, Dec. 2014, pp. 900-905. doi: 10.1109/GLOCOMW.2014.7063547. [23] A. Bayesteh, E. Yi, H. Nikopour, et al.,“Blind detection of SCMA for uplink Grant ⁃Free Multiple Access,”in Proc. IEEE ISWCS 2014, Barcelona, Spain, Aug. 2014, pp. 853-857. doi: 10.1109/ISWCS.2014.6933472. [24] T. Liu, X. Li, and L. Qiu,“Capacity for downlink massive MIMO MU ⁃SCM.A system,”in Proc. IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5. doi: 10.1109/WCSP.2015.7341100. [25] H. Nikopour, E. Yi, A. Bayesteh, et al.,“SCMA for downlink multiple access of 5G wireless networks,”in Proc. IEEE Globalcom 2014, Austin, USA, Dec. 2014, pp. 3940-3945. doi: 10.1109/GLOCOM.2014.7037423. [26] U. Vilaipornsawai, H. Nikopour, and A. Bayesteh,“SCMA for open ⁃loop joint transmission CoMP,”in Proc. IEEE VTC Fall 2015, Boston, USA, Sept. 2015, pp. 1-5. doi: 10.1109/VTCFall.2015.7391126. [27] Y. Li, X. Lei, P. Fan, et al.,“An SCMA ⁃based uplink inter ⁃cell interference cancellation technique for 5G wireless systems,”in Proc. IEEE WCSP 2015, Nanjing, China, Oct. 2015, pp. 1-5, doi: 10.1109/WCSP.2015.7341306 [28] A. Medra and T. N. Davidson,“Flexible codebook design for limited feedback systems via sequential smooth optimization on the grassmannian manifold,” IEEE Transactions on Signal Processing, vol. 62, no. 5, pp. 1305-1318, Mar. 2014. doi:10.1109/TSP.2014.2301137. Manuscript received: 2016⁃06⁃30 SUN Qi (sunqiyjy@chinamobile.com) received the BSE and PhD degree in informa⁃ tion and communication engineering from Beijing University of Posts and Telecom⁃ munications, China in 2009 and 2014, respectively. After graduation, she joined the Green Communication Research Center of the China Mobile Research Institute. Her research focuses on 5G key technologies, including non ⁃orthogonal multiple access, new waveforms, flexible duplex and UDN. WANG Sen (wangsenyjy@chinamobile.com) received the PhD degree in informa⁃ tion and communication engineering from Beijing University of Posts and Telecom⁃ munications, China in 2013. He joined the Green Communication Research Center of the China Mobile Research Institute after graduation. He is now working on the 5G key technologies and standardization. His research interests include massive MI⁃ MO, non ⁃orthogonal multiple access, new waveforms and system level evaluation. HAN Shuangfeng (hanshuangfeng@chinamobile.com) received his MS and PhD de⁃ grees in electrical engineering from Tsinghua University, China in 2002 and 2006 respectively. He joined Samsung Electronics as a senior engineer in 2006 working on MIMO, MultiBS MIMO etc. From 2012, he is a senior project manager in the Green Communication Research Center of the China Mobile Research Institute. His research interests are green 5G, massive MIMO, full duplex, NOMA and EE ⁃SE codesign. Chih⁃Lin I (icl@chinamobile.com) received her PhD degree in electrical engineer⁃ ing from Stanford University, USA. She has been working at multiple world ⁃class companies and research institutes leading the R&D, including AT&T Bell Labs; Di⁃ rector of AT&T HQ, Director of ITRI Taiwan, and VPGD of ASTRI Hong Kong. She received the IEEE Trans. COM Stephen Rice Best Paper Award and is a winner of the CCCP National 1000 Talent Program. In 2011, she joined China Mobile as its Chief Scientist of wireless technologies, established the Green Communications Re⁃ search Center, and launched the 5G Key Technologies R&D. She is spearheading major initiatives including 5G, C ⁃ RAN, high energy efficiency system architec⁃ tures, technologies and devices; and green energy. She was an elected Board Mem⁃ ber of IEEE ComSoc, Chair of the ComSoc Meetings and Conferences Board, and Founding Chair of the IEEE WCNC Steering Committee. She is currently an Execu⁃ tive Board Member of GreenTouch, a Network Operator Council Member of ETSI NFV, a Steering Board Member of WWRF, and a Scientific Advisory Board Member of Singapore NRF. Her current research interests center around“Green, Soft, and Open” BiographiesBiographies October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS34 Special Topic Unified Framework Towards Flexible Multiple Access Schemes for 5G SUN Qi, WANG Sen, HAN Shuangfeng, and Chih⁃Lin I
  • 39. Multiple Access Rateless Network Coding forMultiple Access Rateless Network Coding for Machine⁃to⁃Machine CommunicationsMachine⁃to⁃Machine Communications JIAO Jian1,2 , Rana Abbas2 , LI Yonghui2 , and ZHANG Qinyu1 (1. Harbin Institute of Technology Shenzhen Graduate School, Shenzhen 518055, China; 2. Center of Excellence in Telecommunications, University of Sydney, Sydney, NSW 2006, Australia) Abstract In this paper, we propose a novel multiple access rateless network coding scheme for machine⁃to⁃machine (M2M) communications. The presented scheme is capable of increasing transmission efficiency by reducing occupied time slots yet with high decoding suc⁃ cess rates. Unlike existing state⁃of⁃the⁃art distributed rateless coding schemes, the proposed rateless network coding can dynami⁃ cally recode by using simple yet effective XOR operations, which is suitable for M2M erasure networks. Simulation results and analysis demonstrate that the proposed scheme outperforms the existing distributed rateless network coding schemes in the scenar⁃ io of M2M multicast network with heterogeneous erasure features. rateless network coding; multiple accesss; machine⁃to⁃machine communications (M2M) Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 005 http://www.cnki.net/kcms/detail/34.1294.TN.20161011.1511.002.html, published online October 11, 2016 Special Topic M 1 Introduction achine⁃to⁃machine (M2M) communication sys⁃ tem is expected to support a massive number of devices communicating with each other in a fully automated fashion with minimum or with⁃ out human intervention [1]. Equipped with networked and real⁃ time processing capabilities, these devices can implement a wide range of applications, such as intelligent transportation systems (ITS), healthcare monitoring, smart metering, energy management and smart grids. M2M communication system is generally characterized by a massive number of machine⁃type communication (MTC) devic⁃ es that have no/low mobility, low computational and storage ca⁃ pabilities, and low power budget [2]. Moreover, most MTC de⁃ vices suffer from severe congestion and access delay in an M2M system with a large number of devices [3]-[5]. Therefore, the main motivation behind this paper is to propose a coding strategy that exploits the interference in the channel to in⁃ crease data rates. Our work focuses on the cooperative joint network and coding strategy for MTC devices in multicast set⁃ tings. These MTC devices disseminate messages to multiple re⁃ ceivers simultaneously with the help of relay nodes. The rateless code, originally investigated in [6] for single source broadcasting in a single hop network, is deemed as a milestone for packet erasure codes. It can recover the original k information symbols from any n = k + O( k ln2 (k θ)) received coded symbols with the probability 1⁃θ and the decoding cost of O(k ln(k θ)) of operations, where θ is the allowable failure probability to recover the original message after n coded sym⁃ bols have been received. In addition, the encoding and decod⁃ ing process of the rateless code is complex, including logarith⁃ mic order for Luby Transform (LT) code and linear order for Raptor code. Furthermore, both LT and Raptor codes are able to provide practical capacity ⁃ achieving solutions, if their en⁃ coding degree distributions are sophisticated designed [7], [8]. The rateless code has been widely applied in cooperative communications [9]- [13]. In [9], the complexity, delay, and memory of different state⁃of⁃the⁃art rateless coding algorithms are analyzed for a multi⁃hop network. In [10], a superimposed on ⁃ the ⁃ fly recoding scheme is performed by each transport node in a multi⁃hop tree network, but it is difficult to imple⁃ ment due to the high decoding complexity. The first distributed LT (DLT) code is proposed in [11], and a new degree distribu⁃ tion, named deconvolved soliton distribution (DSD) is de⁃ signed. However, all the source nodes and relays are assumed This works was supported in part by Natural Scientific Research Innovation Foundation in Harbin Institute of Technology under Grant No. HIT. NSRIF 2017051, Shenzhen Basic Research Program under Grant Nos. JCYJ20150930150304185 and JCYJ2016 0328163327348, and National High Technology Research & Development Program of China under Grant No. 2014AA01A704. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 35
  • 40. Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS36 to have exact information regarding the number of sources and encoding distributions to adapt relaying schemes. In [12], the authors utilized density evolution and linear programming frameworks to find an optimal combination at each relay node for any network architecture consisting of four sources. This manner can achieve the asymptotical error floor, but has intrac⁃ table calculation complexity. A relaying protocol for Y ⁃ net⁃ works, namely Soliton⁃Like Rateless Coding (SLRC), is intro⁃ duced in [13]. By enabling probabilistic forwarding and com⁃ bining packets to reduce the overhead between relay and desti⁃ nation, the aggregate distribution at the destination can still maintain a near⁃ideal distribution, even if one source left the network. However, the lack of buffer utilization in SLRC relay limits the total encoding and decoding overheads. The destination cooperation in interference channels is an⁃ other feature of M2M communication where one device can act as a relay for another device. A cooperative communication scheme for two mobile users is proposed in [4], which is poten⁃ tially able to receive and decode each other’s messages based on the signal⁃to⁃interference⁃plus⁃noise ratio (SINR). In [5], The received signal at the destination can be realized as a su⁃ perposition of coded symbols sent from the relay, which is ca⁃ pacity approaching if an appropriate successive interference cancellation (SIC) is used for decoding [14]. In this paper, we propose an adaptive rateless network cod⁃ ing scheme in an M2M erasure network. First, an simple de⁃ gree distribution is designed for rateless coding in all the source devices, and the collided devices transmit simultaneous⁃ ly. Then, an optimal relaying strategy is proposed to forward and combine the encoded packets with appropriate propor⁃ tions, according to different erasure probabilities over the un⁃ derlying edges. This is particularly suitable for M2M communi⁃ cations with strict power limitations, especially when the data size is small. By doing this, the total time slot of transmission is reduced obviously while the high decoding success rates are maintained. Moreover, we compared the current typical rate⁃ less coding relay schemes with our proposed scheme, with the aspects of the complexity and buffer memory. Simulation re⁃ sults show that the proposed rateless network coding scheme outperforms the existing distributed rateless coding schemes under various erasure probability scenarios. 2 Network Model and Rateless Code 2.1 Network Model In [12] and [13], authors have introduced and optimized the applications of rateless coding in the Y⁃network model. We at⁃ tempt to extend the Y⁃network model to a relay multicast mod⁃ el as shown in Fig. 1a. The relay R multicasts the data streams both to destination nodes D1 and D2 and guarantees the two source data to be recovered. Due to the special feature of rate⁃ less coding, the overheads at two destination nodes are appar⁃ ently the same as the one in the Y⁃network and accordingly the DLT and SLRC algorithms are also appropriate for the network model in Fig. 1a. We also consider a butterfly network model (Fig. 1b), which has two sources, one relay and two destinations. Two direct edges are added to send two separate data streams form S1 and S2 respectively. The encoded packets should be processed (re⁃ coding) at R, and they can be converged at both D1 and D2 in the end. The model uses multicast from source to relay and di⁃ rectly from source to destination as well. This is its remarkable difference from the Y⁃network model. We define the edges in this model as e1 to e6. Each edge has an erasure probability εi, which is described as an independent⁃identical⁃distribution (i.i. d) Bernoulli variable. We assume that the packet size and the transmission rate of all the edges are equivalent (one time slot one packet). The rateless coding transmission scheme in this network model is described as follows. 1) Step 1: S1 and S2 generate the encoded packets with a rate⁃ less coding degree distribution [7]; 2) Step 2: S1 multicasts the encoded stream both to R and to D1, and simultaneously, S2 multicasts its stream to R and to D2; 3) Step 3: R generates a new encoded stream by the relaying network coding (NC) scheme with the received encoded packets and multicasts them to D1 and D2 simultaneously; 4) Step 4: Once D1 and D2 receive enough encoded packets, they start to decode the encoded packets from the source and relay nodes to recover the two sources original packets; ▲Figure 1. The proposed network models: (a) relay multicast model; (b) butterfly model. (a) R: relay node S:source node D: destination node (b) S1 S2 D1 D2 R e3 e4 e1 e2 S1 S2 D1 D2 R e1 e2 e3 e4 e5 e6
  • 41. Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 37 5) Step 5: After successful decoding, D1 (D2) transmits a single acknowledgment (ACK) packet indicating the termination of the session. 2.2 Rateless Coding Rateless coding is a modern forward error correction (FEC) technology. It protects from packet ⁃ loss, and can reduce the feedbacks for user acknowledgement due to rarely caring about the erasure probability of the channel. In a single⁃hop scenar⁃ io, the original packets are defined as k data symbols, and the encoded packets required by the decoder are defined as n en⁃ coded symbols. Furthermore, the overhead is defined as ( )n - k k and the number of encoded symbols sent by the send⁃ er is defined as N. Therefore, the expected code rate at the sender is conveyed as R = k N = ( )1 - ε k n , where ε is the era⁃ sure probability of the channel. On the other hand, the encod⁃ ing and decoding complexity of rateless codes is very low, which are expressed as O(ln k) for LT codes and O(k) for Rap⁃ tor codes. As an example of rateless coding, the LT code uses robust soliton distribution (RSD) to achieve the erasure channel ca⁃ pacity under the single hop network. The coding degree distri⁃ bution is a key element for the successful recovery of data sym⁃ bols. For a parameter δ and the length k RSD, μ(k) is defined as: μ(i)= ρ(i)+ τ(i) β , for i = 1,…,k , (1) where ρ(i)={1 i, for i = 1 1 i(i - 1) , for i = 2,…,k , (2) τ(i)= ì í î ïï ïï S ik, for i = 1,…,ë ûk S - 1 S ln(S δ) k, for i = ë ûk S 0, for i = ë ûk S + 1,…,k , (3) β =∑i = 1 k μ(i)+ τ(i) . (4) S is the average number of degree⁃one symbols, namely rip⁃ ple size, which is defined by S = c ln(k δ) k , where c>0. It is worth noting that the LT decoder performs the BP algo⁃ rithm with the prior knowledge of the degree and associated neighbors. Given a block of encoded symbols, the decoder re⁃ cursively decodes the data symbols from the bipartite graph connecting the information and encoded symbols. The BP algo⁃ rithm starts from degree⁃one symbols, by removing their contri⁃ butions from the graph in order to produce a smaller graph with another set of degree⁃one encoded symbols. Then, the new degree⁃one encoded symbols of this smaller graph are removed again, and iteratively the process continues to recover all data symbols, as described in [7] and [8]. 3 Analysis of Relaying Schemes As the rateless code is used in multi⁃source relay network, the erasure probability of different paths (multicast and uni⁃ cast) may influence the relaying strategy and the correspond⁃ ing performance with NC. Specifically, the relay R may receive no packet from S1 or S2 in a time slot due to packet loss in multi ⁃source relay network. Hence, it is an interesting and signifi⁃ cant topic to select proper rateless coding algorithms based on NC and relaying strategy for efficient transmissions on lossy network. In this section, we try to consider the conventional methods in Y ⁃ networks and butterfly networks, and compare their decoding performance at the destination nodes. Moreover, we propose a new optimized ⁃NC scheme to trade off the decod⁃ ing performance by selecting proper forwarding and combining probabilities. 3.1 Comparison of Typical Relaying Schemes We assume that the number of original packets is k and the number of encoded packets generated is N at both the source nodes. In two fixed time slots, the destination nodes D1/D2 of butterfly network can receive one encoded packet from S1/S2 in the first slot, and then receive one from the relay R in the sec⁃ ond slot. It is a limited condition that D1 and D2 only receive the maximum 2N packets when N encoded packets are sent from the source, if and only if all the edges are lossless. D1 and D2 use the BP decoding algorithm to decode the compilations of two encoded streams after 2N slots to recover 2k original pack⁃ ets, respectively. There are the following four typical relaying schemes in this butterfly network model: 1) Store⁃and⁃Forward (SF) The relay R immediately forwards the packets to the next hop as soon as it receives packets. If two packets arrive simul⁃ taneously, R randomly forwards one of them and stores another into the buffer. If the relay R receives no packet, it waits for the next slot. Due to the uncertain storage of packets, this scheme may easily make congestion on R. 2) DLT With S1 (S2) using DSD, R performs random decision proto⁃ col in [11] to combine and forward two received packets. Once the erasure event occurs at one of the edges between sources and relay, R directly forwards another received packet. If no packets arrive, the relay waits for the next slot. These waiting slots at relay lead to low efficiency due to a serious waste of sources. By using considerable low ⁃ degree encoded packets, this scheme could scarcely cover all the original packets of two sources, despite of its simple encoding complexity. 3) SLRC With S1 (S2) using RSD, R uses the SLRC relaying scheme to operate the two encoded packets. It forwards most of the low degree packets (degree⁃one or degree⁃two) directly and com⁃
  • 42. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS38 bines other high degree packets, in order to assure the aggre⁃ gate distribution at destination nodes to be soliton⁃like. With the SLRC, the transmission time slot is not wasted if the pack⁃ et on e1 or e2 is dropped, by R’s choosing two packets from the buffers to combine and forward. Compared with the DLT scheme, this SLRC scheme obtains more gains. 4) eXclusive⁃OR (XOR) This scheme is based on the simple eXclusive ⁃ OR (XOR) operation of classic NC in the butterfly model. The relay R combines the two received packets into one new packet. If only one packet arrives, the relay R sends the received packet di⁃ rectly, and if no packet arrives, R waits for the next time slot. Since the relay carries the two received packets by forward⁃ ing and combining operations, the recoding complexity comes almost from XOR operation which depends mainly on the era⁃ sure probability of the edges. In addition, the relay requires two buffers to store the packets from different sources. There⁃ fore, we compare the four schemes (Table 1). We can find that SF has the least recoding complexity, and the complexity of SL⁃ RC and DLT depends significantly on their forwarding proba⁃ bility. On the other hand, DLT and XOR need the smallest buf⁃ fer size of only one packet for each source, while SLRC must store all the packets not forwarded for the next operation. 3.2 Analysis of Decoding Matrix In this section, we make an intuitive analysis on recovery performance at the destination nodes by the BP decoding ma⁃ trix, in which the BP decoding algorithm could be described as one unitization process, with the columns denoting the encod⁃ ed symbols and the rows denoting the original data symbols. We define A and B as the encoding matrix at S1 and S2 re⁃ spectively, and denote the dimension of the matrix as the sub⁃ scripts. The decoding matrices at D1 and D2 are the aggregation of the forwarding and combining sub⁃matrices A and B, with 2k rows due to the number of data symbols from two sources. Since the lost packets have been removed from the matrix, the number of columns reveals the received encoded symbols by destination nodes exactly. For the SF scheme, the decoding matrices at D1 and D2 are defined as: F D1 2k ×[ ](1 - ε5)N +[ ](1 - ε1)+(1 - ε2) (1 - ε3)N 2 = é ë ê ê ù û ú ú A S1 k ×(1 - ε5)N A S1 k ×(1 - ε1)(1 - ε3)N 2 0 0 0 B S2 k ×(1 - ε2)(1 - ε3)N 2 , (5) F D2 2k ×[ ](1 - ε6)N +[ ](1 - ε1)+(1 - ε2) (1 - ε4)N 2 = é ë ê ê ù û ú ú 0 A S1 k ×(1 - ε1)(1 - ε4)N 2 0 B S2 k ×(1 - ε6)N 0 B S2 k ×(1 - ε2)(1 - ε4)N 2 . (6) From the matrices F D1 and F D2 in (5) and (6), we know that the SF scheme is only appropriate for unicast like source to destination, since the dimensions of sub⁃matrices for A and B are extremely unequal. The lack of combination operations makes the destination nodes unable to recover the whole data symbols. Therefore, this scheme is inefficient for multicast in the butterfly network model. For the SLRC scheme, as an optimized DLT, we only pres⁃ ent its decoding matrices that are defined as: H D1 2k × ( )1 - ε5 + 1 - ε3 N = é ë ê ê êê ê ê ù û ú ú úú ú ú A S1 k ×(1 - ε5)N A S1 k × ν1(1 - ε3)N 0 A S1 k ×(1 -∑i = 1 2 νi)(1 - ε3)N 0 0 B S2 k × ν2(1 - ε3)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε3)N , (7) H D2 2k × ( )1 - ε6 + 1 - ε4 N = é ë ê ê êê ê ê ù û ú ú úú ú ú 0 A S1 k × ν1(1 - ε4)N 0 A S1 k ×(1 -∑i = 1 2 νi)(1 - ε4)N B S2 k ×(1 - ε6)N 0 B S2 k × ν2(1 - ε4)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε4)N , (8) where νi (i=1, 2) is the probability distribution of the relay for⁃ warding packets from S1 and S2, while ν͂ i = 1 - νi is the distribu⁃ tion of the packets into the buffer. For the XOR Scheme, the decoding matrices are defined as: G D1 2k ×[ ](1 - ε5)+ Max{ }(1 - ε1), (1 - ε2) (1 - ε3) N = é ë ê ê ù û ú ú A S1 k ×(1 - ε5)N A S1 k ×(1 - ε1)(1 - ε3)N 0 B S2 k ×(1 - ε2)(1 - ε3)N , (9) G D2 2k ×[ ](1 - ε6)+ Max{ }(1 - ε1), (1 - ε2) (1 - ε4) N = é ë ê ê ù û ú ú 0 A S1 k ×(1 - ε1)(1 - ε4)N B S2 k ×(1 - ε6)N B S2 k ×(1 - ε2)(1 - ε4)N . (10) In (9) and (10), G can be segmented into four sub⁃matrices: Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu ▼Table 1. The complexity and buffer for the four schemes * γi and νi are the distributions of the no⁃forwarding packets for DLT and SLRC DLT: distributed Luby transform SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR SF DLT SLRC XOR Relay recoding complexity 0 ( )1 - ε1 ( )1 - ε2 ( )1 - γi N (1 -∑i = 1 2 νi)N ( )1 - ε1 ( )1 - ε2 N Buffer size of relay N 2 packets for each source One packet for each source ( )1 - νi N packets for each source One packet for each source
  • 43. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 39 the two parts in the left are the forwarding matrix A or B, and the two right parts are the combined matrices. Since the com⁃ bined symbols of this scheme may be lack of degree⁃one and degree⁃two packets, it needs enough columns of single sub⁃ma⁃ trices A or B on the edge e5 or e6 in Fig. 1 to start the BP decod⁃ er. In the decoding process, only if all the data symbols of A have been recovered, B can begin to decode. By comparing these decoding matrices, we can find that the forwarding matrix A or B has occupied such numerous columns in H, as A S1 k × ν1(1 - ε3)N or B S2 k × ν2(1 - ε3)N . The combination part of en⁃ coded symbols as A and B é ë ê ê êê ê ê ù û ú ú úú ú ú A S1 k ×(1 -∑i = 1 2 νi)(1 - ε3)N B S2 k ×(1 -∑i = 1 2 νi)(1 - ε3)N in the SLRC scheme has much less columns than that of é ë ê ê ù û ú ú A S1 k ×(1 - ε1)(1 - ε3)N B S2 k ×(1 - ε2)(1 - ε3)N in the XOR scheme. SLRC can still recode new packets in the re⁃ lay to transmit, even if no packets received due to enough large ε1 and ε2. However, it has not fully utilized the packets directly from D1 and D2 on e5 and e6. Note that, only N encoded packets at most could be sent by the sources and relay, which con⁃ strains the required time slots to be only 2N in the network model, given one packet at each time slot. It renders that the decoding matrix H has many single forwarding columns in A or B which obviously reduces the relevance of encoded symbols from S1 and S2. As a result, the large proportion of forwarding packets by the relay cannot give much help to improve the BP decoding performance, especially on the condition of the rela⁃ tively small erasure probability ε5 and ε6. On the other hand, when ε5 and ε6 become larger, the decod⁃ ing performance is mostly decided by the proportion of forward⁃ ing the single packets and XOR combination of two sources’ packets in the relay. In the XOR scheme, the number of recov⁃ ery data symbols would decline very fast due to the lack of low degree encoded symbols for BP decoding. Besides, the XOR operations would be blocked and degraded since two separate packets from two sources could hardly arrive at the relay simul⁃ taneously in the large ε5 and ε6. 3.3 Proposed Optimized NC Scheme On the basis of above analysis, we have found that the relay⁃ ing schemes should forward the low⁃degree packets for starting BP decoder, by taking into consideration the erasure probabili⁃ ties of the direct edges e1 and e2. On the other side, the relay al⁃ so needs to remain the enough proportion of combinations of the packets in the buffers, in order to prevent from no received packets from S1 and S2 simultaneously. Accordingly, we pro⁃ pose a novel NC scheme with a self⁃adjusted forwarding proba⁃ bility associated with the variations of ε5 and ε6. The basic rule of the proposed scheme is as follows: if ε1 and ε2 increase, the relay immediately forwards more low⁃degree packets; if ε5 and ε6 decrease, the relay combines more packets. The proposed algorithm, named Opt ⁃ NC scheme, has the comparable complexity and buffer requirements with the SLRC scheme. We denote the forwarding probabilities λ and θ for en⁃ coded packets from S1 and S2, which are predetermined to be equivalent to the erasure probabilities ε5 and ε6, respectively. Algorithm 1 shows the steps of Opt⁃NC algorithm. Algorithm 1: Opt⁃NC Scheme (at one time slot) p1: received encoded packets from S1; p2: received encoded packets from S2; d1: degree of p1; d2: degree of p2; λ: forwarding probability of the low degree packets from S1; θ: forwarding probability of the low degree packets from S2; a=rand(); b=rand(); if d1=1˅2 and d2=1˅2 and a<λ and b <θ forward p1 or p2 with equal probability; put another packet into the buffer of another source; else if d1=1˅2 and a <λ forward p1 and put p2 into the buffer of S2; else if d2=1˅2 and b <θ forward p2 and put p1 into the buffer of S1; else put the packets received into the buffers respectively; pb1: random choose one packet in the buffer of S1; pb2: random choose one packet in the buffer of S2; pXOR = pb1XOR pb2; forward pXOR; end if * ˅ means logical operator of OR 4 Simulation Results and Discussion We analyze the performance of the above five algorithms in a butterfly network coding system as Fig. 1b. The encoding de⁃ gree distribution is selected to be RSD with parameters δ = 0.05, c=0.03. The number of data symbols k=100, and the num⁃ ber of encoded symbols from S1 and S2 is indicated to be the same as N. We emulate the encoding and decoding procedure using Monte Carlo experiments with 10,000 times. The ratio between the statistics of decoding failure times and total exper⁃ iment times is defined as decoding failure rate (DFR). In this work, the lowest displayable DFR in our simulation is 10⁃4 . Giv⁃ en the time slots and erasure probabilities of edges, the lower DFR of relaying schemes means outstanding decoding perfor⁃ mance. We give the unicast performance and multicast perfor⁃ Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic
  • 44. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS40 mance respectively to discuss the influences between the sin⁃ gle source and double sources. Fig. 2 gives the performance of five schemes in the erasure free network model. We can find that the Opt⁃NC scheme ap⁃ proaches the unicast and multicast curves of the XOR scheme, which obtain the lowest DFR of 10⁃4 at 400 time slots. The oth⁃ er schemes need at least 600 time slots to recover two sources’ data, due to their inefficiency of buffer utilization at the relay. This simulation result proves that the Opt⁃NC scheme can ap⁃ ply as a typical NC scheme to get the remarkable multicast throughput gains in lossless network data transmission. With vary erasure probabilities from 0 to 0.8 of all edges among ε1 to ε6, and the code length N=360, the decoding per⁃ formance of five schemes is shown in Fig. 3 . The unicast re⁃ sults of the schemes reveal the better performance than the multicast results, since the erasure probabilities of ε1, ε2, ε3, ε4 make the combination operations at the relay inappropriate. It is noted that the SLRC, XOR and Opt⁃NC schemes have simi⁃ lar multicast decoding performance, with the DFR lower than 10⁃ 4 and the erasure probability of 0.2. The simulation results indicate that our adaptive Opt⁃NC scheme integrates the advan⁃ tages of SLRC and XOR, which also reveals outstanding decod⁃ ing performance in lossy network. Fig. 4 shows the DFRs of five schemes with the encoded packets of N=250, ε1 to ε4 null, and ε5 and ε6 from 0.2 to 0.9. Since the multicasts of the SF and DLT are both restricted by the limited number of encoded packets from source, their de⁃ coding performance maintains at an inferior level in spite of ε5 and ε6 increasing. On the other side, the multicast DFR perfor⁃ mance of XOR and that of Opt⁃NC are almost consistency with their unicasts. If the erasure probabilities of e5 and e6 are lower than 0.25, the Opt⁃NC scheme can get the similar DFR with that of the XOR scheme. If ε5 and ε6 increase from 0.25 to 0.9, the Opt⁃NC scheme outperforms the XOR scheme by its dy⁃ namic property. In addition, compared to the SLRC, the Opt⁃ NC scheme also has a better performance as ε5 and ε6 are both lower than 0.45. However, the SLRC gives a lower decoding failure rate as the ε5 and ε6 are both in a range of 0.45 to 0.8. Once the erasure probability increases higher than 0.8 (the edges e5 and e6 are almost interrupted), the Opt⁃NC scheme ap⁃ proaches SLRC⁃multicast with a higher efficiency. In a word, Special Topic Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR ▲Figure 2. All the edges are erasure free, k =100. ▲Figure 3. All the edges have the same erasure probability, k =100, N =360. ▲Figure 4. Edges e5 and e6 are lossy while other edges are lossless, k =100, N =250. 100 10-1 10-2 10-3 10-4 600550500450400350300250200150 Total slots of transmission DFR SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast 100 10-1 10-2 10-3 10-4 0.8 Erasure probability of all edges DFR 0.70.60.50.40.30.20.10 SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast 100 10-1 10-2 10-3 10-4 1.0 ε5 and ε6 DFR 0.2 0.90.80.70.60.50.40.3 SF⁃unicast SF⁃multicast DLT⁃unicast DLT⁃multicast SLRC⁃unicast SLRC⁃multicast XOR⁃unicast XOR⁃multicast Opt⁃NC⁃unicast Opt⁃NC⁃multicast DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR DFR: decoding failure rate DLT: distributed Luby transform NC: network coding SF: store⁃and⁃forward SLRC: soliton⁃like rateless coding XOR: eXclusive⁃OR
  • 45. October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 41 the proposed relaying scheme is an adaptive method to compro⁃ mise the decoding performance of XOR and SLRC. 5 Conclusions In this paper, we have studied rateless network coding ap⁃ plied in machine⁃to⁃machine communications for multiple ac⁃ cess applications. A novel dynamic relaying scheme Opt⁃NC was proposed that exploits the forwarding and combining opera⁃ tions to obtain an enhanced decoding performance of the de⁃ coder at the destination nodes. The Opt⁃NC scheme has adap⁃ tive capability of responding to the vary erasure probability of direct edges. The simulation results show that the proposed re⁃ lay scheme performs close to the optimal XOR scheme in loss⁃ less and lossy network, respectively. Furthermore, the Opt⁃NC scheme can be used in the physical layer by incorporating the XOR operation and superposition practical modulations. Multiple Access Rateless Network Coding for Machine⁃to⁃Machine Communications JIAO Jian, Rana Abbas, LI Yonghui, and ZHANG Qinyu Special Topic References [1] M. Shirvanimoghaddam, Y. Li, and M. Dohler,“Probabilistic rateless multiple access for machine⁃to⁃machine communication,”IEEE Transactions on Wireless Communications, vol. 14, no. 6815-6826, Dec. 2015. [2] K. Zheng, S. Ou, J. Alonso⁃Zarate, et al.,“Challenges of massive access in high⁃ ly dense LTE⁃Advanced networks with machine⁃to⁃machine communications,” IEEE Wireless Communications Magazine, vol. 21, no. 3, pp. 12-18, Jun. 2014. [3] G. Durisi, T. Koch, and P. Popovski. (2015).“Towards massive, ultra⁃reliable, and low ⁃ latency wireless communication with short packets,”CoRR [Online]. Available: https://arxiv.org/abs/1504. 06526 [4] B. W. Khoueiry and M. R. Soleymani,“A novel destination cooperation scheme in interference channels”, in Proc. IEEE 80th Vehicular Technology Conference (VTC 2014 ⁃ Fall), Vancouver, Canada, Sept., 2014. doi: 10.1109/VTC⁃ Fall.2014.6965826. [5] M. Shirvanimoghaddam, M. Dohler, and S. J. Johnson. (2016).“Massive multiple access based on superposition raptor codes for M2M communications,”CoRR [Online]. Available: https://arxiv.org/abs/1602.05671 [6] J. Byers, M. Luby, M. Mitzenmacher, and A. Rege,“A digital fountain approach to reliable distribution of bulk data,”in Proc. ACM SIGCOMM’98, Vancouver, Canada, Jan. 1998, pp. 56-67. doi:10.1145/285237.285258. [7] M. Luby,“LT Codes,”in Proc. IEEE Symposium on the Foundations of Comput⁃ er Science (STOC), Vancouver, Canada, 2002, pp. 271-280. [8] A. Shokrollahi,“Raptor Codes,”IEEE Transactions on Information Theory, vol. 52, no. 6, pp. 2551-2567, Jun. 2006. doi: 10.1109/TIT.2006.874390. [9] P. Pakzad, C. Fragouli, and A. Shokrollahi,“Coding schemes for line networks,” in Proc. ISIT, Adelaide, Germany, 2005, pp. 1853-1857. [10] R. Gummadi and R. S. Sreenivas,“Relaying a fountain code across multiple nodes,”in Proc. SIGCOMM’08, Seattle, USA, 2008, pp. 483-484. [11] S. Puducheri, J. Kliewer, and T. E. Fuja,“The design and performance of dis⁃ tributed LT codes,”IEEE Transactions on Information Theory, vol. 53, no. 10, pp. 3740-3754, Oct. 2007. [12] D. Sejdinovic, R. Piechocki, and A. Doufexi,“AND⁃OR tree analysis of distrib⁃ uted LT codes,”in Proc. ITW, Volos, Greece, 2009, pp. 261-265. [13] A. Liau, S. Yousefi, and I. Kim,“Binary soliton⁃like rateless coding for the Y⁃ network,”IEEE Transactions on Communications, vol. 59, no. 12, pp. 3217- 3222, Dec. 2011. [14] R. Abbas, M. Shirvanimoghaddam, Y. Li, and B. Vucetic,“On SINR⁃based ran⁃ dom multiple access using codes on graph,”in IEEE Global Communications Conference (GLOBECOM), San Diego, USA, 2015. doi: 10.1109/GLO⁃ COM.2015.7417013. Manuscritp received: 2016⁃07⁃14 JIAO Jian (jiaojian@hitsz.edu.cn) received his PhD degree in communication engi⁃ neering from Harbin Institute of Technology (HIT) in 2011. He received his BS de⁃ gree in electrical engineering from Harbin Engineering University in 2005, and his MASc degree in information and communication engineering from HIT Shenzhen Graduate School in 2007. He is an assistant research fellow in the Department of Electrical and Information Engineering of HIT Shenzhen Graduate School. His cur⁃ rent interests include deep space communications, networking and channel coding. Rana Abbas (rana.abbas@sydney.edu.au) is currently a PhD student at the Centre of Excellence in Telecommunications, School of Electrical And Information Engi⁃ neering, The University Sydney, Australia, where she is a recipient of the Australian Postgraduate Awards scholarship and the Norman 1 Price scholarship. She received her bachelor’s degree in electrical engineering from The University of Balamand, Lebanon in 2012 and her master’s degree in electrical engineering from The Uni⁃ versity of Sydney, Australia in 2013. Her research interests include error control codes, machine⁃to⁃machine communications, random multiple access, and coopera⁃ tive networks. LI Yonghui (yonghui.li@sydney.edu.au) received his PhD degree in 2002 from Bei⁃ jing University of Aeronautics and Astronautics. From 1999 to 2003 he was affiliat⁃ ed with Linkair Communication Inc., where he held the position of project manager with responsibility for the design of physical layer solutions for LAS⁃CDMA system. Since 2003 he has been with the Centre of Excellence in Telecommunications, the University of Sydney, Australia. He is now an associate professor at the School of Electrical and Information Engineering, University of Sydney. He is the recipient of the Australian Queen Elizabeth II Fellowship in 2008 and the Australian Future Fellowship in 2012. His current research interests are in the area of wireless com⁃ munications, with a particular focus on MIMO, cooperative communications, coding techniques, and wireless sensor networks. He holds a number of patents granted and pending in these fields. He is an executive editor for European Transactions on Tele⁃ communications (ETT). He received best paper awards at the IEEE International Conference on Communications (ICC) 2014 and the IEEE Wireless Days Conferenc⁃ es (WD) 2014. ZHANG Qinyu (zqy@hit.edu.cn) received his bachelor’s degree in communication engineering from Harbin Institute of Technology (HIT) in 1994, and PhD degree in biomedical and electrical engineering from the University of Tokushima, Japan, in 2003. From 1999 to 2003, he was an assistant professor with the University of Tokushima. From 2003 to 2005, he was an associate professor with the Shenzhen Graduate School, HIT, and was the founding director of the Communication Engi⁃ neering Research Center with the School of Electronic and Information Engineering. Since 2005, he has been a full professor, and serves as the dean of the EIE School. He is on the Editorial Board of some academic journals, such as The Journal on Communications, KSII Transactions on Internet and Information Systems, and Sci⁃ ence China: Information Sciences. He was the TPC Co⁃Chair of the IEEE/CIC ICCC’ 15, the Symposium Co⁃Chair of the IEEE VTC’16 Spring, an Associate Chair for Fi⁃ nance of ICMMT’12, and the Symposium Co ⁃ Chair of CHINACOM’11. He has been a TPC Member for INFOCOM, ICC, GLOBECOM, WCNC, and other flagship conferences in communications. He was the Founding Chair of the IEEE Communi⁃ cations Society Shenzhen Chapter. He has received the National Science Fund for Distinguished Young Scholars, the Young and Middle ⁃ Aged Leading Scientist of China, and the Chinese New Century Excellent Talents in University, and obtained three scientific and technological awards from governments. His research interests include aerospace communications and networks, wireless communications and net⁃ works, cognitive radios, signal processing, and biomedical engineering. BiographiesBiographies
  • 46. Multiple Access Technologies for Cellular MMultiple Access Technologies for Cellular M22MM CommunicationsCommunications Mahyar Shirvanimoghaddam and Sarah J. Johnson (School of Electrical Engineering and Computer Science, The University of Newcastle, NSW 2308, Australia) Abstract This paper reviews the multiple access techniques for machine⁃to⁃machine (M2M) communications in future wireless cellular net⁃ works. M2M communications aims at providing the communication infrastructure for the emerging Internet of Things (IoT), which will revolutionize the way we interact with our surrounding physical environment. We provide an overview of the multiple access strategies and explain their limitations when used for M2M communications. We show the throughput efficiency of different multi⁃ ple access techniques when used in coordinated and uncoordinated scenarios. Non⁃orthogonal multiple access (NOMA) is also shown to support a larger number of devices compared to orthogonal multiple access techniques, especially in uncoordinated sce⁃ narios. We also detail the issues and challenges of different multiple access techniques to be used for M2M applications in cellu⁃ lar networks. Internet of Things (IoT); massive access; machine⁃to⁃machine (M2M) communications; multiple access Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 006 http://www.cnki.net/kcms/detail/34.1294.TN.20161024.1001.002.html, published online October 24, 2016 1 Introduction achine⁃to⁃machine (M2M) communications is expected to become an integral part of cellular networks in the near future. In M2M communi⁃ cations a large number of multi⁃role devices, such as sensors and actuators, wish to communicate with each other and with the underlying data transport infrastructure. To enable such massive communication in wireless networks, ma⁃ jor shifts from current protocols and designs are necessary [1]. Current wireless networks that have been mainly designed and engineered for human⁃based applications, such as voice, vid⁃ eo, and data, cannot be used for M2M communications due to the different nature of their traffic and service requirements [2]. These differences have posed many questions and chal⁃ lenges in the communication society, in both industry and re⁃ search sectors. M2M communications aims at providing the communication infrastructure for emerging Internet of Things (IoT) and in⁃ volves the enabling of seamless information exchange between autonomous devices without any human intervention. M2M de⁃ vices can be either stationary, such as smart meters, or mobile, such as fleet management devices, and they can connect to the network infrastructure using either wired or wireless links. Key challenges of massive M2M communications can be listed as follows [3]: 1) Device cost: For the mass deployment of M2M communica⁃ tions, low cost devices are necessary for most use cases. 2) Battery life: Most M2M devices are battery operated and re⁃ placing batteries is not practical for many applications. 3) Coverage: Deep indoor and regional connectivity is a re⁃ quirement for many applications. 4) Scalability: Network capacity must be easily scaled to han⁃ dle a large number of devices forecasted to arise in the near future. 5) Diversity: Cellular systems must be able to support diverse service requirements for different use cases, ranging from static sensor networks to tracking systems. The wired solutions include cable, xDSL, and optical fiber, and can provide high reliability, high data rate, short delay, and high security. However, they are cost ineffective and do not support mobility and scalability; therefore, not appropriate for M2M applications [3]. On the other hand, Wireless capil⁃ lary (i.e., short range) solutions, such as WLAN and ZigBee, can provide low cost infrastructure and scalability for most M2M applications, but they suffer from small coverage, low da⁃ ta rate, weak security, and severe interference. Wireless cellu⁃ lar, i.e., GSM, GPRS, 3G, LTE⁃A, WiMAX, etc., however of⁃ fers excellent coverage, mobility and scalability support, and good security, and the fact that the infrastructure already exists M Special Topic October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS42
  • 47. makes it a promising solution for M2M communications [3]. Therefore, our focus in this paper is on wireless cellular solu⁃ tions for M2M communications. The mobile industry is standardizing several low power tech⁃ nologies, such as extended coverage GSM (EC⁃GSM), LTE for machine⁃type communication (LTE⁃M), and narrow band IoT (NB⁃IoT). Since GSM is still the dominant mobile technology in many markets, it is expected to play a key role in the IoT due to its global coverage and cost advantages. EC⁃GSM en⁃ ables coverage improvements of up to 20 dB with respect to GPRS on the 900 MHz band [4]. A combined capacity of up to 50,000 devices per cell on a single transceiver has been achieved by defining new control and data channels mapped over legacy GSM. LTE⁃M brings new power saving functional⁃ ity suitable for serving a variety of IoT applications, which ex⁃ tend battery life to 10 years or more. NB⁃IoT is a self contained carrier that can be deployed with a system bandwidth of 200 kHz. These initiatives were undertaken in 3GPP Release 13 for M2M specific applications [3]. Despite all these efforts, further improvements are required in the way that devices communicate with the base station to support a large number of devices and not jeopardizing the hu⁃ man⁃based communication quality. The multiple access (MA) techniques have been identified as a key area where improve⁃ ments for M2M communications are needed. The fact that the radio access strategy in LTE is still based on random access mechanisms turns it into a potential bottleneck for the perfor⁃ mance of cellular networks when the number of M2M devices grows [5]. Moreover, radio resources are orthogonally allocated to the users/devices in the current LTE standards, which is not effective for M2M communications when the number of devic⁃ es goes very large, due to the limited number of radio resourc⁃ es [6]. In this paper, we consider several multiple access technolo⁃ gies and show their performance in coordinated and uncoordi⁃ nated scenarios. Overall, coordinated strategies outperform un⁃ coordinated ones as in coordinated strategies the base station can optimally allocate the radio resources between the devices and support a larger number of devices. We also show that the non⁃orthogonal multiple access (NOMA) scheme achieves the highest throughput in both coordinated and uncoordinated strategies, whereas frequency division multiple access (FDMA) has comparable performance in coordinated scenarios. This suggests that FDMA can be effectively used in coordinated sce⁃ narios to achieve maximum throughput (this has been consid⁃ ered by 3GPP for M2M communications in the NB⁃IoT solu⁃ tion), while in uncoordinated scenarios, NOMA strategies must be considered to effectively support a large number of devices and use the available radio resources in an efficient manner. The remainder of the paper is organized as follows. Section 2 represents the unique characteristics of M2M communications and its challenges in cellular networks. In Section 3, we pro⁃ vide an overview on different multiple access technologies. Co⁃ ordinated and uncoordinated MA techniques are represented in Section 4 and 5, respectively, where we characterize their maximum achievable throughput. Practical issues for imple⁃ menting MA techniques for M2M communications are present⁃ ed in Section 6. Finally, Section 7 concludes the paper. 2 M2M Communications: Characteristics and Challenges Until recently, cellular systems have been designed and en⁃ gineered for human based applications, such as voice, video, and data, with a higher demand on downlink. M2M communi⁃ cations however has different traffic characteristics that in⁃ clude small and infrequent data generated from a very large number of devices, which imposes a higher traffic volume on the uplink. In addition, M2M applications have very diverse service requirements. For instance, in alarm signal applica⁃ tions, a small⁃size message must be delivered to the base sta⁃ tion (BS) within 10 ms, while in other applications, such as smart metering, the delay of up to several hours or even a day is tolerable [7]. Due to limited radio resources and the large number of de⁃ vices involved in M2M communications, wireless networks should minimize the time wasted due to collisions or exchang⁃ ing control messages. The throughput must be large enough to support a large number of devices. Control overhead must be minimized as the payload data in many M2M applications is of small size and the control overhead of conventional approaches in current cellular systems results in an inefficient M2M com⁃ munications [8]. In fact, if the control overhead of a protocol is large, the effective throughput is degraded even though the physical data rate may not be affected. It is also required that the effective throughput remain high irrespective of the traffic level [9]. Scalability is another challenge in M2M communications as it is expected that a large number of devices arise in M2M sce⁃ narios. These devices have dynamic behaviour, i.e., entering and leaving the network frequently; thus the network must easi⁃ ly tolerate the changes in the node density with little control in⁃ formation exchange. Energy efficiency is also one of the most important challenges in M2M communications, as devices in many M2M applications are battery operated and long life times are expected for these devices [10]. More specifically, the energy spent on radio access and data transmission in M2M communications must be minimized to improve the ener⁃ gy efficiency in a large scale. For instance, in high load scenar⁃ ios, exchanging control information may consume more than 50% of the total energy in IEEE 802.11 MAC protocol, which shows its ineffectiveness in dense M2M applications [9]. In many M2M applications, the network latency is a critical factor that determines the effectiveness of the service. For in⁃ stance, in intelligent transportation systems and healthcare monitoring, it is highly important to make the communication Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 43
  • 48. reliable and fast. Channel access delay then needs to be mini⁃ mized to reduce the overall latency in M2M communications. Moreover, in cellular systems, human⁃to⁃human (H2H) devices coexist with M2M devices, and the communication protocol must be designed in such a way to not jeopardize the quality of human⁃based communications. Resource management and allo⁃ cation are challenging tasks in M2M communications which co⁃ exist with H2H applications, as H2H applications have com⁃ pletely different service requirements [11]. These unique characteristics of M2M communications intro⁃ duce a number of networking challenges in cellular networks. The fundamental issue arises from the fact that most M2M ap⁃ plications involve a huge number of devices. The question is then how the available radio resources have to be shared among devices such that their service requirements are simul⁃ taneously met. 3 Overview of Multiple Access Techniques for M2M Communications Multiple access techniques can be divided into two broad categories, depending on how the radio resources are allocated to the devices. These include 1) uncoordinated, where the de⁃ vices transmit data using slotted random access and there is no need to establish dedicated resources, and 2) coordinated, where devices transmit on separate resources pre⁃allocated by the base station. In coordinated MA, the base station knows a priori the set of devices that have data to transmit. The BS can also acquire channel state information (CSI) of these devices based on which it allocates resources to optimize system throughput. CSI to the devices can be obtained by each device sending an upload pilot signal. Multiple access techniques can be also divided into orthogo⁃ nal and non⁃orthogonal approaches. In orthogonal MA (OMA), radio resources are orthogonally divided between devices, where the signals from different devices are not overlapped with each other. Instances of OMA (Fig. 1) are time division multiple access (TDMA), frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFD⁃ MA), and single carrier FDMA (SCFDMA). First and second generation cellular systems are mainly developed using OMA approaches, which avoid intra ⁃ cell interference and simplify air interface design. However, OMA approaches have no abili⁃ ty to combat the inter⁃cell interference; therefore careful cell planning and interference management techniques are re⁃ quired to solve the interference problem [12]. Non⁃orthogonal MA (NOMA) techniques have been adopted in second and third generation cellular systems. NOMA allows overlapping among the signals from different devices by ex⁃ ploiting power domain, code domain, and interleaver pattern. Code division multiple access (CDMA) is the well⁃known exam⁃ ple of NOMA which has been adopted in second and third gen⁃ eration cellular systems. CDMA is robust against inter⁃cell in⁃ terference, but suffers from intra⁃cell interference [12]. CDMA is also not suitable for data services which require high single⁃ user rates. Rather than CDMA which exploits code domain, NOMA in current study in general exploits power domain. NO⁃ MA is also shown to provide better performance than OMA [12]. In NOMA, signals from multiple users are superimposed in the power⁃domain and successive interference cancellation (SIC) is used at the BS to decode the messages. It is also shown that NOMA can achieve the multiuser capacity region both in the uplink and downlink [12]. In this paper, we compare NOMA and OMA strategies in both coordinated and uncoordinated scenarios, and show that NOMA can provide the system with higher capacity to support M2M devices, especially in the uncoordinated scenario. This is achieved by exploiting the power domain, rather than frequency⁃ domain or time⁃domain as in FDMA and TDMA, respectively. CDMA: code division multiple access FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple access Figure 1. ▶ Different multiple access schemes. Time Power TDMA Frequency (a) Time Power FDMA Frequency (b) Time Power CDMA Frequency (c) Time Power NOMA Frequency (d) October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS44 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
  • 49. For the analysis in this paper, we consider a single cell cen⁃ tered by base station and devices uniformly distributed around it in a circular region with radius R . The uplink load seen by the base station is modeled by a Poisson point process with mean λ arrivals per second. We further assume a time slotted system with a slot duration of τs . We perform our analysis on a typical radio resource with slot duration τs and bandwidth W . Each device packet is assumed to have a payload of L bits. The channel from a device located at distance r from the base station is modelled by g =(r/R) -γ , where γ denotes the path loss exponent and we ignore shadowing and small scale fading [13]. The received signal⁃to⁃noise ratio (SNR) for a de⁃ vice transmitting with power Pt over bandwidth W is then given by [14]: μr = Pt Pmax W Wt μg , (1) where Pmax is the maximum transmit power and μ is the ref⁃ erence SNR, defined as the average received SNR from a de⁃ vice transmitting at maximum power Pmax over bandwidth W located at the cell edge. Without loss of generality, we assume ordered channel gain g1 ≥ g2 ≥ … ≥ gK , where K is the num⁃ ber of devices. 4 Coordinated Multiple Access Strategies In this section, we consider the coordinated multiple access strategies, i.e., TDMA, FDMA, and NOMA, and compare their throughput efficiency. In this section, we assume that the BS has perfect CSI to all the devices. 4.1 Optimal Throughput FDMA Strategy In FDMA, the spectrum is partitioned between the devices and each device will transmit in a portion of the spectrum. Fig. 1b shows the FDMA strategy, where the whole spectrum has been divided between 6 devices, and each device will use its allocated bandwidth for the data transmission. Using Shannon’s capacity formula, the minimum bandwidth required for the transmission of L bits by the ith device over time τs is given by the solution of the following equation [13]: L τsWmini = log2 æ è çç ö ø ÷÷1 + μ W Wmini gi . (2) The maximum load that can be supported in a resource block of duration τs and bandwidth W is given by: Kmax = max ì í î ü ý þ K:∑i = 1 K Wmini ≤ W . (3) 4.2 Optimal Throughput TDMA Strategy In TDMA, the whole spectrum is used by each device in sep⁃ arate time instances. Fig. 1a shows the TDMA scheme, where the same time duration is allocated for 6 devices, and each de⁃ vice will only transmit in its allocated time slot using the whole spectrum. TDMA is an interesting MA strategy due to its sim⁃ plicity, but it is not efficient for M2M applications with a large number of devices. Moreover, with increasing the number of devices, each device’s transmission will be delayed which is not appropriate for delay⁃sensitive M2M applications. Assuming a capacity approaching code and using Shannon’ s capacity equation, the time required for a device located at distance r from the base station to deliver its packet to the destination is given by [13]: τ ≥ L W log2(1 + μr) , (4) and the minimum time required to deliver the message is ob⁃ tained when the device is transmitting with full power Pmax : τmini = L W log2(1 + μgi) . (5) Similar to FDMA, the maximum number of devices which can be supported in a resource block of duration τs and band⁃ width W can then be found as follows: Kmax = max ì í î ü ý þ K:∑i = 1 K τmini ≤ τs . (6) 4.3 Optimal Throughput NOMA Strategy Unlike TDMA and FDMA, devices in the NOMA strategies are assumed to transmit in the same resource block and their transmissions interfere with each other. We assume that the BS perform successive interference cancellation (SIC), where it starts the decoding with the device with the largest channel gain and treats the signals from other devices as additive noise. After decoding the first device, its signal will be removed from the received signal and the BS continues the decoding for the second device and treats the remainder as additive noise. This process is continued until all the devices are successfully de⁃ coded. Under this decoding strategy, the Shannon Capacity for⁃ mula for the ithdevice is given by: L = Wτs log2 æ è ç ç ö ø ÷ ÷ 1 + Pi μgi 1 +∑j = i + 1 K Pi μgj , (7) and the required transmit power can be calculated as follows: Pi μgi = æ è çç ö ø ÷÷2 L Wτs - 1 æ è çç ö ø ÷÷1 + ∑j = i + 1 K Pi μgj . (8) By substituting, i = K , we have: PK = 2 L Wτs - 1 μgK , (9) Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 45
  • 50. and by going backwards and finding the transmit power for the ith device, we have: Pi = 2 (K - i)L Wτs æ è çç ö ø ÷÷2 L Wτs - 1 μgi . (10) The maximum load that the BS can support in a resource block of bandwidth W and duration τs can be found as fol⁃ lows: Kmax = max{ }K:Pi ≤ Pmax for i = 1,2,…,K . (11) 4.4 Comparison Between Coordinated MA Techniques Fig. 2 shows the maximum throughput versus arrival rates for different coordinated MA techniques. NOMA can achieve very high throughput when the arrival rate is very large. FDMA performs very close to the NOMA strategy and can support all the active devices for the arrival rates up to 14,000 packets per second. The advantage of NOMA comes from the fact that the devices can use the whole spectrum thus achieving a higher throughput compared to FDMA. Only a fraction of the spec⁃ trum is used by each device in FDMA. Also, TDMA cannot support many devices, which shows that it is not an effective MA strategy for M2M communications. It is clear that the time slot duration τi and subchannel bandwidth Wi cannot be arbitrarily small in TDMA and FD⁃ MA, respectively. As can be seen in Fig. 2, if we put some con⁃ straints on the minimum time slot duration or subchannel band⁃ width, the number of devices which can be supported by FD⁃ MA and TDMA would be limited. For example, if the minimum time slot duration for TDMA is set to be 1 ms, the maximum number of devices which can be supported in a time slot of du⁃ ration 1 s is 1000. Similarly, if the minimum subchannel band⁃ width in FDMA is set to be 1 kHz, the maximum number of de⁃ vices which can be supported by the BS will be 1000. This shows that in practical systems where the minimum subchan⁃ nel bandwidth and time slot duration cannot be very small, the maximum throughput of TDMA and FDMA will be limited. In such cases, NOMA can bring more benefits to the system as it can support a larger number of devices without dividing the ra⁃ dio resource into subchannels or time slots. 5 Uncoordinated Multiple Access Strategies In this section, we assume that the base station does not have CSI to the devices, which is particularly the case for M2M communications with a large number of devices, where it is almost impractical for the base station to estimate the chan⁃ nel to every device with random activities. The only informa⁃ tion we assume is available at the BS is the traffic load which can be obtained using different load estimation algorithms. 5.1 Uncoordinated FDMA In this scheme, we assume that the base station chooses a se⁃ lection probability pc and broadcasts this information to the devices. Each device which has data to transmit only switches on its transmitter with probability pc . We refer to these devic⁃ es as active devices. Let Nc denote the number of active de⁃ vice. We further assume that the BS uniformly divides the spectrum into Nw subchannels, and each device randomly chooses a subchannel for its transmission. We also assume that each device only transmits on a selected subchannel if the max⁃ imum transmit power required to deliver its message to the BS is less than Pmax , assuming no collision on the selected sub⁃ channel. More specifically, the ith device is transmitting in a subchannel if the following condition holds: æ è ç ç ö ø ÷ ÷2 LNw Wτs - 1 ≤ Nw μgi . (12) Therefore, the probability that a device is transmitting can be calculated as follows: P æ è ç ç ö ø ÷ ÷ æ è ç ç ö ø ÷ ÷2 LNw Wτs - 1 ≤ Nw μgi = æ è ç çç ç ö ø ÷ ÷÷ ÷ Nw μ 2 LNw Wτs - 1 2 γ , (13) which is due to the fact that the devices are uniformly distribut⁃ ed in the cell and the probability that a device is located at dis⁃ tance r is given by 2r R2 . The average number of active de⁃ vices which can deliver their messages, considering no colli⁃ sion, can be found as follows: Np = Nc æ è ç çç ç ö ø ÷ ÷÷ ÷ Nw μ 2 LNw Wτs - 1 2 γ . (14) As the devices randomly choose a sub ⁃ channel for their FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple acces ▲Figure 2. Average throughput versus arrival rates for different coordi⁃ nated MA techniques. Total available bandwidth is W = 1 MHz, time slot duration is τs = 1 sec, and the packet length is L = 1000 bits. 18􀆯000 16􀆯000 14􀆯000 12􀆯000 10􀆯000 8􀆯000 6000 4000 2000 0 104 103 102 101 Arrival rate (λ) Throughput(packets/s) FDMA NOMA TDMA FDMA, Wmin =1 KHz TDMA, τmin =1 ms October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS46 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
  • 51. transmission, more than one device can select the same sub⁃ channel, which leads to collision. The base station cannot de⁃ code any of the devices that are simultaneously transmitting on that particular subchannel. The probability of collision can be calculated as follows [14]: Pc = 1 - æ è ç ö ø ÷1 - 1 Nw Np - 1 . (15) The average number of devices which can successfully deliv⁃ er their messages to the BS is given by NP(1 - Pc) . We assume that the BS finds the optimal values for pc and Nw such that the number of devices which can be supported by the BS is maximized. 5.2 Uncoordinated TDMA Similar to FDMA, we assume that the BS assigns an access probability pc to the devices. Let Nc denote the number of ac⁃ tive device. We further assume that the BS uniformly divides the time into Nt time slots, and each device randomly chooses a time slot for its transmission. We also assume that the each device only transmits in a selected time slot if the maximum transmit power required to deliver its message to the BS is less than Pmax , assuming no collision on the selected time slot. More specifically, the ith device is transmitting in a time slot, if the following condition holds: æ è ç ç ö ø ÷ ÷2 LNt Wτs - 1 ≤ μgi . (16) Therefore, the probability that a device is transmitting can be calculated as follows: p æ è ç ç ö ø ÷ ÷ æ è ç ç ö ø ÷ ÷2 LNt Wτs - 1 ≤ μgi = æ è ç çç ç ö ø ÷ ÷÷ ÷ μ 2 LNt Wτs - 1 2 γ , (17) which is due to the fact that the devices are uniformly distribut⁃ ed in the cell and the probability that a device is located at dis⁃ tance r is given by 2r/R2 . The average number of active de⁃ vices which can deliver their messages, considering no colli⁃ sion, can be found as follows: Np = Nc æ è ç çç ç ö ø ÷ ÷÷ ÷ μ 2 LNt Wτs - 1 2 γ . (18) The average number of devices which can successfully deliv⁃ er their messages to the BS is given by NP(1 - Pc) , where Pc is given by (15) by replacing Nw with Nt . We assume that the BS finds the optimal values for Pc and Nt such that the num⁃ ber of devices which can be supported by the BS is maximized. 5.3 Uncoordinated NOMA We consider that each device performs power control such that the received SNR at the BS for each device is γ0 . A de⁃ vice will only transmit if and only if the transmit power re⁃ quired to achieve the SNR γ0 at the base station is less than Pmax . Let Np denote the number of devices which can trans⁃ mit, i.e., their required transmit power is less than Pmax . The achievable rate for the devices considering the successive in⁃ terference cancellation at the BS can be calculated as follows: Rmin = log2 æ è çç ö ø ÷÷1 + γ0 1 +(Np - 1)γ0 . (19) A message of length L can be delivered by Np devices if WτsRmin ≥ L . Using (19), the required SNR γ0 for successfu⁃ lly delivering a message of length L at the BS is derived as fol⁃ lows: γ0 = 1 1 2 L Wτs - 1 - Np . (20) Accordingly, the number of devices which can be supported at the BS is upper bounded as follows: Np ≤ 1 2 L Wτs - 1 . (21) 5.4 Comparison Between Uncoordinated MA Techniques Fig. 3 shows the maximum number of devices which can be supported by the base station versus different arrival rates for uncoordinated MA strategies. The minimum time slot duration FDMA: frequency division multiple access NOMA: non⁃orthogonal multiple access TDMA: time division multiple acces ▲Figure 3. Average throughput versus arrival rates for different unco⁃ ordinated MA techniques. Total available bandwidth is W = 1 MHz, time slot duration is τs = 1 s, and the packet length is L = 1000 bits. The minimum time slot duration for TDMA is considered to be 1 ms and the minimum subchannel bandwidth in FDMA is considered to be 1 kHz. 1400 104 Arrival rate (λ) Throughput(packets/s) 103 102 101 1200 1000 800 600 400 200 NOMA TDMA FDMA Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 47
  • 52. for TDMA is considered to be 1 ms, which corresponds to Nt = 1000 , and the minimum subchannel bandwidth in FDMA is considered to be 1 kHz, which corresponds to Nw = 1000 . As shown in this figure, NOMA can support much larger num⁃ ber of devices compared to the FDMA and TDMA strategies. This is due to the high collision probability in uncoordinated FDMA and TDMA in high arrival rates, while in NOMA a large number of devices can simultaneously transmit in the same resource block by exploiting the power domain. This shows the advantage of NOMA in uncoordinated scenarios. Therefore, NOMA can be an excellent choice for M2M applica⁃ tions with a large number of devices and random traffic. More⁃ over, FDMA outperforms TDMA in moderate loads but they perform similarly in low and high arrival rates. It is important to note that in NOMA the constraints on mini⁃ mum time slot duration or subchannel bandwidth do not affect the throughput efficiency. This is due to the fact that in NOMA all the devices are transmitting in the whole bandwidth in all slot duration. One could consider some limitations in the mini⁃ mum power difference between the devices, which mostly de⁃ pends on the hardware capability to distinguish different power levels which is out of scope of this paper. 6 Practical Considerations of Massive NOMA for M2M Communications and Future Directions NOMA can bring many benefits to cellular systems which in⁃ clude, but are not limited to, the following. NOMA can effec⁃ tively use the spectrum and provide higher throughput by ex⁃ ploiting power domain and non⁃orthogonal multiplexing. It also provides robust performance gain in high mobility scenarios. NOMA is also compatible with OFDMA and can be applied on top of OFDMA for downlink and SC⁃FDMA for uplink. It can be also combined with multi ⁃ antenna techniques to improve the system performance. Using NOMA, multiple users can si⁃ multaneously transmit in the same subband without being iden⁃ tified by the destination a priori. The devices can attach their terminal identities to their messages and the base station can identify the devices after decoding their messages. The RA pro⁃ cedure can be eliminated and therefore the access delay and signaling overhead will be significantly reduced [12]. Although NOMA can improve spectrum efficiency and sys⁃ tem capacity, there are many practical challenges for this tech⁃ nology to be potentially used in real wireless systems for M2M communications. Here, we outline the main practical consider⁃ ation of massive NOMA for M2M communications. First, in uncoordinated strategies the base station needs to estimate the arrival rate to effectively detect the devices. In un⁃ coordinated FDMA, the BS needs to know the number of devic⁃ es to find the optimal access probability and the number of sub⁃ bands. In NOMA, the problem is much more complicated as the BS runs the SIC and needs to know the number of devices with different power levels. For simplicity, one could consider that the devices perform power control such that only one pow⁃ er level is received at the BS, but this may have some implica⁃ tions on the actual performance of the system as the overall sys⁃ tem data rate will be dominated by the device with the lowest SINR; and thus will not effectively use the available spectrum. However, even with this simplification and suboptimal power allocations, NOMA outperforms FDMA in uncoordinated sce⁃ narios and can support a large number of devices under high loads. Second, channel estimation at the devices is necessary in un⁃ coordinated strategies employing NOMA techniques. This is due to the fact that the devices are not identified by the BS be⁃ forehand and they are simultaneously transmitting at the same resource block. To enable the BS to detect the devices and de⁃ code their messages, the devices need to perform channel esti⁃ mation and adjust their power so the BS only deals with some known power levels rather than unknown channel gains. On the other hand, to effectively perform SIC, the multipath effect must be carefully taken into consideration as multipath will spread the signal over time, which decreases the effective sig⁃ nal to noise ratio for each device, and makes the BS unable to perform SIC. One can consider several techniques, such as time reversion [15], to eliminate the multipath effect by treat⁃ ing the channel between each device and the BS as the natural match filter. This has been shown an effective way to combat multipath effect for several fixed location M2M applications [16]. Third, NOMA requires synchronization among the devices at the symbol level. This is very challenging as providing time synchronicity between a large number of devices distributed in a large environment is tedious. However, the devices in many M2M applications are deployed in fixed locations, so each de⁃ vice can determine its propagation delay using different dis⁃ tance estimation strategies or using control information periodi⁃ cally sent by the BS. Fourth, as the number of devices transmitting in each re⁃ source block in uncoordinated NOMA is random, the physical data rate cannot be determined beforehand. One could consid⁃ er a very low rate code at each device, but it might be ineffi⁃ cient when used in low⁃to⁃moderate loads. An effective strate⁃ gy is then to use rateless codes to automatically adapt to the traffic condition. Authors in [17] have proposed to use analog fountain codes to enable massive multiple access for M2M communications and achieve very high throughput even in high loads. Moreover, as shown in [18], binary rateless codes can be effectively used to enable NOMA for M2M communica⁃ tions. These coding strategies were mainly proposed to maxi⁃ mize the throughput in M2M communications and for delay sensitive applications with very short messages, more ad⁃ vanced coding techniques should be combined with rateless ideas to enable low latency massive multiple access in M2M communications. October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS48 Special Topic Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson
  • 53. Last but not least, NOMA is still in its early stage of its de⁃ velopment and more research work must be done to clearly identify its effectiveness in real scenarios. From an information theoretic point of view, it achieves the capacity region of the multiple access channel and thus is optimal in terms of throughput. But in real M2M applications when NOMA is joint⁃ ly considered with medium access control layer in real world scenarios, it might not be as efficient as OMA techniques, which have been considered as effective multiple access tech⁃ niques for a long time and several issues and challenges have been solved over the years. 7 Conclusions In this paper, we provided an overview of multiple access techniques for emerging machine⁃to⁃machine communications in cellular systems. The unique challenges of M2M communi⁃ cations were represented, where we identified scalability, ener⁃ gy efficiency, and reliability, as the most important features for every potential multiple access technology which is considered for M2M communications. We provided a simple study on the throughput efficiency of multiple access techniques in both co⁃ ordinated and uncoordinated scenarios. NOMA was shown to provide the highest throughput in both coordinated and uncoor⁃ dinated scenarios, whereas FDMA has shown comparable per⁃ formance with NOMA in coordinated scenarios. NOMA is shown to be scalable in uncoordinated scenarios and can sup⁃ port a large number of devices. It can be also combined with different access management schemes to control the load over the base station. We also provided some of the practical issues in NOMA which needed to be considered for the use of NOMA strategies for M2M communications in future cellular systems. References [1] H. Shariatmadari, R. Ratasuk, S. Iraji, et al.,“Machine⁃type communications: current status and future perspectives toward 5G systems,”IEEE Commununica⁃ tions Magazine, vol. 53, no. 9, pp. 10-17, Sept. 2015. doi: 10.1109/MCOM.2015. 7263367. [2] G. Wu, S. Talwar, K. Johnsson, N. Himayat, and K. Johnson,“M2M: From mo⁃ bile to embedded internet,”IEEE Commununications Magazine, vol. 49, no. 4, pp. 36-43, Apr. 2011. doi: 10.1109/ MCOM.2011.5741144. [3] Ericsson. (Jan. 2016). Cellular networks for massive IoT. Tech. Rep. Uen 284 23⁃ 3278 [Online]. Available: https://www.ericsson.com/res/docs/whitepapers/wp iot. pdf [4] Service Requirements for Machine⁃Type Communications (MTC); Stage 1, 3GPP TS 22.368 V.13.0.0, Jun. 2014. [5] A. Laya, L. Alonso, and J. Alonso⁃Zarate,“Is the random access channel of LTE and LTE⁃A suitable for M2M communications? a survey of alternatives,”IEEE Commununications Surveys & Tutorials, vol. 16, no. 1, pp. 4-16, 2014. doi: 10.1109/SURV.2013.111313.00244. [6] G. Naddafzadeh⁃Shirazi, L. Lampe, G. Vos, and S. Bennett,“Coverage enhance⁃ ment techniques for machine ⁃ to ⁃ machine communications over LTE,”IEEE Communications Magazine, vol. 53, no. 7, pp. 192-200, Jul. 2015. doi: 10.1109/ MCOM.2015.7158285. [7] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi,“Internet of things for smart cities,”IEEE Internet of Things Jounral, vol. 1, no. 1, pp. 22-32, Feb. 2014. doi: 10.1109/JIOT.2014.2306328. [8] D. Wiriaatmadja and K. W. Choi,“Hybrid random access and data transmission protocol for machine⁃to⁃machine communications in cellular networks,”IEEE Transactions on Wireless Communications, vol. 14, no. 1, pp. 33-46, Jan. 2015. doi: 10.1109/TWC.2014.2328491. [9] A. Rajandekar and B. Sikdar,“A survey of MAC layer issues and protocols for machine⁃to⁃machine communications,”IEEE Internet of Things Journal, vol. 2, no. 2, pp. 175-186, Apr. 2015. doi: 10.1109/JIOT.2015.2394438. [10] M. Hasan, E. Hossain, and D. Niyato,“Random access for machine⁃to⁃machine communication in LTE ⁃ advanced networks: issues and approaches,”IEEE Communications Magazine, vol. 51, no. 6, pp. 86-93, Jun. 2013. doi: 10.1109/ MCOM.2013.6525600. [11] K. Zheng, S. Ou, J. Alonso⁃Zarate, et al.,“Challenges of massive access in high⁃ ly dense LTE⁃advanced networks with machine⁃to⁃machine communications,” IEEE Wireless Communications Magazine, vol. 21, no. 3, pp. 12-18, Jun. 2014. doi: 10.1109/MWC.2014.6845044. [12] A. Li, Y. Lan, X. Chen, and H. Jiang,“Non⁃orthogonal multiple access (NO⁃ MA) for future downlink radio access of 5G,”China Communications, vol. 12, Supplement, pp. 28-37, Dec. 2015. [13] H. S. Dhillon, H. C. Huang, H. Viswanathan, and R. A. Valenzuela,“Power⁃ef⁃ ficient system design for cellular⁃based machine⁃to⁃machine communications,” IEEE Transactions on Wireless Communications, vol. 12, no. 11, pp. 5740- 5753, Nov. 2013. doi: 10.1109/TWC.2013.100713.130025. [14] H. S. Dhillon, H. C. Huang, H. Viswanathan, and R. A. Valenzuela,“Funda⁃ mentals of throughput maximization with random arrivals for M2M communica⁃ tions,”IEEE Transactions on Communications, vol. 62, no. 11, pp. 4094-4109, Nov. 2014. doi: 10.1109/TCOMM.2014.2359222. [15] B. Wang, Y. Wu, F. Han, Y. H. Yang, and K. J. R. Liu,“Green wireless com⁃ munications: A time ⁃ reversal paradigm,”IEEE Journal on Selected Areas in Communications, vol. 29, no. 8, pp. 1698-1710, Sept. 2011. doi: 10.1109/JSAC. 2011.110918. [16] Y. Chen, F. Han, Y. H. Yang, et al.,“Time ⁃ reversal wireless paradigm for green internet of things: An overview,”IEEE Internet of Things Jounral, vol. 1, no. 1, pp. 81-98, Feb. 2014. doi: 10.1109/JIOT.2014.2308838. [17] M. Shirvanimoghaddam, Y. Li, M. Dohler, B. Vucetic, and S. Feng,“Probabilis⁃ tic rateless multiple access for machine ⁃ to ⁃ machine communication,”IEEE Transactions on Wireless Communications, vol. 14, no. 12, pp. 6815 - 6826, Dec. 2015. doi: 10.1109/TWC.2015.2460254. [18] M. Shirvanimoghaddam, S. J. Johnson, and M. Dohler. (2016). An efficient mas⁃ sive access strategy based on superposition Raptor codes for M2M communica⁃ tions. CoRR [Online]. Available: http://arxiv.org/pdf/1602.05671v1.pdf Manuscript received: 2016⁃06⁃30 Mahyar Shirvanimoghaddam (fmahyar.shirvanimoghaddam@newcastle.edu.au) re⁃ ceived the BSc degree with 1st Class Honours from University of Tehran, Iran, in September 2008, the MSc degree with 1st Class Honours from Sharif University of Technology, Iran, in October 2010, and the PhD degree from The University of Syd⁃ ney, Australia, in January 2015, all in electrical engineering. He then held a re⁃ search assistant position at the Centre of Excellence in Telecommunications, School of Electrical and Information Engineering, The University of Sydney, before coming to the University of Newcastle, Australia, where he is now a postdoctoral research as⁃ sociate at the School of Electrical Engineering and Computer Science. His general research interests include channel coding techniques, cooperative communications, compressed sensing, machine⁃to⁃machine communications, and wireless sensor net⁃ works. Sarah Johnson (sarah.johnsong@newcastle.edu.au) received the BE (Hons) degree in electrical engineering in 2000, and PhD in 2004, both from the University of Newcastle, Australia. She then held a postdoctoral position with the Wireless Signal Processing Program, National ICT Australia before returning to the University of Newcastle where she is now an Australian Research Council Future Fellow. Her re⁃ search interests are in the fields of error correction coding and network information theory. She is the author of a book on iterative error correction published by Cam⁃ bridge University Press. BiographiesBiographies Multiple Access Technologies for Cellular M2M Communications Mahyar Shirvanimoghaddam and Sarah J. Johnson Special Topic October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 49
  • 54. Software Defined OpticalSoftware Defined Optical Networks and ItsNetworks and Its Innovation EnvironmentInnovation Environment LI Yajie1 , ZHAO Yongli1 , ZHANG Jie1 , WANG Dajiang2 , and WANG Jiayu2 (1. Beijing University of Posts and Telecommunications, Beijing 100876, China) 2. ZTE Corporation, Shenzhen 518057, China) Software defined optical networks (SDONs) integrate software defined technology with optical communication networks and represent the promising development trend of future optical networks. The key technologies for SDONs include software⁃ defined optical transmission, switching, and networking. The main features include control and transport separation, hard⁃ ware universalization, protocol standardization, controllable optical network, and flexible optical network applications. This paper introduces software defined optical networks and its innovation environment, in terms of network architecture, protocol extension solution, experiment platform and typical applications. Batch testing has been conducted to evaluate the performance of this SDON testbed. The results show that the SDON testbed has good scalability in different sizes. Meanwhile, we notice that controller output bandwidth has great influence on lightpath setup delay. optical networks; software defined networking; innovation en⁃ vironment DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 007 http://www.cnki.net/kcms/detail/34.1294.TN.20160928.1048.002.html, published online September 28, 2016 Abstract Keywords W 1 Introduction ith the emerging of new network services, the interaction of all kinds of information grows day by day. It is an eternal theme for optical networks to satisfy the transmission demands for high speed, wide broadband, large capacity, and long⁃dis⁃ tance transmission. The changes of servics properties brings a new challenge: intelligence of optical networks. For example, high burst services require optical networks to have dynamic adaptability; large⁃scale networking requires optical networks to be scalable; and variable bandwidth provisioning requires optical networks to be flexible. To realize the intelligent opti⁃ cal network, the industry has carried out a long⁃term research and exploration. So far, the intelligent optical network has gone through three important stages of development. 1) Automatic Switching Optical Networks (ASON) An ASON is divided into three planes: the transmission plane, control plane and management plane. With the control plane based on Generalized Multi ⁃ Protocol Label Switching (GMPLS) protocol, ASON adopts distributed signaling and rout⁃ ing to solve the connection control problem and satisfy the function demands of automatic switching [1]- [4]. However, ASON has obvious limitations in many aspects, including large⁃ scale connection control, complex path calculation, network openness, devices interworking, and cost reduction. Besides, the GMPLS standard is very complex, which greatly affects the application and promotion of ASON. 2) Path Computation Element (PCE) Architecture for ASON In order to better adapt to the characteristics of multi⁃layer multi⁃domain large⁃scale optical networks, the Internet Engi⁃ neering Task Force (IETF) separates the path calculation func⁃ tion from the control layer and develops an independent unit, i.e., PCE [5]-[8]. In order to satisfy the function demands of large⁃scale multi⁃layer/domain, PCE adopts the distributed sig⁃ naling and centralized routing to solve the problem of path se⁃ lection and calculation for inter⁃layer and inter⁃domain path. However, with unitary function of path calculation, PCE needs to cooperate with other technology in applications. 3) Software Defined Optical Networks (SDONs) SDONs can offer a unified schedule and control for various kinds of optical layer resources according to the requirements of users or operators. With programmable software and dynam⁃ ic customization, the SDON solves the problem of function ex⁃ tension and therefore realizes rapid response to requests, effi⁃ cient utilization of resources and flexible service provisioning. The SDON well supports service processing, control strategy and programmable transmission device, which achieves pro⁃ grammable tuning of optical network elements [9]. Therefore, the SDON is more suitable for multi⁃layer/domain and multi⁃ constraint optical networks, and it can effectively improve oper⁃ ational efficiency and reduce cost. The article introduces the SDON innovation environment from the perspectives of architecture, protocol extension, exper⁃ imental platform and typical applications. Section 2 describes the hierarchical control architecture and the process of cross⁃ Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS50 This work was supported by ZTE Industry⁃Academia⁃Research Cooperation Funds under Grant No. Surrey⁃Ref⁃9953.
  • 55. domain connection provisioning in detail. Section 3 depicts the workflow of connection provisioning in multi ⁃ domain optical networks. Section 4 shows the necessary extension work of OpenFlow 1.3 protocol for optical networks. The experimental environment and typical applications of SDON are respectively discussed in section 5 and section 6. Section 7 is performance evaluation of the SDON testbed and the last section summariz⁃ es the paper. 2 Hierarchical Architecture for SDON As shown in Fig. 1, considering the cross⁃layer distribution of multi⁃domain optical network resources, this paper proposes an OpenFlow enabled hierarchical control architecture in or⁃ der to solve the problem of programmable control in optical net⁃ work. With the advantage of software⁃defined networking, the architecture uniformly abstracts optical transmission network resources and content resources, and provides them with the multi ⁃ domain controller through the northbound interface. In this way, the uniform control of cross⁃layer resources is real⁃ ized. The hierarchical architecture consists of three layers: the physical layer, control layer and application layer. 1) The physical layer mainly includes data⁃center and inter⁃ data⁃center optical transmission networks. OpenFlow⁃enabled IP Routers (OF ⁃ Router) and optical transmission equipment with OpenFlow agents (OF⁃ROADM) are deployed in the net⁃ work. 2) The application layer mainly includes various applica⁃ tions such as dynamic migration of virtual machines, virtual network provisioning, and spectrum defragmentation. It is con⁃ nected with the control layer through the Restful API interface. All the service requests are triggered from this layer. 3) The control layer is mainly composed of optical control⁃ lers and multi⁃domain controller. In the optical controller, the protocol analysis module ana⁃ lyzes the underlying optical transmission equipment via the OpenFlow protocol extended for optical transmission devices. It collects the status of OF⁃ROADM in the optical network and abstracts the network topology information. Then the abstract⁃ ed topology information is sent to the network abstraction mod⁃ ule and stored in the optical database (ON ⁃ TED). The OTN manager manage the optical transmission equipment, such as lightpath setup and deletion, and resources allocation. The op⁃ tical network controller packages the network status and topolo⁃ gy information via the protocol encapsulation module and makes a notification to the multi⁃domain controller. The multi⁃domain controller integrates the network informa⁃ tion collected by the optical controller through the southbound Control Virtual Network Interface (CVNI) interface and moni⁃ tors the network status. With the northbound Restful ⁃ API, it parses the application requests sent by the application layer. It consists of nine function modules and one resource integration module. Resource integration is completed by the heteroge⁃ neous network database (Het⁃TED). The application database (App⁃TED) and ON⁃TED in network are set into the same data⁃ base, with the purpose of supplying the network resources in⁃ formation for the corresponding module in the network. The nine function modules are respectively described as follows. 1) Application monitor: It monitors the computing resources in the network and reports the information of computing re⁃ sources to service selection engine. 2) Service selection engine (SSE): According to the status of application resources and network resources requests, it se⁃ lects the most appropriate application resources to meet the vir⁃ tual requests. 3) Application resource manager: It manages the application resources in the network, and keeps real⁃time synchronous up⁃ date with the resources information in the application resourc⁃ es database (App⁃TED). 4) Request resolver: It parses the requests sent by the appli⁃ cation layer and forward them to the corresponding module for processing. 5) Virtual network manager: It manages the virtual network requests sent by the application layer according to the status of application resources and network resources, and selects the API: Application Program Interface APP⁃TED: application traffic engineering database AR: application resources CVNI: Control Virtual Network Interface ODL: OpenDaylight OF: OpenFlow OFP: OpenFlow protocol ON⁃TED: optical traffic engineering database OTN: optical transport network PCE: path computation element ROADM: Reconfigurable Optical Add/Drop Multiplexer SSE: service selection engine TED: traffic engineering database VN: virtual network ▲Figure 1. Hierarchical network architecture. Application 1, Application 2, Application 3, Application 4App layer Multi⁃domain control layer Optical control layer Physical layer Multi⁃domain controller Restful API Rq resolver VN manager AR manager SSE Application monitor Wrapper ODL core Policy generator Policy analyzing engine App⁃TED Domian 1⁃TED Domian 2⁃TED PCE Multi⁃domain TED Req/Rep PCE ON⁃TED Network abstraction OFP analysis ODL core Resolver OTN manager CVNI Optical controller CVNI Resolver Optical controller PCE OTN manager ON⁃TED Network abstraction OFP analysis ODL core OpenFlow agent OpenFlow agent OpenFlow agent OpenFlow agent OF⁃routerOF⁃router OpenFlow+Optical extension OpenFlow+Optical extension Data center Data center OF⁃ROADM Optical networks OF⁃ROADM Optical networks Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 51
  • 56. most appropriate network links to meet the virtual network re⁃ quests with the corresponding strategy. 6) Policy generator: According to the network requests from the application layer, it generates the corresponding strategy information of resource provisioning for the heterogeneous net⁃ work controller. 7) Policy analyzing engine: It parses the strategy generated by the strategy generator module, and sends it to the corre⁃ sponding module for network resource allocation. 8) PCE: As a core component of the multi⁃domain controller, PCE is used in response to the request of path computation. The calculation is based on the input path information, strate⁃ gy information and request information, etc. It may return two kinds of calculation results, the appropriate path information computed by multi⁃domain or the failure information. 9) Wrapper: It packages the resource allocation information with the CVNI protocol and sends it to the optical controller via the CVNI interface. 3 Workflow of Connection Provisioning Fig. 2 shows the workflow of providing an end⁃to⁃end con⁃ nection in the multi⁃domain optical networks. After a Transmis⁃ sion Control Protocol (TCP) connection is established, the multi ⁃ domain controller completes the handshake with the optical controller by using OpenFlow messages, then periodically sends packages to keep the connection alive. The multi ⁃ do⁃ main controller requests the abstract topology as well as the de⁃ tailed port information. After receiving a request for connec⁃ tion setup the optical controller completes the path computa⁃ tion and resource allocation in the local domain via the domain⁃ specified protocol. Once the process is finished, the optical controller sends a“success”reply to the multi⁃domain control⁃ ler. When the multi ⁃ domain controller collects all the“suc⁃ cess”messages from optical domains, a“success”notification will be sent to the application layer. At this point, a connection or lightpath is considered to be established successfully. 4 Protocol Extensions for SDON Based on OpenFlow 1.3 protocol, CVNI is an interface proto⁃ col between the multi⁃domain controller and optical layer con⁃ troller. Several OpenFlow messages in CVNI have been extend⁃ ed to satisfy the requirements of optical networks. The multi⁃ domain controller sends a GET_CONFIG_REQUEST message to the optical controller to get the location of network nodes and the optical controller replies a GET_CONFIG_REPLY message. The MULTIPART_REQUET messages is used by the multi⁃domain controller to obtain topology resources including ports and links information. The MULTIPART_REPLY mes⁃ sage carries topology information from the optical controller to the multi⁃domain controller. The multi⁃domain controller em⁃ ploys FLOW_MOD messages to complete connection setup and deletion. The match field and action field in an extended FLOW_MOD message respectively represent the input optical port and output optical port. Note that the multi⁃domain con⁃ troller sends a BARRIER_REQUEST message to the single⁃do⁃ main controller in order to verify whether the optical cross con⁃ nection is deployed successfully. The single⁃domain controller then sends a BARRIER_REPLY message to notify the multi⁃ domain controller that the connection is created or deleted suc⁃ cessfully. Due to space limitation, only FLOW_MOD message extension is illustrated in Fig. 3. 5 Experimental Platform for SDON As shown in Fig. 4, an all⁃optical network innovation (AO⁃ NI) experimental platform for SDON is distributed in three ge⁃ ography locations connected by optical fiber links. Two of them are located in Room 342 and Room 423 in the Sci⁃ ence&Research building of Beijing University of Posts and Telecommunications (BUPT ), and the third location is at 21Vi⁃ anet Company in Jiuxianqiao, Beijing. Two data centers are re⁃ spectively deployed in Room 342 and 21Vianet Company, and Room 423 serves as the access network for users, which com⁃ poses a typical network environment with the application of da⁃ ta center. The AONI platform supports three typical network scenarios, i.e., the inter⁃data center network, user access to da⁃ ta center network and intra ⁃ data center network. The AONI platform focuses on how to embody the advantages of optical switching network in these three scenarios. The platform sup⁃ ports both optical burst switching and optical circuit switching, and supports both flexible grid high⁃speed optical transmission and fixed grid transmission. Therefore, the AONI platform not only provides efficient transmission and switching in the future but also remains compatible with traditional networks. The high capacity optical burst switching (OBS) is mainly used for TCP: Transmission Control Protocol ▲Figure 2. Workflow of connection provisioning in multi⁃domain optical networks. Optical controller #1 Multi⁃domain controller Optical controller #N Application TCP three⁃way handshake TCP three⁃way handshake OpenFlow handshake & keep alive OpenFlow handshake & keep alive Topology resource reply Topology resource request Topology resource request Topology resource reply Connect creation request Connection request in local domain Connection request in local domain Connection reply in local domain Connection reply in local domain Connection creation reply Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS52
  • 57. the adaption to the high burstiness character⁃ istic of intra⁃data center services. The flexible grid high⁃speed optical transmission andopti⁃ cal circuit switching are mainly applied to the inter⁃data center to realize large⁃grained vari⁃ able bandwidth switching. The all⁃optical ac⁃ cess and convergence layer uses fixed grid transmission and switching to achieve flexible access of broadband services. Thus the archi⁃ tecture of AONI includes intra⁃data center all⁃ optical interconnection, all⁃optical access lay⁃ er, all⁃optical convergence layer and all⁃opti⁃ cal core layer. Such a platform can highly sim⁃ ulate real scenarios of all ⁃ optical switching wide area network (WAN) in the future. 6 Typical Applications of SDON The SDON is a promising solution to high intelligence of next generation optical net⁃ work and has broad application prospects. The typical applications include bandwidth on demand (BoD) provisioning, virtual ma⁃ chine (VM) online migration, spectrum defrag⁃ mentation, and virtual optical networks (VON) provisioning. The homepage of AONI applications is shown in Fig. 5. BoD applications and VM migration are im⁃ plemented based on the physical topology shown in Fig. 4. For lack of flexible⁃grid opti⁃ cal devices, a multi ⁃ domain logical topology (Fig. 6) is designed for VON provisioning and spectrum defragmentation. Both the physical topology and the logical topology are under control of the SDN controller. Each domain in⁃ ◀Figure 4. AONI: all optical network innovation environment. BUPT: Beijing University of Posts and Telecommunications BV⁃OXC: bandwidth⁃variable optical cross connect DC: data center OXC: optical cross connect S⁃NE: static⁃network element S⁃ROADM: static⁃reconfigurable optical add/drop multiplexer ▲Figure 3. FLOW_MOD message extension of CVNI protocol. China Unicom Cernet OpenFlow switch Room 423, BUPT Access layer S⁃NE S⁃ROADM BV⁃OXC Port A Port B Port C Cernet OpenFlow switchRoom 342, BUPT DC, BUPT Energy⁃efficient OXC Energy⁃efficient OXC Optical interconnection in DC Aggregation layer Core layer Chaoyang district, Beijing Jiuxianqiao DC 10GE optical interface 40G optical interface 1GE optical interface Flexible all⁃optical networks Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 53
  • 58. cludes eight standalone OpenFlow⁃Agents (OF⁃AGs). Running on high ⁃ performance Linux servers, each OF ⁃ AG is pro⁃ grammed based on Open⁃vSwitch. 6.1 BoD Applications BoD applications help users have a global understanding of underlying optical networks and accomplish a series of opera⁃ tions in optical networks, including connection setup, connec⁃ tion deletion, connection query, connection modification and so on. A lightpath connection is built and on ⁃ demand band⁃ width is allocated according to users requirements. Besides the instant operation, users are able to make an appointment to car⁃ ry out above operations by setting starting and ending time. Fig. 7 shows a connection named“service 1”is created from node 20.20.20.14 to node 20.20.20.21 with required band⁃ width. The detailed information about this connection is listed in the lower part of Fig. 7b, including routing, current status, creation delay and so on. 6.2 VM Online Migration VM migration plays an important role in data backup and load balance of data centers. A VM migration application en⁃ ables online migration of virtual machines among different da⁃ ta centers. With transmission advantages of optical networks, it just takes a short time to complete the migration process. In ad⁃ dition, the online migration pattern has no impact on users’ac⁃ cess to resources in the migrating virtual machine. In Fig. 8, a VM, 863VM, is migrated from server 10.108.50.40 to server 10.108.51.124 and the migration path is 20.20.20.14 ⁃ 20.20.20.15 ⁃ 20.20.20.12. Meanwhile, users can query re⁃ source utilization information of the selected servers, such as CPU and memory status. 6.3 VON application Optical network virtualization technologies support the dy⁃ namic provisioning of VONs in the same network infrastruc⁃ ture and achieve high⁃efficiency utilization of network resourc⁃ es. Because of its centralized control manner, software⁃defined networking (SDN) is regarded as a promising technology for re⁃ alizing VON provisioning. In the AONI testbed, network opera⁃ tors can provide virtual optical networks for different custom⁃ ers. The topology of VON can either be pre⁃configured by oper⁃ ▲Figure 5. Homepage of AONI applications. ▲Figure 6. Multi⁃domain logical topology. ▲Figure 7. Web view of BoD application: (a) before connection setup; (b) after connection setup. Domain 1 Domain 3 Domain 2 (a) (b) Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS54
  • 59. ators or be customized by users. In Fig. 9, a triangle VON to⁃ pology is successfully mapped to multi ⁃ domain networks. Meanwhile, 1+1 protection is available for services deployed in the VON. The green path in Fig. 9b stands for the working route of the service while the purple path represents the protec⁃ tion path. 6.4 Spectrum Defragmentation The frequent setup and release of lightpaths in a dynamic network scenario will fragment the optical spectrum into non⁃ aligned, isolated and small ⁃ sized spectrum segments. Spec⁃ trum fragments result in low spectrum utilization and high blocking probability since these fragments could hardly be oc⁃ cupied for new incoming requests. With the application of spectrum defragmentation, users can have a good knowledge of spectrum utilization and trigger spectrum defragmentation if necessary to optimize spectrum resources. In Fig. 10, there are 50 connections or lightpaths deployed in the multi⁃domain net⁃ work shown in Fig. 6. It is obvious that the spectrum utilization is effectively improved with the implementation of spectrum de⁃ fragmentation 7 Performance Evaluation of SDON Testbed Batch testing has been conducted to evaluate the perfor⁃ mance of this SDON testbed. Ten thousands lightpath requests are generated following Poisson distribution, and their source⁃ destination pairs per execution are randomly chosen. The hold⁃ ing time of lightpath requests follows exponential distribution. To verify the scalability of the SDON testbed, we compare the blocking probabilities of different network sizes. As shown in Fig. 11, the number of network nodes ranges from 200 to ▲Figure 8. Web view of VM migration: (a) before the migration; (b) after the migration. ▲Figure 9. Web view of VON provisioning: (a) before the process; (b) after the process. (a) (b) (a) (b) Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 55
  • 60. 1000. For each network size, the blocking probability increas⁃ es with traffic load. With the same traffic load, 1000⁃nodes net⁃ work has the lowest blocking probability since it has the high⁃ est network capacity. In addition, the relationship between the controller output bandwidth and the lightpath setup delay is studied. The con⁃ troller output bandwidth can be adjusted by the VMware Ether⁃ net bandwidth modulator. As shown in Fig. 12, the output bandwidth of controller is set to five different values, including 400 kbps, 500 kbps, 700 kbps, and 1 Mbps. The average delay of lightpath setup is calculated for each case. We can see that the output bandwidth of controller has great influence on light⁃ path setup delay. With the growth of output bandwidth, the av⁃ erage setup delay decreases significantly from 300 ms to 50 ms. 8 Conclusions With the advantage of programmable network elements, the SDON realizes service customization, adaptive modulation for⁃ mat, flexible bandwidth allocation and dynamic provisioning of virtual network resources with centralized control manner. This paper introduces SDON and its innovation environment— AONI in terms of network architecture, protocol extension solu⁃ tion, experiment platform, typical applications and perfor⁃ mance evaluation. The SDON represents the development di⁃ rection of optical networks and has broad application prospects in the future. ▲Figure 10. Web view of spectrum defragmentation: (a) before defragmentation; (b) after DFefragmentation. ▲Figure 11. Blocking probabilities of different network sizes. ▲Figure 12. Lightpath setup delay of different output bandwidth. References [1] Requirements for Generalized Multi⁃Protocol Label Switching (GMPLS) Routing for the Automatically Switched Optical Network (ASON), IETF RFC4258, Nov. 2005. (a) (b) 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 Blockingprobability 12001000800600200 Traffic load (Erlangs) 200 Nodes 400 Nodes 600 Nodes 800 Nodes 1000 Nodes 400 300 Setupdelay(ms) Traffic load (Erlangs) 250 200 150 100 50 0 1000800600400200 infinite 1 Mbps 700 kbps 500 kbps 400 kbps Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu Review October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS56 0
  • 61. [2] Requirements for Generalized MPLS (GMPLS) Signaling Usage and Extenstions for Automatically Switched Optical Network (ASON), IETF RFC4139, Jul. 2005. [3] Requirements for GMPLS⁃Based Multi⁃Region and Multi⁃Layer Networks (MRN/ MLN), IETF RFC5212, Jul. 2008. [4] Y. Ji, D. Ren, H. Li, X. Liu, and Z. Wang,“Analysis and experimentation of key technologies in service⁃oriented optical internet,”Science China Information Sci⁃ ences, vol. 54 no. 2, pp. 215-226, Feb. 2011. doi: 10.1007/s11432⁃010⁃4168⁃5. [5] A Path Computation Element (PCE)⁃Based Architecture, IETF RFC4655, Aug. 2006. [6] Path Computation Element (PCE) Communication Protocol Generic Require⁃ ments, IETF RFC4657, Sept. 2006. [7] Requirements for Path Computation Element (PCE) Discovery, IETF RFC4674, Oct. 2006. [8] Path Computation Element Communication Protocol (PCECP) Specific Require⁃ ments for Inter ⁃ Area MPLS and GMPLS Traffic Engineering, IETF RFC4927, Jun. 2007. [9] J. Zhang, H. Yang, Y. Zhao, et al.,“Experimental demonstration of elastic opti⁃ cal networks based on enhanced software defined networking (eSDN) for data center application,”Optics Express, vol. 21, no. 22, pp. 26990- 27002, Nov. 2013. doi:10.1364/OE.21.026990. Manuscript received: 2016⁃03⁃31 LI Yajie (yajieli@bupt.edu.cn) is a PhD candidate in State Key Laboratory of Infor⁃ mation Photonics and Optical Communication, Beijing University of Posts and Tele⁃ communications (BUPT), China. His research interest is software defined optical networks. ZHAO Yongli (yonglizhao@bupt.edu.cn) received his PhD degree from BUPT. He is an associate professor in State Key Laboratory of Information Photonics and Opti⁃ cal Communication, BUPT. His research interest is optical transport networks. ZHANG Jie (lgr24@bupt.edu.cn) received his PhD degree from BUPT. He is a pro⁃ fessor in State Key Laboratory of Information Photonics and Optical Communica⁃ tion, BUPT. His research interest is optical transport networks. WANG Dajiang (wang.dajiang@zte.com.cn) works in wireline product operation of BN product team, ZTE Corporation. His research interest is optical transport net⁃ works. WANG Jiayu (wang.jiayu1@zte.com.cn) received his master degree from BUPT. He is a SDON R&D representative from BN product team, ZTE Corporation. His re⁃ search interest is optical transport networks. BiographiesBiographies Review Software Defined Optical Networks and Its Innovation Environment LI Yajie, ZHAO Yongli, ZHANG Jie, WANG Dajiang, and WANG Jiayu October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 57 New Members of ZTE Communications Editorial Board Dr. CHEN Yan is a professor in the Department of Electrical Engineering and Computer Science at Northwestern University, USA. He is also an adjunct professor in the College of Computer Science at Zhejiang University, China. He got his PhD in Computer Science at University of California at Berkeley, USA in 2003. His research interests include network security, measurement and diagnosis for large scale networks and distributed systems. He won the Department of Energy (DoE) Early CAREER award in 2005, the Department of Defense (DoD) Young Investigator Award in 2007, and the Best Paper nomination in ACM SIGCOMM 2010. Based on the Google Scholar, his papers have been cited for over 9000 times and his h⁃index is 40. Dr. SONG Wenzhan is the Georgia Power Mickey A. Brown Professor of Engineering in the University of Georgia, USA. Dr. Song is a distinguished scientist and educator on cyber⁃physical systems informatics and security in energy, environment and health applications, where decentralized sensing, computing, communication and security play a critical role and need a transformative study. He has an outstanding record of leading large multidisciplinary research projects on those issues with multi⁃million grant support from NSF, NASA, USGS, and industry, and his research was featured in MIT Technology Review, Network World, Scientific America, New Scientist, National Geographic, etc. Dr. Song is a recipient of NSF CAREER Award (2010), Outstanding Research Contribution Award (2012) at GSU, Chancellor Research Excellence Award (2010) at WSU. He was also a recipient of 2004 National Outstanding Oversea Student Scholarship by China (only 40 in USA) during PhD study. Dr. Song also has a outstanding publication record and serves many premium conferences and journals as editor, chair or TPC member. He is also an inaugural member of OpenFog consortium involving industry and academic leaders.
  • 62. Depth EnhancementDepth Enhancement Methods for CentralizedMethods for Centralized Texture⁃Depth PackingTexture⁃Depth Packing FormatsFormats YANG Jar􀆼Ferr, WANG Hung􀆼Ming, and LIAO Wei􀆼Chen (Department of Electrical Engineering, Institute of Computer and Communication Engineering, National Cheng Kung University, 1 University Road, Taiwan 701, China) To deliver three ⁃ dimension (3D) videos through the current two⁃dimension (2D) broadcasting systems, the frame⁃compati⁃ ble packing formats properly including one texture frame and one depth map in various down ⁃ sampling ratios have been proposed to achieve the simplest and most effective solution. To enhance the compatible centralized texture⁃depth packing (CTDP) formats, in this paper, we further introduce two depth enhancement algorithms to further improve the quality of CT⁃ DP formats for delivering 3D video services. To compensate the loss of color YCbCr 444 to 420 conversion of colored ⁃ depth, two efficient depth reconstruction processes based on texture and depth consistency are proposed. Experimental re⁃ sults show that the proposed enhanced CTDP depacking pro⁃ cess outperforms the 2DDP format and the original CTDP de⁃ packing procedure in synthesizing virtual views. With the help of the proposed efficient depth reconstruction processes, more correct reconstructed depth maps and better synthesized quality can be achieved. Before the available 3D broadcasting systems, which adopt truly depth and texture dependent cod⁃ ing procedure, we believe that the proposed CTDP formats with depth enhancement could help to deliver 3D videos in the current 2D broadcasting systems simply and efficiently. 3D videos; frame⁃compatible; 2D⁃plus⁃depth; CTDP Abstract Keywords DOI: 10.3969/j. issn. 1673􀆼5188. 2016. 04. 008 http://www.cnki.net/kcms/detail/34.1294.TN.20160802.1732.004.html, published online August 2, 2016 O 1 Introduction ver past decades, more and more three⁃dimension⁃ al (3D) videos have been produced in the formats of stereo or multiple views with their correspond⁃ ing depth maps. People desire to have more truth⁃ ful and exciting experience through the true 3D visualizations. In order to fit the traditional two⁃dimensional (2D) television (TV) programs, we need to modify the 3D videos to accommo⁃ date the certain constraints. Frame⁃packing is one of possible solutions to introduce 3D services in the current cable and ter⁃ restrial 2D TV systems. There are several well⁃known formats for packing the stereo views into 2D frame such as side⁃by⁃ side (SbS), top ⁃ and ⁃ bottom (TaB), and checkerboard frame ⁃ compatible formats [1]- [4]. However, there exist two major problems, which slow down the development of the 3D TV ser⁃ vices, in the existing frame⁃packing methods. The frame⁃com⁃ patible packing 3D videos of the stereo views mean that two texture images are gathered in one frame, which may make se⁃ rious annoying effects on traditional 2D displays. Besides, ste⁃ reo packing formats cannot support multi⁃view naked⁃eye 3D displays unless the stereo videos are further processed by real⁃ time stereo matching methods [5], [6] and depth image⁃based rendering (DIBR) algorithms [7], [8]. To support multiview 3D displays, the 2D⁃plus⁃depth packing (2DDP) frame⁃compatible format, which arranges the texture in the left and the depth in the right, is suggested [9]. Once the color texture and depth ar⁃ ranged in the SbS fashion, the 2DDP format will bring even worse annoying visualization in 2D displays than the stereo packing formats. Recently, MPEG JCT⁃3V team proposed the latest coding standard for 3D video with depth [9]. However, it still needs some time to be deployed in current digital video broadcasting systems, which are with 2D and 3D capabilities. To deal with the above problems, a novel frame—compati⁃ ble centralized texture⁃depth packing (CTDP) formats for deliv⁃ ering 3D video services is proposed [10]. With AVS2 and HEVC video coders, the proposed CTDP formats [10] show bet⁃ ter objective and subjective visual quality in 2D and 3D dis⁃ plays than the 2DDP format. In the CTDP format, the sub⁃pixel is utilized to store the depth information, while the texture in⁃ formation is arranged in the center of the frame to raise the 2D⁃ compatible visual quality. However, the rearrangement will de⁃ grade the quality of the reconstructed depth map, especially when the video format with YCbCr space is 420 format with 4 Y components, one Cb component and one Cr component for each 4 color pixels. To further increase the visual quality, an efficient depth reconstruction process is also proposed in this paper. The frame structure of the CTDP method in cooperation with the current broadcasting system is shown in Fig. 1. With⁃ out any extra hardware, the 2D TV displays can also exhibit an acceptable 2D visual quality. For glasses or naked⁃eye 3D dis⁃ plays, we only need a simple CTDP depacking circuit followed by DIBR kernel to synthesize stereo or multiple views if the view⁃related sub⁃pixel formation of a naked⁃eye 3D display is given. The rest of the paper is organized as follows. The CTDP for⁃ mats are overviewed in Section 2. The proposed depth recon⁃ struction process is described in Section 3. Experimental re⁃ sults to demonstrate the effectiveness of the proposed system are shown in Section 4. Finally, we conclude this paper in Sec⁃ Research Paper October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS58
  • 63. tion 5. 2 Centralized Texture􀆼Depth Packing Formats To achieve system compatibility, the basic concept of the CTDP method [10] is similar to frame compatible concept to pack texture and depth information together while keeping the same resolution as 2D videos. To solve the 2D visualization is⁃ sue, we can arrange the texture in the center and the depth in two sides of the packed frame. 2.1 Colored􀆼Depth Frame The depth frame is only a gray image with Y components. To pack the depth frame, the colored⁃depth frame is suggested to represent it [10]. Thus, the colored⁃depth frame can be treated as the normal color texture frame, which can be directly encod⁃ ed by any 2D video encoders with three times efficiency. As shown in Fig. 2, three depth horizontal lines are treated as hor⁃ izontal R, G, and B subpixel lines in the RGB colored⁃depth frame. Since the nearby depth values are very close, the RGB colored⁃depth frame will exhibit nearly gray visual sensation. After color subpixels packing in the vertical direction, the ver⁃ tical resolution of RGB colored⁃depth frame becomes one third of the original resolution. In Fig. 2, for example, the nine depth lines have been packed into three RGB colored⁃ depth lines. For the most video coders, the coding and decoding processes are conducted in YCbCr color space. Therefore, we apply the RGB to YCbCr color space conversion as é ë êê ù û úú Y Cb Cr = é ë êê ù û úú 0.2568 0.5041 0.0979 -0.1482 -0.2910 0.4392 0.4392 -0.3678 -0.0714 é ë êê ù û úú R G B + é ë êê ù û úú 16 128 128 (1) to transfer it to the YCbCr colored⁃depth frame [11]. It is noted that the sub⁃pixels in RGB space are with full resolution of (4, 4, 4). If the YCbCr space is with (4, 4, 4) format, the color space transformation will not change the depth results with about +/- 0.5 error due to the round⁃off errors in color space conversions. However, for the most video coders, the sub⁃pix⁃ els in YCbCr space could be in (4, 2, 0) or (4, 2, 2) format, where Cb and Cr components will be further downsampled. Even without coding errors, the YCbCr colored ⁃ depth frame might have slightly translation errors. 2.2 Centralized Texture􀆼Depth Packing Without loss of generality for frame⁃compatible packing, we assume that the vertical CTDP packing formats are desired. Then, we need to reduce the vertical resolutions of texture and depth separately such that the total packed resolution will re⁃ mind the same, where the original horizontal resolution is H. If the reduction factors for texture and depth resolutions are a and b, we should choose reduction factors to satisfy α +(1/3)β = 1 to achieve the frame compatible requirement [10]. For example, the reduction factors (a = 3/4, b = 3/4) , (a = 5/6, b = 1/2), (a = 7/8, b = 3/8), (a = 11/12, b = 1/4), and (a = 15/16, b = 3/16) will satisfy the above frame compatible re⁃ quirement. Fig. 3 shows the flowchart of the computation of generating the texure ⁃ 5/6 CTDP format. First, we downscale the vertical resolution of texture and depth frames into five ⁃ sixths and one⁃second of the original resolution, respectively. By using the colored⁃depth concept, the resized depth frame with 1/2H can be further represented into RGB subpixels as suggested in Section 2.1 to reduce the vertical size to 1/6H. Then, we can split the depth frame evenly into two separated parts with the size of 1/12H. To make better coding efficiency and better 2D visualization, these two split colored ⁃ depth frames should be flipped vertically. The flipped depth frames will have better alignments to the texture frame and better visu⁃ alization for 2D displays with visual shadow sensation. Finally, we obtain the texture⁃5/6 CTDP frame by combining the first ▲Figure 1. The broadcasting architecture by using the proposed enhanced CTDP format. DIBR: depth image⁃based rendering ▲Figure 2. Rearrangement of the depth frame into RGB colored depth frame in vertical direction. Color and depth packing Video encoder Video decoder Multiview DIBR Color and depth depacking with depth enhancement 2DTV 3DTV No supporting hardware needed TV broadcasting network Stereo multiview 3D display formation Original depth frame (9 gray depth lines) 1 2 3 4 5 6 7 8 9 1R 1G 1B Colored depth frame (3 color depth lines) 2R 2G 2B 3R 3G 3B 3 depth pixels to 1 color depth pixel (RGB subpixels arrangement) Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 59
  • 64. flipped depth part (1/12H), the resized texture frame (5/6H), and the other flipped depth part (1/12H) from top to bottom se⁃ quentially. The ratio of downscaling can also be changed to generate the other CTDP formats [12]-[15]. For example, the reduction ratio of the texture frame could be 7/8 or 15/16. For texture⁃7/ 8 and texture⁃15/16 reduction ratios, the vertical resolutions of depth frames will be respectively downscaled to 3/8 and 3/16 to satisfy (2). Except the resizing factor, the packing proce⁃ dures for texture⁃7/8 and texture⁃15/16 are similar to that of texture⁃5/6. If we want to attain horizontal CTDP formats, all the resizing of texture and depth frame, the color⁃packed depth frame, slipping, and flipping procedures should be per⁃ formed in the horizontal direction. The packed frame can be obtained by com⁃ bining the first flipped depth part, the resized texture frame, and the other flipped depth part from left to right se⁃ quentially. The outlooks of the original texture, depth, and the CTDP frames with different ratios and different orien⁃ tations are shown in Fig. 4. It is noted that in the proposed CTDP format, the width/height of the flipped depth part will be always in the horizontal/vertical CTDP format, which helps avoid the compression artifact in texture and depth boundary. Please refer to [13] for more details of the arrangement. 2.3 Depacking CTDP Formats With respect to the packing proce⁃ dure in Fig. 3, the flow diagram for de⁃ packing the texture ⁃ 5/6 CTDP format is shown in Fig. 5. Once we receive the CTDP format, we should first split the packed frame into three parts: the top flipped depth part, the central texture, and the bot⁃ tom flipped depth part. For two flipped depth parts, we perform another vertical flipping and combined them into the whole texture⁃packed depth frame. The YCrCb colored⁃depth frame might need to upsample Cr and Cb components back to (4, 4, 4) format first. Then, we can convert it to (4, 4, 4) RGB colored ⁃ depth frame by é ë êê ù û úú R G B = é ë êê ù û úú 1.1644 -0.0001 1.5960 1.1644 -0.3917 -0.8130 1.1644 2.0173 -0.0001 æ è ç ç ö ø ÷ ÷ é ë êê ù û úú Y Cb Cr - é ë êê ù û úú 16 128 128 . (2) After the color space conversion, The RGB colored ⁃depth frame (1/6H) can be finally recovered to the re⁃ sized depth frame (1/2H). After 6/5 upscaling texture and 2/1 depth frames in the vertical direction, we finally depack the origi⁃ nal texture and depth frames. Of course, a possible DIBR method should be used to generate all the necessary views. As for the other texture reduction ratios such as 7/8 and 15/16, all the procedures will be the same except the resizing factors of depth will be 3/8 and 3/16, respectively. 3 Depth Enhancement Algorithms From the previous section, it is known that when the YCbCr space is (4, 2, 0) or (4, 2, 2) format, the YCbCr colored⁃depth frame will induce translation errors along the depth edges. To ▲Figure 3. The computation of the proposed frame compatible texture⁃5/6 CTDP format. (a) ▲Figure 4. Schematics of original (a) texture, (b) depth; (c) vertical texture⁃5/6 CTDP; (d) vertical texture⁃7/8 CTDP; (e) vertical texture⁃15/16 CTDP; (f) horizontal texture⁃5/6; (g) CTDP, horizontal texture⁃7/8 CTDP; (h) horizontal texture⁃15/16 CTDP; and (i) 2DDP frame compatible formats. 5/6H resized texture 1H 1H 1/2H 1/6H 1/12H 1/12H 1/12H 5/6H 1/12H 1H Downscaling in vertical direction Downscaling in vertical direction Pixel rearrangement to RGB subpixels channel YCbCr color format conversion Vertical splitting & flipping Combination Y′ Cb′ = Cr′ 0.2568 0.5041 0.0979 -0.1482 -0.2910 0.4392 0.4392 -0.3678 -0.0714 Y Cb Cr 16 128 128 + (b) (c) (d) (e) (f) (g) (h) (i) Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS60
  • 65. further reduce the depth edge errors, in this paper, we propose two efficient depth enhancement processes. The enhancement processes can be incorporated with the original depacking pro⁃ cess as shown in Fig. 6. The enhancement processes include YCbCr calibration, texture⁃similarity⁃based depth up⁃sampling and pattern⁃based down⁃sampling. Details of the enhancement algorithms are addressed in the following subsections. 3.1 YCbCr Calibration When the YCbCr color space is (4, 4, 4), the color space transformation between RGB color space and YCbCr color space will only contain round⁃off errors in color space conver⁃ sions. However, for the most video coders, the sub⁃ pixels in the YCbCr color space might be (4, 2, 0) or (4, 2, 2) formats, where Cb and Cr components will be further down⁃sampled in order to save the bandwidth in broadcasting systems. At the depack⁃ ing side, we need to calibrate the translation errors between YCbCr (4, 4, 4) and YCbCr (4, 2, 0) and (4, 2, 2). For simplicity, we will illustrate our pro⁃ posed system in YCbCr (4, 2, 0), however, the simi⁃ lar manner can still be applied for YCbCr (4, 2, 2). Before we start to calibrate the YCbCr data, we first define some anchor pixels, which are shown in Fig. 7. The anchor pixels denote the pixels which have the correct Cb and Cr subpixel values. The diagram of missing components in YCbCr (4, 2, 0) for all surrounding pixels is shown in Fig. 8. Each color means a set which Cb and Cr subpixel components are down ⁃ sampled. The black area means the missing Cb and Cr subpixels and they can be given by: Cbcal(a,b)= argCbC min |YC - Y(a,b)| , (3) and Crcal(a,b)= argCrC min |YC - Y(a,b)| , (4) where YC is a vector of the neighbor anchor pixels of the pix⁃ els Y(a, b). 3.2 Texture􀆼Similarity􀆼Based Depth Up􀆼Sampling In order to preserve the continuity of the edge, the direction⁃ al vectors are utilized to calculate the edge direction in the low⁃ resolution (LR) depth and the corresponding high resolution (HR) texture image. The directional vec⁃ tors of LR depth image and HR texture image can be formed as:   VdL =∑Ω exp(- DE(xL,yL)- DΩ σV ) × uΩ , (5) and  Vc =∑Ω exp(- Y(x,y)- YΩ σV ) × uΩ , (6) where   VdL and  Vc denote the directional vectors of the pixels in LR depth image and HR texture image, respectively, σV represents the standard deviation ▲Figure 5. The computation of the proposed texture⁃5/6 CTDP depacking procedure. ▲Figure 6. Flowchart of the depth⁃enhanced CTDP de⁃packing system. ▲Figure 7. Anchor pixels in YCbCr (4, 2, 0). 1H 1/12H 1/12H View and depth splitting Vertical flipping & combination Inverse YCbCr color format conversion Get depth pixel from RGB subpixels channel Upscaling in vertical direction R G = B 1.1644 -0.0001 1.5960 1.1644 -0.3917 -0.8130 1.1644 2.0173 -0.0001 Y Cb Cr 16 128 128 - 1/12H 5/6H 1/12H 1H 1H 1/6H 1/2H Upscaling in vertical direction View and depth splitting Horizontal flipping & combination View and depth splitting Horizontal flipping & combination YCbCr calibration Inverse YCbCr color format conversion Get depth pixel from RGB subpixels channel Upscaling in horizontal direction (bicubic) Depth refinement Upscaling in horizontal direction H W⁃WRD*2 H W⁃WRD*2WRD WRD H W W H Y Cb Cr Y YY Y :Yanc Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 61
  • 66. of the directional vector function, Ω denotes the 8 neighbor pixels of the target pixel (Fig. 9), DE represents the combined depth, which is obtained from previous step, Y is the bright⁃ ness of the texture image, and uΩ is the unit vector corre⁃ sponding to the neighbor pixels Ω in 8 directions. Before up⁃sampling the depth image, the directional vectors are first transformed from Cartesian coordinate system to Spherical coordinate system. The transform function is given by: r = x∂ 2 + y∂ 2 , (7) and θ = arctan( y∂ x∂ ) , (8) where x∂ and y∂ denote the coordinate of reconstructed depth at high resolution. For example, at vertical texture⁃11/12 CT⁃ DP, x∂ = 4x and y∂ = y . However, the resolution of directional vectors in depth image is smaller than the resolution of direc⁃ tional vectors in texture image. The bilinear interpolation [16] is utilized to scale up the depth directional vector to the resolu⁃ tion of the texture image. After that, The interpolated depth im⁃ age is formed as: where Tup denotes the normalized factor, p is the target pixel which needs to be scale up, q is the neighbor pixels of the tar⁃ get pixel, and Vd(θ) is the value of θ in the scaled VdL(θ) . ψ denotes the Gaussian weight function and can be giv⁃ en as: ψ(n)= exp(- n2 σψ ) . (10) The basic concept of the depth interpolation is to compare the directional vectors of the depth image and the texture image. The weighted summation of the LR depth is utilized to interpolate the HR depth if the directional vectors of the depth image and the texture image are similar. Otherwise, the pixels in HR depth are regarded as holes, which are filled in the step of hole⁃filling. The function of hole⁃filling is given as: Dhole - filling(x,y)= ì í î argDup ξ(min(ΔPc(θ))), if (x,y)∍ holes Dup(x,y), else , (11) where ΔPc(θ) denotes the difference of the degree between Pcθ and 8 neighbor pixels. ξ represents the selection fun⁃ ction of the hole⁃filling and it can be formed as: ξ(m)={Y(m), if||Y - Y(m)|| < THY ξ(m)+ 1, else , (12) where Y denotes the brightness of the target pixel, Y(m) de⁃ notes the brightness of the neighbor pixels in m direction, THY is the threshold to control the selection range, and ξ(m)+ 1 represents the next pixel in m direction. 3.3 Pattern􀆼Based Down􀆼Sampling In order to contain texture image and depth image in one sin⁃ gle frame, both depth image and texture image need to be down ⁃sampled. For the depth image, the bilinear and bi⁃cubic con⁃ volution methods are utilized to down⁃sample the depth image. However, the weighted summation strategy in bilinear and bi⁃ cubic convolution leads to the blur of the down⁃sampled data. Hence, we propose two sampling patterns to down⁃sample the depth image without fusing the data. There are the direct line pattern and slant line pattern. 1) Direct line pattern The sampling strategy of direct line pattern is to grab pixels in the straight line direction. According to the characteristic of the CTDP format, the reduction of the resolution is only in ei⁃ ther horizontal or vertical direction. The function of direct line pattern is given as: Ddown(x,y)= Dorigin(∂hor × x -[∂hor /2],∂ver × y -[∂ver /2]) , (13) where ∂hor and ∂ver are the factors of down⁃sampling ratios in horizontal direction and vertical direction, respectively. For CTDP format usage, either ∂hor or ∂ver is equal to 1, while the other one denotes the down⁃sampling ratio in packing proce⁃ dure. [x] is the floor function, which means the largest integer not greater than x. The direct line pattern in horizontal direc⁃ tion with 2, 4, 8 down⁃sampling ratio is shown in Fig. 10. 2) Slant line pattern ▲Figure 8. Missing Cb and Cr subpixels in YCbCr (4, 2, 0). ◀Figure 9. The diagram of the neighbor pixels, Ω . (9)Dup = ì í î ï ï 1 Tup ∑DE(q)× ψ(DE(p)- DE(q)), if||Vd(θ)- Vc(θ)|| < π 8 or Vc(r) < 1 hole, else , YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY YCbY Cr YY Ω 135° Ω 90° Ω 45° Ω 0° Ω 180° Ω 225° Ω 275° Ω 315° Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS62
  • 67. The sampling strategy of slant line is to grab pixels in 45 de⁃ gree direction. The function of direct line pattern is given as: Ddown(x,y)= Dorigin(∂hor × x -(∂hor - y),y) , (14) or Ddown(x,y)= Dorigin(x,∂ver × y -(∂ver - x)) . (15) Equ. (14) is utilized to down⁃sample the depth image in hori⁃ zontal direction, while the down⁃sampling of the vertical direc⁃ tion follows (15). The slant line sampling pattern is suitable for down⁃sampling the depth image both in vertical and horizontal direction, which is shown in Fig. 11 with 2, 4, 8 down⁃sam⁃ pling ratios. With the down⁃sampling by the direct line pattern, the up⁃ sampling function in de⁃packing procedure needs to be modi⁃ fied as: Because of the pattern⁃based sampling strategy, the pixels of the up⁃sampled depth are directly copied from the LR depth if there are located at position of the direct line pattern. 4 Experimental Results 4.1 Performance Evaluation of CTDP Format with Respect to 2DDP Format In order to verify the coding performances of the proposed CTDP formats with respect to the 2DDP format, we conducted a set of experiments to evaluate performances of packing meth⁃ ods in cooperation with a specific video coder (AVS2) in terms of the peak signal⁃to⁃noise ratio (PSNR), bitrate qualities of the depacked texture and depacked depth frames, and their synthe⁃ sized virtual views. In the experimental simulations, we use five MPEG 3D video sequences, which are Poznan Hall, Poznan Street, Kendo, Balloons, and Newspaper sequences as shown in Figs. 12a-12e, respectively. The AVS2 coding conditions are followed by the instruction suggested by the AVS workgroup while the QPs are set to 27, 32, 38, and 45 for Intra frames [17]. Under All Intra (ai), Low Delay P (ldp), Random Access (ra) test conditions, Tables 1 and 2 show the average BDPSNR and BDBR [18] performance for different kinds of CTDP formats with respect to the 2DDP format achieved by AVS2. For calculating the PSNR of the 2DDP format, we first separate the texture and depth frames from the 2DDP frame and upsample them to the original image size W × H. By using the recovered texture and depth frames from 2DDP frame and the original uncompressed texture and depth frames, the PSNR can therefore be calculated. Similarly, the PSNR of CTDP format is calculated by using the texture and depth frames recovered from CTDP frame and the original uncompressed texture and depth frames. From Tables 1 and 2, we can see that the proposed texture⁃5/6, 7/8, and 15/16 CTDP formats have much better PSNR and bitrate saving in texture when comparing with the 2DDP format, which means our CT⁃ DP format can achieve better visual quality in 2D displays when only texture frames are viewed. In addition, the depth quality for CTDP formats will become worse while the resizing factors getting bigger. Besides the comparisons of original tex⁃ ture and depth achieved by different packing formats, we also compare the quality of synthesized virtual view with respect to the 2DDP format. It is noted that the reference synthesized vir⁃ ▲Figure 10. The direct line pattern in horizontal direction of (a) down⁃sampling factor 2; (b) down⁃sampling factor 4; and (c) down⁃sampling factor 8. 10 11 (16)Dup = ì í î ïï ïï DE(p), if p ∈ sampled data 1 T ∑DE(q)× ψ(DE(p)- DE(q)), else if ||Vd(θ)- Vc(θ)|| < π 8 or Vc(r) < 1 hole, else . ▲Figure 11. The slant line pattern of (a) down⁃sampling factor 2; (b) down⁃sampling factor 4; (c) down⁃sampling factor 8. 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (a) (c) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (b) Missing pixel Sampled pixel 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (a) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (b) 10 11 127 8 94 5 61 2 3 1 2 3 4 5 6 7 8 9 10 11 12 (c) Missing pixel Sampled pixel Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 63
  • 68. tual view for calculating the PSNR is also obtained by the origi⁃ nal uncompressed texture and depth frames. The DIBR setting for virtual view synthesis is shown in Table 3. As to the quali⁃ ty of the synthesized virtual view, the texture⁃5/6 and 7/8 CT⁃ DP formats after the DIBR process show better BDPSNR and BDBR performances than 2DDP format. It is noted that all syn⁃ thesized views do not perform any depth enhancement and depth preprocessing, and the hole filling used in the DIBR pro⁃ cess is the simple background extension. In summary, the texture qualities BDPSNR and BDBR in Ta⁃ bles 2 and 3 can be treated as the objective quality indices in 2D displays, while the virtual view qualities can be the objec⁃ tive quality indices in 3D displays. The results show that the proposed texture⁃5/6 and 7/8 CTDP format will be the better choices for the broadcasters. The texture⁃3/4 CTDP format has better 3D performance while texture⁃7/8 CTDP format achieves better 2D performance. 4.2 Performance Evaluation of Depth Enhancement for CTDP Format To verify the proposed depth enhancement mechanism, we first show the reconstructed depth from original and depth⁃en⁃ hanced CTDP formats. The RD curves for different ratios of CTDP formats are shown in Fig. 13. It can be seen that the pro⁃ posed refined CTDP format can always achieve better perfor⁃ mance. The gains between the depth⁃enhanced CTDP and the original CTDP formats are increased while the ratio of texture is increased. For the subjective evaluation, the partial portions of the re⁃ constructed depth for Shark sequence are shown in Fig. 14. It can be seen that the depth can be reconstructed well especial⁃ ly for the edge region by using the depth enhancements. In the following, we will compare the synthesis results. The partial portions of the generated views are shown in Fig. 15. From the results, the proposed CTDP format can successfully preserve the edges well of the synthesis views without the jag⁃ gy noise. 4.3 Comparison with Different Depth Interpolation Methods The comparison results of different depth interpolation meth⁃ ods are shown in Table 4 for Shark sequence at all⁃intra (ai) coding condition with QP=32. The symbols of Bi and BC de⁃ note the bilinear and bi⁃cubic convolution interpolation meth⁃ ods, respectively. The methods of JBU [19] and FEU [20] are the texture⁃similarity based depth interpolation methods. The proposed depth up ⁃ sampling method has better PSNR and SSIM results for reconstructed depth images in vertical⁃11/12 CTDP and vertical⁃23/24 CTDP formats. For the vertical⁃5/6 CTDP format, the proposed depth up⁃sampling method can al⁃ so provide better reconstructed depth images. The comparison results of partial reconstructed depth with different depth interpolation methods are shown in Fig. 16. The reconstructed depth images of bilinear and bi⁃cubic convo⁃ lution interpolation methods have serious jaggy noise among the edges. It can be seen that the proposed depth up⁃sampling method can outperform other methods with better edges. 5 Conclusions In this paper, we proposed depth enhancement processes for ▲Figure 12. Five texture and depth frame compatible packing formats. ▼Table 1. Averaged BDPSNR performances BDPSNR ai ra ldp avg Recovered texture performance 5/6 2.2693 2.5868 2.2079 2.3546 7/8 2.2587 2.71322 2.42228 2.464733 15/16 2.34168 2.86354 2.54206 2.582427 Synthesized virtual view performance 5/6 0.6150 0.7371 0.3920 0.5813 7/8 0.3664 0.7489 0.6005 0.5720 15/16 ⁃0.7732 ⁃0.2754 ⁃0.3928 ⁃0.4805 ▼Table 2. Averaged BDBR performances BDBR ai ra ldp avg Recovered texture performance 5/6 ⁃55.3282 ⁃61.0706 ⁃48.7147 ⁃55.0378 7/8 ⁃46.5149 ⁃59.5852 ⁃54.3501 ⁃53.4834 15/16 ⁃47.9069 ⁃61.6926 ⁃56.2303 ⁃55.2766 Synthesized virtual view performance 5/6 ⁃22.9490 ⁃24.6686 ⁃9.9111 ⁃19.1762 7/8 ⁃7.92186 ⁃24.6019 ⁃19.6368 ⁃17.3869 15/16 56.0625 30.0742 36.5666 40.9011 ▼Table 3. DIBR settings for virtual view synthesis Sequence Poznan hall Poznan street Kendo Balloons Newspaper Resolution 1920*1088 1920*1088 1024*768 1024*768 1024*768 Frames 200 250 300 300 300 Coded view 6 4 3 3 4 Synthesized view 5 3 4 5 6 (a) Poznan hall (b) Poznan street (c) Keno (d) Balloons (e) Newspaper Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS64
  • 69. CTDP formats [10]. The CTDP formats can be comfortably and directly viewed in 2DTV displays without the need of any extra computation. However, the CTDP formats slightly suffer from the depth discontinuities for high texture ratios. Comparing to ▲Figure 13. RD curves of reconstructed depth from the original CTDP depacking process and the proposed depth⁃enhanced CTDP depacking process for (a) texture⁃5/6; (b) texture⁃11/12; (c) texture⁃23/24 formats. CTDP: compatible centralized texture⁃depth packing PSNR: peak signal⁃to⁃noise ratio ▲Figure 14. Partial portions of reconstructed depth in S10 Shark with the original CTDP depacked: (a) texture⁃5/6; (b) texture⁃11/12; (c) tex⁃ ture⁃23/24 formats and the proposed depth enhanced CTDP depacked; (d) texture⁃5/6; (e) texture⁃11/12; (f) texture⁃23/24 formats. ▲Figure 15. Partial synthesis views of S02 Poznan Street with original CTDP depacking: (a) texture⁃5/6; (b) texture⁃11/12; (c) texture⁃23/24 and the proposed depth⁃enhanced CTDP depacking; (d) texture⁃5/6; (e) texture⁃11/12; (f) texture⁃23/24. ▼Table 4. The PSNR and SSIM comparison of different depth interpolation methods in S10 Shark at All Intra (ai) QP=32 PSNR (dB) Vertical⁃5/6 CTDP Vertical⁃11/12 CTDP Vertical⁃23/24 CTDP SSIM Vertical⁃5/6 CTDP Vertical⁃11/12 CTDP Vertical⁃23/24 CTDP Bi 33.7164 31.7651 29.8525 Bi 0.9361 0.9147 0.8914 BC 33.5138 31.6581 29.7836 BC 0.9334 0.9122 0.8879 JBU 33.6252 32.0857 30.3490 JBU 0.9361 0.9192 0.8959 FEU 33.7135 31.7850 29.8953 FEU 0.9368 0.9158 0.8925 Proposed 33.5239 32.4422 30.5411 Proposed 0.9397 0.9270 0.9068 (a) (b) (c) (d) (e) (f) (a) (b) (c) (d) (e) (f) 40003500300025002000150010005000 41.0 40.5 40.0 39.5 39.0 38.5 38.0 37.5 37.0 36.5 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) (a) 350030002500200015005000 39.5 39.0 38.5 38.0 37.5 37.0 36.5 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) 36.0 35.5 35.0 (b) 4000 38.0 PSNR(dB) Bit rate (Kbps) CTDP (bilinear) CTDP (bicubic) Depth⁃enhanced CTDP (direct line pattern) Depth⁃enhanced CTDP (slantline pattern) (c) 450040003500300025002000150010005000 37.5 37.0 36.5 36.0 35.5 35.0 34.5 34.0 33.5 33.0 Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen Research Paper October 2016 Vol.14 No. 4 ZTE COMMUNICATIONSZTE COMMUNICATIONS 65 1000
  • 70. the 2DDP format, the CTDP formats with the same video cod⁃ ing systems, such as AVS2 (RD 6.0) and HEVC [10], show bet⁃ ter coding performances in texture and depth frames and syn⁃ thesized virtual views. To further increase the visual quality, in this paper, the depth enhancement methods, including YCbCr calibration and texture ⁃ similarity ⁃ based depth up ⁃ sampling, are proposed. Experimental results reveal that the proposed depth enhancement can efficiently help to increase the depack⁃ ing performances of the CTDP formats to achieve better recon⁃ structed depth images and better synthesis views as well. With the aforementioned simulation results, we believe that the pro⁃ posed depth enhanced CTDP depacking methods will be a greatly⁃advanced system for current 2D video coding systems, which can provide 3D video services effectively and simply. ▲Figure 16. The comparison of partial reconstructed depth with differ⁃ ent depth interpolation methods for vertical ⁃ 11/12 CTDP format: (a) ground truth; (b) bilinear; (c) bi ⁃ cubic convolution; (d) JBU [19]; (e) FEU [20]; (f) proposed. References [1] J.⁃F. Yang, H.⁃M. Wang, K.⁃I. Liao, L. Yu, and J.⁃R. Ohm,“Centralized texture⁃ depthpacking formats for effective 3D video transmission over current video broadcasting systems,”IEEE Transactions on Circuits and Systems for Video Technology, submitted for publication. [2] Dolby Laboratories, Inc. (2015). Dolby Open Specification for Frame⁃Compatible 3D Systems [Online]. Available: http://www.dolby.com [3] ITU. (2015). Advanced Video Coding for Generic Audio—Visual Services [Online]. Available: http://www.itu.int [4] G. Sullivan, T. Wiegand, D. Marpe, and A. Luthra,“Text of ISO/IEC 14496⁃10 advanced video coding (third edition),”ISO/IEC JTC 1/SC 29/WG11, Redmond, USA, Doc. N6540, Jul. 2004. [5] G. J. Sullivan, A. M. Tourapis, T. Yamakage, and C. S. Lim,“ISO/IEC 14496⁃10: 200X/FPDAM 1,”ISO/IEC JTC 1/SC 29/WG11, Apr. 2009. [6] T. Kanade and M. Okutomi,“A stereo matching algorithm with an adaptive win⁃ dow: theory and experiment,”IEEE Transactions on Pattern Analysis and Match⁃ ing Intelligence, vol. 16, no. 9, pp.920-932, Sept. 1994. doi:10.1109/34.310690. [7] K. Zhang, J. Lu, and G. Lafruit,“Cross⁃based local stereo matching using orthog⁃ onal integral images,”IEEE Transactions on Circuits and Systems for Video Tech⁃ nology, vol. 19, no. 7, pp.1073- 1079, Jul. 2009. doi: 10.1109/TCS⁃ VT.2009.2020478. [8] S.⁃C. Chan, H.⁃Y. Shum, and K.⁃T. Ng,“Image⁃based rendering and synthesis,” IEEE Signal Processing Magazine, vol. 24, no. 6, pp. 22-33, Nov. 2007. doi: 10.1109/MSP.2007.905702. [9] T.⁃C. Yang, P.⁃C. Kuo, B.⁃D. Liu, and J.⁃F. Yang,“Depth image⁃based rendering with edge⁃oriented hole filling for multiview synthesis,”in Proc. International Conference on Communications, Circuits and Systems, Chengdu, China, Nov. 2013, vol. 1, pp. 50-53. doi: 10.1109/ICCCAS.2013.6765184. [10] Philips 3D Solutions,“3D interface specifications, white paper,”Eindhoven, The Netherlands, Dec. 2006. [11] Studio Encoding Parameters of Digital Television for Standard 4:3 and Wide⁃ Screen 16:9 Aspect Ratios, ITU⁃R BT.601⁃5, 1995. [12] J.⁃F. Yang, K.⁃Y. Liao, H.⁃M. Wang, and Y.⁃H. Hu,“Centralized texture⁃depth packing (CTDP) SEI message syntax,”Joint Collaborative Team on 3D Video Coding Extensions of ITU⁃T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11, Strasbourg, France, Doc. no. JCT3V⁃J0108, Oct. 2014. [13] J.⁃F. Yang, K.⁃Y. Liao, H.⁃M. Wang, and C.⁃Y. Chen,“Centralized texture⁃ depth packing (CTDP) SEI message,”Joint Collaborative Team on 3D Video Coding Extensions of ITU⁃T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11, Geneva, Switzerland, Doc. no. JCT3V⁃K0027, Feb. 2015. [14] J.⁃F. Yang, H.⁃M Wang, Y.⁃A. Chiang, and K. Y. Liao,“2D frame compatible centralized color depthpacking format (translated from Chinese),”AVS 47th Meeting, Beijing, China, AVS⁃M3225, Dec. 2013. [15] J.⁃F. Yang, H.⁃M. Wang, K.⁃Y. Liao, and Y.⁃A. Chiang,“AVS2 syntax message for 2D frame compatible centralized color depth packing formats (translated from Chinese),”AVS 50th Meeting, Nanjing, China, AVS⁃M3472, Oct. 2014. [16] H. C. Andrews and C. L. Patterson,“Digital interpolation of discrete images,” IEEE Transaction on Computers, vol. 25, no. 2, 1976. [17] X.⁃Z. Zheng,“AVS2⁃P2 common test conditions (translated from Chinese),” AVS 46th Meeting, Shenyang, China, AVS⁃N2001, Sep. 2013. [18] G. Bjontegaard,“Calculation of average PSNR differences between RD ⁃ curves,”Austin, USA, Doc. VCEG⁃M33 ITU⁃T Q6/16, Apr. 2001. [19] J. Kopf, M. F. Cohen, D. Lischinski, and M. Uyttendaele,“Joint bilateral upsampling,”ACM Transaction on Graphics, vol. 26, no. 3, Article 96, Jul. 2007. doi:10.1145/1275808.1276497. [20] S.⁃Y. Kim and Y.⁃S. Ho,“Fast edge⁃preserving depth image upsampler,”Jour⁃ nal of Consumer Electronics, vol. 58, no. 3, pp. 971- 977, Aug. 2012. doi: 10.1109/TCE.2012.6311344. Manuscript received: 2015⁃11⁃12 YANG Jar⁃Ferr (jefyang@mail.ncku.edu.tw) received his PhD degree from the Uni⁃ versity of Minnesota, USA in 1988. He joined the National Cheng Kung University (NCKU) started from an associate professor in 1988 and became a full professor and distinguished professor in 1995 and 2007. He was the chairperson of Graduate Insti⁃ tute of Computer and Communication Engineering during 2004-2008 and the direc⁃ tor of the Electrical and Information Technology Center 2006-2008 in NCKU. He was the associate vice president for Research and Development of the NCKU. Cur⁃ rently, he is a distinguished professor and the director of Technologies of Ubiqui⁃ tous Computing and Humanity (TOUCH) Center supported by National Science Council (NSC), Taiwan, China. Furthermore, he is the director of Tomorrow Ubiqui⁃ tous Cloud and Hypermedia (TOUCH) Service Center. During 2004-2005, he was selected as a speaker in the Distinguished Lecturer Program by the IEEE Circuits and Systems Society. He was the secretary, and the chair of IEEE Multimedia Sys⁃ tems and Applications Technical Committee and an associate editor of IEEE Trans⁃ action on Circuits and Systems for Video Technology. In 2008, he received the NSC Excellent Research Award. In 2010, he received the Outstanding Electrical Engi⁃ neering Professor Award of the Chinese Institute of Electrical Engineering, Taiwan, China. He was the chairman of IEEE Tainan Section during 2009-2011. Currently, he is an associate editor of EURASIP Journal of Advances in Signal Processing and an editorial board member of IET Signal Processing. He has published 104 journal and 167 conference papers. He is a fellow of IEEE. WANG Hung ⁃ Ming (ming@video5.ee.ncku.edu.tw) received the BS and PhD de⁃ grees in electrical engineering from National Cheng Kung University (NCKU), Tai⁃ wan, China in 2003 and 2009, respectively. He is currently a senior engineer of No⁃ vatek Microelectronics Corp., Taiwan, China. His major research interests include 2D/3D image processing, video coding and multimedia communication. LIAO Wei⁃Chen (a800812momo@gmail.com) received the BS and MS degrees in electrical engineering from National Cheng Kung University (NCKU), Taiwan, Chi⁃ na in 2013 and 2015, respectively. His major research interests include image pro⁃ cessing, video coding and multimedia communication. BiographiesBiographies (a) (b) (c) (d) (e) (f) Research Paper Depth Enhancement Methods for Centralized Texture⁃Depth Packing Formats YANG Jar⁃Ferr, WANG Hung⁃Ming, and LIAO Wei⁃Chen October 2016 Vol.14 No. 4ZTE COMMUNICATIONSZTE COMMUNICATIONS66
  • 71. • Remit of Journal ZTE Communications publishes original theoretical papers, research findings, and surveys on a broad range of communica⁃ tions topics, including communications and information system design, optical fiber and electro-optical engineering, micro⁃ wave technology, radio wave propagation, antenna engineering, electromagnetics, signal and image processing, and power engineering. The journal is designed to be an integrated forum for university academics and industry researchers from around the world. • Manuscript Preparation Manuscripts must be typed in English and submitted electronically in MS Word (or compatible) format. The word length is ap⁃ proximately 3000 to 8000, and no more than 8 figures or tables should be included. Authors are requested to submit mathe⁃ matical material and graphics in an editable format. • Abstract and Keywords Each manuscript must include an abstract of approximately 150 words written as a single paragraph. The abstract should not include mathematics or references and should not be repeated verbatim in the introduction. The abstract should be a self- contained overview of the aims, methods, experimental results, and significance of research outlined in the paper. Five care⁃ fully chosen keywords must be provided with the abstract. • References Manuscripts must be referenced at a level that conforms to international academic standards. All references must be num⁃ bered sequentially in-text and listed in corresponding order at the end of the paper. References that are not cited in-text should not be included in the reference list. References must be complete and formatted according to ZTE Communications Editorial Style. A minimum of 10 references should be provided. Footnotes should be avoided or kept to a minimum. • Copyright and Declaration Authors are responsible for obtaining permission to reproduce any material for which they do not hold copyright. Permission to reproduce any part of this publication for commercial use must be obtained in advance from the editorial office of ZTE Com⁃ munications. Authors agree that a) the manuscript is a product of research conducted by themselves and the stated co-au⁃ thors, b) the manuscript has not been published elsewhere in its submitted form, c) the manuscript is not currently being con⁃ sidered for publication elsewhere. If the paper is an adaptation of a speech or presentation, acknowledgement of this is re⁃ quired within the paper. The number of co-authors should not exceed five. • Content and Structure ZTE Communications seeks to publish original content that may build on existing literature in any field of communications. Au⁃ thors should not dedicate a disproportionate amount of a paper to fundamental background, historical overviews, or chronolo⁃ gies that may be sufficiently dealt with by references. Authors are also requested to avoid the overuse of bullet points when structuring papers. The conclusion should include a commentary on the significance/future implications of the research as well as an overview of the material presented. • Peer Review and Editing All manuscripts will be subject to a two-stage anonymous peer review as well as copyediting, and formatting. Authors may be asked to revise parts of a manuscript prior to publication. • Biographical Information All authors are requested to provide a brief biography (approx. 100 words) that includes email address, educational back⁃ ground, career experience, research interests, awards, and publications. • Acknowledgements and Funding A manuscript based on funded research must clearly state the program name, funding body, and grant number. Individuals who contributed to the manuscript should be acknowledged in a brief statement. • Address for Submission magazine@zte.com.cn 12F Kaixuan Building, 329 Jinzhai Rd, Hefei 230061, P. R. China ZTE Communications Guidelines for Authors
  • 72. ZTECOMMUNICATIONS •Cambridge Scientific Abstracts (CSA) •China Science and Technology Journal Database •Chinese Journal Fulltext Databases •Inspec •Ulrich’s Periodicals Directory •Wanfang Data—Digital Periodicals ZTE Communications has been indexed in the following databases: