CHANNEL ESTIMATION FOR THE ISDBT B FBMC SYSTEM USING NEURAL NETWORKS: A PROPOSAL OF APPLICATION OF BACK-PROPAGATION TRAINING ALGORITHM

Dhinaharan Nagamalai et al. (Eds) : ITCCMA, CSITY, AIFZ, NWCOM, SIGPRO - 2017
pp. 59– 69, 2017. © CS & IT-CSCP 2017 DOI : 10.5121/csit.2017.70705
CHANNEL ESTIMATION FOR THE ISDB-
TB FBMC SYSTEM USING NEURAL
NETWORKS: A PROPOSAL OF
APPLICATION OF BACK-PROPAGATION
TRAINING ALGORITHM
Jefferson Jesus Hengles Almeida, P. B. Lopes,
Cristiano Akamine and Nizam Omar
Postgraduate Program in Electrical Engineering and Computing
Mackenzie Presbyterian University, Sao Paulo, Brazil
ABSTRACT
Due to the evolution of technology and the diffusion of digital television, many researchers have
studied more efficient transmission and reception methods. This fact occurs because of the
demand of transmitting videos with better quality using new standards such 8K SUPER Hi-
VISION. In this scenario, modulation techniques such as Filter Bank Multi Carrier, associated
with advanced coding and synchronization methods, are being applied, aiming to achieve the
desired data rate to support ultra-high definition videos. Simultaneously, it is also important to
investigate ways of channel estimation that enable a better reception of the transmitted signal.
This task is not always trivial, depending of the characteristics of the channel. Thus, the use of
artificial intelligence can contribute to estimate the channel frequency response, from the
transmitted pilots. A classical algorithm called Back-propagation Training can be applied to
find the channel equalizer coefficients, making possible the correct reception of TV signals.
Therefore, this work presents a method of channel estimation that uses neural network
techniques to obtain the channel response in the Brazilian Digital System Television, called
ISDB-TB, using Filter Bank Multi Carrier.
KEYWORDS
Channel estimation, Artificial intelligence, ISDB-TB, FBMC, Neural Networks.
1. INTRODUCTION
Digital TV standards in current use allow for the transmission of standard or high definition video
content. However, consumers are demanding more resolution for more realistic experiences while
watching TV. Therefore, researchers all over the world are working in the development of the
concepts that will enable the broadcast of ultra-high definition content. These concepts include
novel modulation schemes, powerful channel estimation, intelligent receptors, antenna arrays, etc.
Filter Bank Multi Carrier (FBMC) is a modulation technique that has been applied as an
alternative to the Orthogonal Frequency Division Multiplexing (OFDM) [1]. This trend is due to
the fact that FBMC does not use the Cyclic Prefix (CP), increasing significantly the system data
rate [2]. Thus, when FBMC is applied to digital television system such as Integrated Services

60 Computer Science & Information Technology (CS & IT)
Digital Broadcasting Terrestrial B (ISDB-TB) associated with channel coding and synchronization
techniques such as Low Density Parity Check (LDPC) and Bootstrap, 8K transmission may be
made viable. Nevertheless, the transmission of more information per second using the same
frequency bandwidth leads to an increase in bit error rate if improved channel estimation
algorithms, powerful error correcting codes and novel equalizers are not used. However, the
channel estimation becomes a little more complex, due to the characteristics of the used filters
and degrading components present on the channel [3]. In this article, an intelligent channel
estimation algorithm using Neural networks is reported.
The channel estimation is a crucial stage for the perfect reception of digital TV signals because of
the interferences that are generated by several sources on the channel. In the specific case of the
terrestrial broadcast, these are Additive White Gaussian Noise (AWGN), multipath, and others
[4]. Therefore, it is necessary to use different techniques and processes that make possible the
removal and minimization of this effects, allowing the reception of the transmitted signal in an
appropriated way.
Several techniques can be used to estimate the frequency response of the channel. Among them,
we can highlight those that use pilots associated to interpolation methods that adequately
minimize the AWGN and multipath effect [5]. However, this process is not always trivial, and
can be improved with the use of Artificial Intelligence (AI).
The AI can be understood as a set of algorithms to solve complex problems, being able to resolve,
to make decisions, and to develop a method of learning, according to the situation to which it is
applied [6]. In this context, Neural Networks (NN)are used to solve problems through the
simulation the connection between brain neurons, using specific activation and training
algorithms [7].
The herein proposed channel estimation method uses a NN, trained through the Back-propagation
algorithm to calculate the channel response and to permit the correct equalization in a scenario
with AWGN and multi paths. This technique is applied to an FBMC version of the ISDB-TB
digital TV standard. The overall system is simulated on GNUR adio environment. The presented
results show the feasibility of this new system even when transmission id performed in a channel
with severe multipath interference.
2. ISDB-TB
The Brazilian Digital Television System (SBTVD),ISDB-TB, is the terrestrial digital TV standard
adopted by 18 countries in the world. It employs a bandwidth of 6, 7, or 8 MHz and can transmit
One-Segment (1SEG), Standard Definition Television (SDTV) and High Definition Television
(HDTV), according to the combination chosen for the 13 Segments available. In the Figure 1 is
possible to see the bandwidth segmentation used in 6 MHz. The 3 modes of operation have
different parameters that are detailed in Table I [8].
Table 1. ISDB-TB transmission parameters.
Mode
Parameters
Carriers Useful Carriers Pilots Symbol period IFFT
1 1405 1248 157 0,252 ms 2048
2 2809 2496 313 0,504 ms 4096
3 5617 4992 625 1,008 ms 8192

Computer Science & Information Technology (CS & IT) 61
Figure 1. Segmentation of the ISDB-TB channel.
3. FBMC
The Filter Bank Multi Carrier (FBMC) modulation technique consists of dividing the available
band into small equally spaced small segments [9]. For this purpose, filter banks complying with
(1) are used.
= − = ∑ ℎ (1)
where, is the frequency, is the number of subcarriers, is the number of filter coefficients,
and = 0 … − 1.
Applying the Z transform, the polyphase decomposition, and making" =
#$%
, (2) can be
derived.
& = ∑ " '
& '
' &' (2)
By expressing this equation in matrix notation, (3) is obtained.
(
&
⋮
&
* = +
1 ⋯ 1
⋮ ⋱ ⋮
1 ⋯ "
. #
/ +
&
⋮
& &
/ (3)
The implementation of this equation is depicted in the block diagram in the Figure 2.
Figure 2. FBMC system model.

The filters are designed according to the zero inter-symbol interference Nyquist criterion to avoid
phase and amplitude distortions [10]. FBMC employs Offset Quadrature Modulation (OQAM), so
that the orthogonality is obtained between symbols and not between subcarriers [11]. Thus, it is
not necessary to use the CP, making it possible to increase the data rate of the system.
4. PILOT BASED CHANNEL ESTIMATION
To estimate the frequency response of the channel, ISDB-TB performs the constant transmission
of pilots. The position of the pilots depends on a Pseudorandom Binary Sequence (PRBS)
sequence that has a generator polynomial equal to 0 + 02
+ 1 [8].
After the transfer function (Hp) is found using (4), where 3' an4' are the pilot amplitudes
received and transmitted through the − 5ℎ subcarrier, an interpolation method that can be linear,
cubic, among others, is used to estimate the responses of the other subcarriers.
' =
67
89
(4)
In the case of linear estimation, we use (5).
= 1 − : ∙ ' + : ∙ ' + 1 5
where :is a constant determined by the relation between the distance of the position of the
subcarrier where it is desired to estimate the response of the channel to the position of the nearest
pilot.
In the case of cubic interpolation, (6) is used.
= = : ∙ ' + : ∙ ' + 1 + > : ∙ & + ? : ∙ & + 1 (6)
where = : , : , > : , and ? : are constants related to a and & @ is the second derivative
obtained fromthe pilot information matrix [12].
5. NEURAL NETWORKS
Neural networks can be understood as algorithms that seek to simulate the functioning of the
human brain, starting from the construction of small computational entities that act as a human
neuron [13]. To do so, we use units called perceptrons (Figure3) which have as input parameters
an input (0), a gain (A), an activation function ( ) and an output (B).
Figure 3. Perceptron.
The activation function can be linear or not depending on the desired application. The most
common are the logarithm and the sigmoid. Perceptrons can be combined to form layers that are
interconnected to generate larger and more complex networks. For the network to work correctly,
it is necessary to perform the network training, using a set of known inputs and outputs, so that
the gains are properly adjusted and the actual inputs generate the desired responses [14]. Among
the training techniques, one of the most used is the Back-propagation Training Algorithm.

5.1. Back-propagation Training Algorithm
This technique uses a generalization of the Least Mean Square [15]. The activation function is
defined as (7).
C =
DEF 7
Initially random weights (A) are defined for the inputs (0). Then from the desired response (H)
the error is calculated by (8).
IIJ = 0.5 ∙ H − B 8
where B is the output of the activation function.
Then the weights are updated, using (9).
A 5 + 1 = A 5 − M ∙ IIJ ∙ 0 9
where 5 is the previous iteration of the algorithm has been set and M is a chosen gain. Finally, the
algorithm is repeated until the desired response is obtained at the output of the system and the
weights are properly adjusted.
6. PREVIOUS DEVELOPED MODEL OF ISDB-TB USING FBMC
In [2] and [16], a modified ISDB-TB system using FBMC was developed, using GNU Radio
Companion (GRC) as simulation environment. The channel estimation algorithm used did not
employ any Artificial Intelligence feature. For this reason, the present work expands those articles
by using a different approach based on Neural Networks.
6.1. GRC
The GRC is a computational tool that allows the development of processing blocks to for
simulating communications systems [17]. It is an open source and free software that makes
possible the interface between the created model and software radio peripherals [18]. It uses a
Graphical User Interface (GUI) that facilitates the software handling [19] [ 20].
The block codes are created using the C/C++ or Phyton languages and the interconnection among
these blocks is described only in Phyton [21].
The processed data sources on GRC are of complex (8 bytes), float (4 bytes), int (2 bytes), or byte
(1 byte) types. The used terminology of GRC is presented on Table II [22].
Table 2. GRC terminology.
Name Definition
Block Processing Unit with ins or outs
Port Block in or out
Source Data generator
Sink Data consumer
Connection Data flow from an output to an input
Flow Graph Set of blocks and connections
Item Data unit
Stream Continuous flow of items

Name Definition
IO Signature In and out description
Then using GRC and programming language the system can be simulated.
6.2. Flow Graph of ISDB-TB FBMC
The transmitter implemented on GRC is presented in the diagram shown on Figure 4.
Figure 4. ISDB-TB FBMC transmitter.
As it can be seen at the transmission side, an information source generates data that is modulated,
formatted according to the standard, processed by OQAM pre-processing, multiplied by Beta,
modulated through the IFFT and Synthesis filters and transmitted through the channel.
The receiver diagram is shown in the Figure 5.
Figure 5. ISDB-TB FBMC receiver.
At the reception, the data goes through the analysis filters and FFT, multiplied by conjugated
Beta, processed in OQAM post processing. After the zeros and pilots are removed, the data is
decoded to calculate the Bit Error Rate (BER).
The GRC environment flow graph is depicted in Figure 6.

Figure 6. Flow Graph of ISDB-TBFBMC.
7. PROPOSED IA ESTIMATION METHOD
To accomplish the AI estimation, a simple NN with one perceptron for each real and imaginary
part of received symbol is used. When the system initiates, four FBMC symbols are sent as
training sequence and the weights of NN are trained using the Back-propagation method. Then
regular operation starts and the received data symbols are equalized by the trained system.
The Flow Graph used was shown in Figure 5. Inside the “ISDBT_B_deframe” hierarchical block,
three different channel estimators were implemented: the linear interpolation and the cubic
interpolation, both at the time and frequency, and a neural network estimator trained with the
Back-propagation technique (Figure 7).
Figure 7. Content of the hierarchical block “ISDBT_B_deframe”.
Thus, it was connected each channel estimator ate the system and the results could be collected.

5. RESULTS
The analysis were made using the ISDB-TB FBMC in mode 3 as in Table I. The Bit Error Rate
(BER) curves were observed on two scenarios. The first is characterized by AWGN and
modulation level equal to 64 (64-QAM) or 6 bits per symbol (Figure 8). The second uses the
Brazil A digital TV channel model [21], which applies 6 paths with 0, 0.15, 2.2, 3.05, 5.86, and
5.93 microseconds of delay and 0, 13.8, 16.2, 14.9, 13.6, and 16.4dB of attenuation respectively,
using the modulation level 4 (4-QAM) (Figure 9). In this last case, it was necessary to reduce the
QAM modulation due to the long time required to perform the real-time simulation.
It can be observed that the use of neural networks brought to the system an increase of robustness
at 10 O
BER level around to 2.1H in the case only of AWGN and around to 2H in the case
which there is AWGN and multipath.
Figure 8. BER curves of ISDB-TBFBM Cusing 64-QAM with AWGN inserted.
Figure 9. BER curves of ISDB-TBFBM Cusing 4-QAM with AWGN and multipath inserted.

6. CONCLUSION
Current digital TV standards were established to enable the broadcast of standard or high
definition video. Nevertheless, nowadays consumers are demanding even higher definition
content. For this reason, researchers are working on new standards that will enable a higher
information transmission rate in terms of bits/s/Hz. This is the case of FBMC which was also
proposed for forthcoming 5G cellular standard. But this increase in bit rate on the same frequency
bandwidth implies in an increase in bit error rate if novel channel estimation and equalization are
not created.
The use of artificial intelligence applied to channel estimators for FBMC opens a new field of
research. The possibility of employing smart algorithms that can learn even in the presence of
nonlinear interference is paramount to the success of more spectrally efficient modulation
techniques.
In this paper, it was showed that the application of a simple neural network to the problem of
channel estimation in the FBMC modified ISDB-Tb digital TV standard is feasible. The
presented technique achieved an increase in 15%in robustnessin a channel with several multipath
interferences. It was also shown that the Back-propagation training algorithm allows the
estimation of the channel frequency response and contributes to minimize the bit error rate
In future, other kind of neural networks will be investigated, such as a recursive network, since it
can improve the results and required computing effort.
ACKNOWLEDGEMENTS
The authors would like to thank the MACKPESQUISA, Coordination for the Improvement of
Higher Level Personnel (CAPES) and National Counsel of Technological and Scientific
Development (CNPq) for the partial financial subsides for this research.
REFERENCES
[1] Bellanger M, Mattera D, Tanda M. A filter bank multicarrier scheme running at symbol rate for future
wireless systems. Wireless Telecommunications Symposium (WTS), 2015, 2015; 1–5,
doi:10.1109/WTS. 2015.7117247.
[2] Almeida J. J. H, Akamine C, Lopes P. B. A proposal for the next generation of isdb-tb using fbmc in a
sdr implementation on gnu radio environment. 2016 8th IEEE Latin-American Conference on
Communications (LATINCOM), 2016; 1–6, doi:10. 1109/LATINCOM.2016.7811601
[3] Bellanger M. FBMC physical layer: a primer. PHYDYAS, January 2010; :1–31URL: http://www.ict-
phydyas.org/teamspace/internal-folder/FBMC-Primer_06-2010.pdf.
[4] Andreas M. F. Wireless Communications. Second edition, John Wiley & Sons Ltd: California, 2011.
[5] Ishini, A. K., Akamine, C.; 2009. Técnicas de estimação de canal para o sistema isdb-tb. Revista de
Radiodifusão.
[6] Russel S, Norvig, P. Inteligência Artificial. 2a ed., Rio de Janeiro: Campus, 2004.
[7] Lippmann R, An introduction to computing with neural nets, in IEEE ASSP Magazine, vol. 4, no. 2,
pp. 4-22, Apr 1987. doi: 10.1109/MASSP.1987.1165576
[8] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS. NBR 15601: Televisão digital terrestre -
Sistema de transmissão. Rio de Janeiro, 2008.

[9] Cherubini G, et al. Filter bank modulation techniques for very high speed digital subscriber lines.
IEEE Communications Magazine, v. 38, n. 5, p. 98-104, May 2000.
[10] Bellanger M. G. Specification and design of a prototype filter for filter bank based multicarrier
transmission. Acoustics, Speech, and Signal Processing, 2001. Proceedings. (ICASSP ’01). 2001
IEEE International Conference on, vol. 4, 2001; 2417–2420 vol.4, doi:
10.1109/ICASSP.2001.940488.
[11] Siohan P, Siclet, C, Lacaille N. Analysis and design of OFDM/OQAM systems based on filterbank
theory. IEEE Transactions on Signal Processing, v. 50, n. 5, p. 1170-1183, May 2002. ISSN 1053-
587X.
[12] Kang S. G, Ha Y. M, Joo E. K. A comparative investigation on channel estimation algorithms for
ofdm in mobile communications. IEEE Transactions on Broadcasting, v. 49, n. 2, p. 142-149, June
2003.
[13] Haykin S. Redes Neurais: Princípios e Prática. Bookman, 2001, Hamilton, Ontário, Canadá. 2ª
Edição.
[14] Burse K, Yadav R. N, Shivastava S. C. Channel Equalization Using Neural Networks: A Review, in
IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 40,
no. 3, pp. 352-357, May 2010.
[15] Lippmann, R. An introduction to computing with neural nets, in IEEE ASSP Magazine, vol. 4, no. 2,
pp. 4-22, Apr 1987. doi: 10.1109/MASSP.1987.1165576
[16] Almeida J. H,Akamine C, Lopes P. B. A proposal for the next generation of ISDB-Tb using FBMC in
a SDR implementation on GNU radio environment, IEEE Latin-American Transactions.2017, July.
Available
in:http://www.revistaieeela.pea.usp.br/issues/vol15issue7July2017/Vol15issue7July2017TLA.htm
[17] Maheshwarappa MR, Bridges CP. Software defined radios for small satellites. Adaptive Hardware
and Systems (AHS), 2014 NASA/ESA Conference on, 2014; 172–179,
doi:10.1109/AHS.2014.6880174.
[18] Stoica RA, Severi S, de Abreu GTF. On prototyping ieee802.11p channel estimators in real-world
environments using gnuradio. 2016 IEEE Intelligent Vehicles Symposium (IV), 2016; 10–15,
doi:10.1109/IVS.2016.7535356.
[19] Larroca F, Guridi PF, Sena GG, Gonzlez-Barbone V, Belzarena P. gr-isdbt: An isdb-t 1-segment
receiver implementation on gnu radio. Computing Conference (CLEI), 2015 Latin American, 2015;
1–8, doi:10.1109/CLEI.2015.7360050.
[20] Müller A. DAB Software Receiver Implementation. Technical Report 2008. URL:
http://people.ee.ethz.ch/˜andrmuel/files/gnuradio/.
[21] Vachhani K, Mallari RA. Experimental study on wide band fm receiver using gnuradio and rtl-sdr.
Advances in Computing, Communications and Informatics (ICACCI), 2015 International Conference
on, 2015; 1810–1814, doi:10.1109/ICACCI.2015.7275878.
[22] Gnuradio. http://gnuradio.org/redmine/projects/gnuradio/wiki 2016. Acessed in 07/11/2016.
[23] ITU Radio communication Study Groups: Document 6E/TEMP/131-E, Guidelines And Techniques
For The Evaluation Of DTTB Systems, 19 March 2003.

Computer Science & Information Technology (CS & IT)
AUTHORS
Jefferson Jesus Hengles Almeida
degree in Electrical Engineering from Mackenzie Presbyterian University, São Paulo,
Brazil, in 2014. Received his M.Sc Degree in Electrical Engineering from Mackenzie
Presbyterian University, São Paulo, B
degree in Electrical Engineering in Mackenzie Presbyterian University. His current
research involves broadcasting areas, digital television transmission systems
software defined radio.
Paulo Batista Lopes was received the B.Sc. and M.Sc. in EE from the Universidade
Federal do Rio de Janeiro, Brazil, in 1978 and 1981, respectively, and the Ph.D. in EE
from Concordia University, Montreal, Canada, in 1985. From 1985 to 1988, he was with
Elebra and CMA, two Brazilian companies, working on the design of several
Communication equipments. From 1988 to 1999, he was with Texas Instruments as a
DSP specialist. In 1999, he moved to Motorola
Semiconductor) as a Sales and Appli
Universidade Presbiteriana Mackenzie as a professor in the School of Engineering. His
research interests are Circuit Theory, Digital Signal Processing, and Analog Circuit Design.
Cristiano Akamine received his B.Sc. degree in Electrical Engineering from Mackenzie
Presbyterian University, São Paulo, Brazil, in 1999. He received his M.Sc. and Ph.D.
degree in Electrical Engineering from the State University of Campinas (UNICAMP),
São Paulo, Brazil, in 2004 and 2
Systems, Software Defined Radio and Advanced Communication Systems at Mackenzie
Presbyterian University. He has been a researcher in the Digital TV Research Laboratory
at Mackenzie Presbyterian University sin
with many digital TV systems. His research interests are in system on chip for broadcast
TV and Software Defined Radio.
Nizam Omar, Mechanical Engineer ITA 1974. Master in Applied Mathematics, ITA
1979, Ph.D. in Computer Science -
Presbyterian University in Artificial Intelligence and its applications in Education,
Engineering and Economics.
Computer Science & Information Technology (CS & IT)
was born in Cotia, on May 1992. Received his B.Sc.
degree in Electrical Engineering from Mackenzie Presbyterian University, São Paulo,
Brazil, in 2014. Received his M.Sc Degree in Electrical Engineering from Mackenzie
Presbyterian University, São Paulo, Brazil, in 2016. He is currently studying his Ph.D
degree in Electrical Engineering in Mackenzie Presbyterian University. His current
research involves broadcasting areas, digital television transmission systems studies,and
was received the B.Sc. and M.Sc. in EE from the Universidade
Federal do Rio de Janeiro, Brazil, in 1978 and 1981, respectively, and the Ph.D. in EE
from Concordia University, Montreal, Canada, in 1985. From 1985 to 1988, he was with
CMA, two Brazilian companies, working on the design of several
Communication equipments. From 1988 to 1999, he was with Texas Instruments as a
DSP specialist. In 1999, he moved to Motorola-SPS (later to become Freescale
Semiconductor) as a Sales and Application manager. Since 2009, he has been with
Universidade Presbiteriana Mackenzie as a professor in the School of Engineering. His
research interests are Circuit Theory, Digital Signal Processing, and Analog Circuit Design.
is B.Sc. degree in Electrical Engineering from Mackenzie
Presbyterian University, São Paulo, Brazil, in 1999. He received his M.Sc. and Ph.D.
degree in Electrical Engineering from the State University of Campinas (UNICAMP),
São Paulo, Brazil, in 2004 and 2011 respectively. He is a professor of Embedded
Systems, Software Defined Radio and Advanced Communication Systems at Mackenzie
Presbyterian University. He has been a researcher in the Digital TV Research Laboratory
at Mackenzie Presbyterian University since 1998, where he had the opportunity to work
with many digital TV systems. His research interests are in system on chip for broadcast
, Mechanical Engineer ITA 1974. Master in Applied Mathematics, ITA
- PUC RIO 1989. He is Professor at the Mackenzie
Presbyterian University in Artificial Intelligence and its applications in Education,
69

CHANNEL ESTIMATION FOR THE ISDBT B FBMC SYSTEM USING NEURAL NETWORKS: A PROPOSAL OF APPLICATION OF BACK-PROPAGATION TRAINING ALGORITHM

More Related Content

What's hot (20)

Similar to CHANNEL ESTIMATION FOR THE ISDBT B FBMC SYSTEM USING NEURAL NETWORKS: A PROPOSAL OF APPLICATION OF BACK-PROPAGATION TRAINING ALGORITHM (20)

Recently uploaded (20)

CHANNEL ESTIMATION FOR THE ISDBT B FBMC SYSTEM USING NEURAL NETWORKS: A PROPOSAL OF APPLICATION OF BACK-PROPAGATION TRAINING ALGORITHM