Uwicore vtc07 effect of channel-quality indicator delay on hsdpa performance

Effect of Channel-Quality Indicator Delay
on HSDPA Performance
David Martín-Sacristán1, Jose F. Monserrat1, Javier Gozálvez2 and Narcís Cardona1
1
Polytechnic University of Valencia (UPV) – Mobile Communications Group, Spain
2
University Miguel Hernández – Signal Theory and Communications Division, Spain

damargan@teleco.upv.es, jomondel@dcom.upv.es, j.gozalvez@umh.es, ncardona@dcom.upv.es

Abstract- This paper evaluates the effect of the channel This paper assesses the intrinsic delay of the Channel
estimation inaccuracy on the performance of an HSDPA Quality Indicator (CQI) reporting process and its effect on the
system. This study provides some results from system level HSDPA system performance. As it is demonstrated in this
simulations that have been conducted over a very complete paper, this delay has a severe impact in the efficiency of the
dynamic simulator which models an HSDPA system full link adaptation process and also affects the performance of the
compliant with specifications. This emulator allows scheduling mechanisms since the upper limit of the resources
performing multi-user transmission and link adaptation allocated to each user depends on this report. This paper
with a limited modulation and coding scheme (MCS) accomplishes this analysis in a mixed traffic scenario which
selection based on the channel quality indicator (CQI) consists of both best effort (web) and real time traffic (H.263
modes. Many factors such as the user equipment (UE) video telephony). A proper modelling of the HSDPA system
speed, the employed scheduling algorithm or the traffic constraints is implemented and a complete characterization of
load have been considered in the assessment. Moreover the interferences is proposed. That is, to the best authors’
intrinsic constraints of a WCDMA system like HSDPA knowledge, the singularity of this research as compared with
have been also taken into account, i.e. the maximum for example [6] or [7]. The former considers an ideal traffic
number of channelization codes and the maximum source and simplifies the interference variability whereas the
latter uses a very simple model of the channel estimation
transmitted power have been modelled jointly with a
inaccuracy.
complete and dynamic interference characterization.
The presented results prove that not only the mobile speed
Keywords- HSDPA, CQI, Scheduling, Link Adaptation. determines to what extent the CQI reporting delay has an
effect on the system performance but also the scheduling
I. INTRODUCTION algorithm and the system load among other factors.
In order to allow high data rate transmission in Wideband This paper is organised as follows. First, the LA mechanism
Code Division Multiple Access (WCDMA) third generation of HSDPA is briefly cleared up highlighting the time elapsed
mobile networks, High Speed Downlink Packet Access since the user sends its CQI report till the Node-B transmits
(HSDPA) and its uplink counterpart HSUPA have been according to this information. In Section III, all the fast packet
recently standardized in 3GPP. HSDPA is already in a very scheduling algorithms compared in this paper are described.
mature state, and first commercial systems are being rolled out Next, in Section IV the evaluation environment is presented.
in the course of this year. HSDPA achieves high data rates of Finally, the results of different simulations are presented and
up to 14 Mbps by means of adaptive modulation and coding, discussed and the most important conclusions are summarised.
fast scheduling mechanisms (each TTI or Transmission Time
Interval of 2 ms) and a powerful Hybrid ARQ mechanism. II. LINK ADAPTATION BASED ON USER REPORTS
Link adaptation (LA) is a process of paramount importance to In HSDPA the user equipment (UE) is responsible for
optimise system functioning and therefore user equipment reporting periodically to the Node-B the downlink channel
reports channel state either cyclically or in a triggered-based quality. This channel information is numerically represented
manner by means of the Channel Quality Indicator (CQI).
by means of the CQI, whose definition is explained in detail in
Several simulation results for HSDPA have been published
[8]. Numerically CQI extends from 1 to 30, increasing its
in conferences and journal papers during last years. A good
value when channel quality augments. Each CQI can be
survey of HSDPA principles and performance simulations can
be found in [1]. In [2] a closed formulation was proposed for translated into a combination of transmission parameters since
the calculation of CQI whereas in [3] and [4] different [8] establishes a relation among each CQI and a concrete value
reporting schemes were analyzed evaluating the effect of the transport block size (TBS), number of simultaneous
produced by increasing report periods. Several scheduling channelisation codes, modulation and code rate. These
algorithms have also been evaluated (see e.g. [5]) studying parameters were specifically chosen to configure a 1dB
how the algorithm behaves for a constant bit rate traffic granularity in the carrier to interference ratio (CIR) among
pattern. consecutive CQIs for a Block Error Rate (BLER) level of 10%.

1550-2252/$25.00 ©2007 IEEE 804

As the channel state information is provided by the UE to the Robin scheduler prioritises users with an oldest last serving
Node-B there is a non-negligible delay between channel time, MaxCIR gives priority to the users with the highest last
estimation and the reception of this information in the Node-B. reported CQI (the best channel quality) while Proportional Fair
Moreover, some time pass since the Node-B receives the CQI prioritises users with the highest fairness factor ( Ff ) in the
until it uses this information in the LA and scheduling current time interval k defined as:
mechanisms. Besides, at least 2 slots more pass until the UE
Ff i (k ) = Ri (CQI (k )) / Ri (k ) (1)
begins to receive the data scheduled in the Node-B due to the
constant delay between the HS-SCCH channel which signals where Ri (CQI (k )) is the maximum data rate that UE i can
the start of data transmission and the HS-PDSCH which
handles the data. The whole delay is around 6 ms, that is to say, transmit provided the last reported CQI and Ri ( k ) is the mean
three TTIs. data rate of the UE i in the time k, which is updated every TTI
All of the previous delays are inherent in the HSDPA the UE has data waiting for transmission in the buffer
functioning and hence are not controllable. This paper deals according to the next formula:
with the only delay which can be controlled by the system, i.e.
the reporting period delay.
In general, the higher the reporting delay is the higher
(
Ri (k ) = 1 − 1
T ) R (k − 1) + ( 1T ) R (k )
i i (2)

inaccurate the channel estimation is. This inaccuracy depends where T is the number of TTI considered in the averaging
on how the channel and therefore the CIR level vary in the period, Ri (k ) is the updated mean rate, Ri (k − 1) is the old
time domain. This paper assesses the effect of this inaccuracy
on the system performance depending on different factors as mean rate and Ri (k ) is the last instantaneous data rate.
for example the scheduling algorithm employed. Once the candidate list has been ordered the scheduler
allocates resources. Generally the retransmissions are first
III. FAST PACKET SCHEDULING served and afterwards the new transmissions if there are codes
and power left. The scheduler processes a lot of information to
Fast packet scheduling performed in Node B is one of the make a decision as for instance the channel estimation reported
main features of HSDPA. Its implementation is not specified by each user, the buffer size, the power available in the Node-
and investigation in this field can provide a great B for transmission and, what is more important, the difference
differentiation among different HSDPA systems. between this quantity and what the UE considered in the
In each TTI the scheduler makes a decision about to which channel estimation.
UEs the Node-B will transmit in the next TTI and the Once a user has been served, the process is repeated with the
characteristics of this transmission (TBS, number of next UE in the prioritised list until there is no more power or
channelization codes, code rate and power). codes left. The objective of this scheme is to allocate the
The scheduling decision is made taking into account a list of minimum power and the most efficient combination of
candidate UEs which have data to be delivered in the Node-B transport block size and code rate to ensure a block error rate
buffer. The scheduler can differentiate if these data are new or (BLER) of around 10% if possible.
are waiting for retransmissions, giving a highest priority to All the afore-mentioned scheduling algorithms belong to the
retransmissions. A maximum number of code multiplexed first group and provide a more or less good performance when
users is fixed ‘a priori’ in the simulations, setting a limit to the handling best effort traffic but, in order to fulfill the QoS of
number of multiplexed UEs in each TTI. real time (RT) users, other strategies must be considered.
Category 10 UEs have been considered, which are the most Therefore, another kind of resource allocating schemes is
flexible equipments, and only 30 modes of transmission, those needed. The simplest option is to serve first the RT users and
corresponding to the 30 modes defined by the CQI table for later the best effort traffic. This scheme is able to maintain the
this category, are employed. In spite of this restriction, the QoS for RT users if a call admission control (CAC)
number of modes is enough to correctly consider the flexibility mechanism prevents congestion in the cell. This simple
of HSDPA. Moreover, it has been demonstrated that such a differentiation of services provides great results when
resource allocation model can be more robust to the channel considering a mixed scenario with RT and best effort users as
estimation inaccuracy than a more flexible one when the compared with the case in which there is not differentiation.
channel estimation inaccuracy is within some limits [7]. This aspect is clearly appreciated in the results presented in
In this paper two kinds of resource allocating schemes are section V.
analyzed. The former does not guarantee any quality of service
(QoS) since it employs scheduling algorithms that are not QoS IV. EVALUATION ENVIRONMENT
aware. The latter is based on separating different types of
services what can preserve real time services from best effort To conduct this investigation an evolved version of the
services. emulator presented in [9] has been employed, emulating
Within the first group some classic scheduling algorithms HSDPA packet data transmission.
such as Round Robin, Max-CIR and Proportional Fair are The simulator models a multi-tier macrocellular architecture
evaluated. Depending on the algorithm, a different prioritising with one hexagonal central cell with radius of 2800 or 1000
scheme is established among the scheduling candidates. Round meters and 3 additional cell tiers. Simulations are only

805

conducted in downlink since HSDPA is specific of this the experienced CIR of this block in each slot of a TTI. After
direction. Users are on the move within the cell radius. completing the transmission of a whole transport block, the
The available number of codes has also been carefully taken three associated CIR values are averaged and a single CIRavg
into account, and the same assumptions as in [10] have been value is obtained, which represents the quality experienced by
made, assuming a value of 1 for the soft handover overhead. the transport block. A LUT is employed to map the CIR value
The maximum number of HS-SSCH codes has been set to 4, in a BLER value and to decide whether a block is correctly
therefore code multiplexing of 4 users per TTI is allowed. received. When a transport block is received in error, it is not
The HARQ mechanism of HSDPA has been implemented in discarded but stored in the receiver buffer and combined with
a very realistic way. A stop and wait (SAW) protocol with 6 retransmissions according to a specific method. The employed
parallel processes has been considered in order to control the simulator uses the Chase Combining (CC) scheme in which
transmission to each UE. retransmitted blocks are identical to that of the first
The available power is modelled assuming a power transmission.
consumption for the non data channels of HSDPA according to In the simulator web traffic has been modelled as a best
that used in [10]. The total Node-B power is 43 dBm. effort traffic source. The web browsing service has been
Channel modelling comprises path loss, shadowing and also modelled as in [12]. It follows an ON/OFF pattern and a rate of
fast fading. Fast fading modelling is quite important when around 55 kbps per user is expected due to the chosen
considering technologies, such as HSDPA, that base their radio parameters in [12]averaging over a long period of time.
operation on link adaptation techniques. Real-time services have also been included in the simulations
Intra-cell interference on a CDMA system is modelled by through the emulation of real-time H.263 video transmissions
means of an orthogonality factor, which is usually denoted as following the model presented in [13].
α . In absence of multi-path fading, the codes are perfectly
orthogonal and therefore α = 1 . In the worst case α = 0 , V. SYSTEM PERFORMANCE
meaning that orthogonality is entirely destroyed. Typical A meticulous simulation study has been carried out to assess
values of α are between 0.4 and 0.9. In this research a value the joint effect of CQI delay and processing and scheduling.
of 0.8 has been chosen. Thus, the HSDPA carrier to Moreover different scenarios have been considered varying
interference (CIR) level can be expressed as follows: user speed and traffic patterns. The user speed is 3 km/h for the
Pi pedestrian users and 50 km/h for the vehicular users.
ψi
LP ⋅ LS ii
ii

CIRHSDPA = (3) A. Performance with Saturated Traffic Sources
( PTi − P ) ⋅ (1 − α )ψ + P
LP ii ⋅ LS ii
i ∑ L ij ⋅ L ijψ j + N 0 ⋅W
Tj
Initially, to determine the maximum cell capacity of the
j∈Ω P S HSDPA system, users are simulated considering their traffic
where Pi is the addition of the power transmitted in all the sources saturated. Each user has always 80 kbits pending for
channels allocated by the reference cell to the user of interest, transmission in the serving Node-B buffer.
PTi is the total power transmitted by the reference cell, Ω is the Regarding the cell throughput, the MaxCIR algorithm should
set of cells interfering the user and PTj is the total power achieve the highest performance since it allocates more
transmitted by these interferers. In this expression, the resources to the users with the best channel quality, i.e. users
parameters PTi and PTj also include the base station power with the highest available data rate. This fact can be observed
reserved for other channels different from the HSDPA High in Table 1 and Table 2, which show the mean cell throughput
Speed Downlink Shared Channel (HS-DSCH). for the MaxCIR, RR and PF algorithms. The cell throughput is
The simulation tool models all the interfering paths from defined as the total amount of bits correctly received per
each interfering base station to each UE. Other investigations second in a cell. Both tables summarise the results obtained
consider a fixed level of interference coming from other cells after 1800 seconds of system emulation with 15 users
or a fixed ratio between other cells interference and own cell randomly distributed in the cell. Clearly the MaxCIR algorithm
interference. The more realistic approach implemented in this outperforms the other algorithms while the RR scheduler
work allows a more accurate modelling of the interference obtains the worst results as expected.
variability and therefore of the channel quality variability. This In addition, the relation between the CQI reporting period
accurate modelling is of paramount importance to assess the and the cell throughput is clarified in Table 1. It can be
system performance most of all when radio access appreciated a reduction in the cell throughput when the
technologies based on link adaptation, as for instance HSDPA, reporting period increases. This effect is due to the fact that the
are under consideration. CIR level changes dynamically due to the user movement,
In the simulations presented in this paper several look-up shadowing and fast fading and additionally due to the non
tables (LUT) CIR vs BLER have been employed, one for each negligible interference variability. Given that the reporting
CQI value. These LUT are obtained from the European period increment is near milliseconds and the interference
Network of Excellence NEWCOM [11]. variability is highly reduced in a saturated scenario, the fast
In these LUTs it is also included the effect of the HARQ fading changeability is the most important factor which
retransmissions with chase combining. In order to decide if a justifies the reduction in the cell throughput. This difference is
single block is correctly decoded or not the simulator computes higher in the pedestrian scenario than in the vehicular one. For

806

example, the difference in throughput is between 7% and 14% C. Performance with RT traffic.
for the pedestrian case and between 0.2% and 3% for the To measure the level of QoS for the H.263 users, the user
vehicular case. This effect can be explained regarding the satisfaction concept (US) is introduced representing the
channel coherence time. In case of pedestrian users, fast fading percentage of H.263 frames transmitted before the next video
is uncorrelated after 32ms whereas for vehicular users this time frame is generated. The scenario for RT traffic considers a
is reduced down to 2 ms. For this reason, in a vehicular fixed number of 8 H.263 users transmitting at 64 kbps and
scenario the reporting period increment has not a relevant moving in a cell with a radius of 2800m. Figure 1 shows the
effect on the system performance since already with 2ms the results after 2 hours of simulation for RR and MCIR
fast fading is uncorrelated and only the rest of factors, with algorithms. The set of simulated CQI reporting periods is 2, 8,
slower variability, affect the inaccuracy of the channel 16, 32 and 64 ms.
estimation. From the results it can be concluded that the MaxCIR
algorithm outperforms the functioning of the RR for the
TABLE I. CELL THROUGHPUT IN MBPS FOR 1000M CELL RADIUS simulated environment in both the pedestrian and the vehicular
pedestrian vehicular
scenarios. It is worth noting that the user satisfaction decreases
as the reporting period increases but in a different way
CQI reporting
period (ms)
MCIR PF RR MCIR PF RR depending on the mobility. In the pedestrian scenario the slope
2 8.1466 4.6822 4.2911 7.5443 4.7024 4.1867 is more pronounced than in the vehicular scenario.
256 7.6087 4.1181 3.6732 7.5286 4.5412 4.1783
TABLE II. NORMALIZED DELAY FOR MAXCIR ALGORITHM
B. Performance with Best Effort traffic. pedestrian vehicular
N. of CQI reporting
In the next scenario only web browsing users are considered. Users period (ms)
MCIR MCIR
The study is focused in the effect of the CQI reporting period mean 95% mean 95%
depending on the number of users, the scheduling algorithm 2 11.13 23.01 10.30 41.85
and the user speed. 20 16 13.56 26.83 11.10 44.16
64 15.53 31.27 11.60 49.61
Table 2 collects some significant results in terms of the
2 15.84 29.36 15.60 70.34
normalized delay experienced in the transmission of each web 30 16 17.63 34.22 15.47 71.15
object when the MaxCIR algorithm is used. The normalized 64 20.07 39.00 17.85 84.50
delay is defined as the delay in milliseconds required for 2 27.53 45.16 24.35 126.90
transmitting one kbit of information. The mean value and the 50 16 28.86 48.97 28.10 144.40
95th percentile are shown in the Table. It can be observed that 64 30.24 51.02 30.62 156.45
the higher the reporting period the higher the normalized delay,
that is to say, the system performance is deteriorated since a TABLE III. 95TH PERCENTILE OF THE NORMALIZED DELAY
higher normalized delay means more time to transmit the same pedestrian vehicular
data. Besides, increasing the number of users entails a higher
N. of CQI reporting
normalized delay what is quite obvious in interference-limited Users period (ms)
MCIR PF RR MCIR PF RR
systems as HSDPA. Moreover, these results reinforce the idea 2 29.36 32.99 36.55 70.34 51.90 44.89
stated before, that is, in case of a vehicular scenario the 30 16 34.22 34.52 42.43 71.15 52.03 45.92
degradation experienced with increasing reporting periods is 64 39.00 45.12 49.22 84.50 52.98 48.32
greater when the interference variability is higher, i.e. when
more users are active in the cell. 0.9
RR pedestrian
The Table 3 compares the performance of the three 0.89 MCIR pedestrian

scheduling algorithms with a fixed number of users. The 0.88
RR vehicular
MCIR vehicular
MaxCIR algorithm provides the best results in the pedestrian
scenario. PF improves the RR performance since its 0.87
User Satisfaction

functioning is channel state aware and hence it can take 0.86
advantage of the good channel estimation. In spite of the
0.85
greater fairness of PF as compared with MaxCIR, in this
scenario with 30 users it is not enough to improve the MaxCIR 0.84

performance in terms of delay. On the other hand, in the 0.83
vehicular scenario all the algorithms suffer degradation in their
performance, more pronounced in the case of the MaxCIR 0.82

algorithm since this algorithm is totally channel state 0.81

dependant and in the vehicular environment the channel 0.8
estimation present lower accuracy. Again the degradation with 2 8 16 32 64
CQI reporting period (ms)
increasing reporting periods is lower in the vehicular scenario
for all the algorithms. Figure 1. User satisfaction vs CQI reporting period with 8 H.263
users at 64 kbps for RR and MaxCIR algorithms

807

TABLE IV. USER SATISFACTION WITHOUT SERVICE DIFFERENTIATION Therefore, it is possible to optimise the CQI reporting period to
MCIR PF RR
reduce the number of CQI transmissions while maintaining the
CQI reporting
users QoS and the overall system performance.
H263 WWW H263 WWW H263 WWW
period (ms)
2 0.8247 0.9715 0.8155 0.9623 0.7847 0.9540
ACKNOWLEDGMENT
16 0.7965 0.9693 0.7789 0.9648 0.7521 0.9512 This work has been partially funded by CICYT (Spanish
32 0.7885 0.9654 0.7705 0.9559 0.7367 0.9513 National Science and Technology Council) and the FEDER
64 0.7762 0.9672 0.7296 0.9488 0.7325 0.9434 program of the European Commission under the project
TEC2005-08211-C02.
TABLE V. USER SATISFACTION WITH SERVICE DIFFERENTIATION
REFERENCES
MCIR PF RR
[1] H.Holma and A.Toskala, HSDPA/HSUPA for UMTS. John
CQI reporting Wiley and Sons, Ltd, 2006.
H263 WWW H263 WWW H263 WWW
period (ms)
[2] F. Brouwer et alt., “Usage of Link-Level Performance Indicators
2 0.8741 0.9305 0.8624 0.9040 0.8464 0.8969
for HSDPA Network-Level Simulations in E-UMTS”, in Proc.
16 0.8749 0.9221 0.8614 0.9070 0.8423 0.8774
IEEE ISSSTA, Aug. 2004.
32 0.8570 0.9124 0.8604 0.9035 0.8320 0.8801
64 0.8636 0.9033 0.8579 0.8932 0.8211 0.8682
[3] Soo-Yong Jeon; Dong-Ho Cho, “Channel Adaptive CQI
Reporting Schemes for HSDPA Systems”, IEEE
Communications Letters, Vol.10, Issue 6, pp 459-461, Jun. 2006.
C. Performance in a mixed traffic scenario [4] Soo-Yong Jeon; Dong-Ho Cho, “An Enhanced Channel-Quality
Finally a mixed traffic scenario is considered with 8 H.263 Indication (CQI) Reporting Scheme for HSDPA Systems”, IEEE
users and 8 web users. Only pedestrian users are considered. Communications Letters, Vol. 9, Issue 5, pp 432-434, May. 2005.
The cell radius is again 2800 meters. [5] P.J.A. Gutiérrez, “Packet Scheduling and Quality of Service in
HSDPA”, Ph.D.dissertation, Aalborg University, Oct. 2003.
The user satisfaction in WWW is defined as the percentage
[6] T. Kolding. Link and system performance aspects of
of web pages transmitted in less than 4 seconds. Tables 4 and 5
proportional fair scheduling in WCDMA/HSDPA. IEEE Proc.
collect the user satisfaction experimented by the H263 users VTC 2003 vol 3, pp 1717--1722, October 2003.
and the web users in the simulated scenarios. From these [7] M. Döttling, A. Seeger, M. Sikora, “Impact of Imperfect Channel
results it can be concluded that the MaxCIR algorithm presents Quality Feedback and User Terminal Capabilities on Throughput
the best performance and RR shows again the worst of HSDPA”. COST 273, Barcelona, Spain, January 2003.
functioning. The PF represents an intermediate point between [8] 3GPP, Physical layer procedures (FDD), 3GPP TS25.214 V7.1.0.
these two algorithms since it allocates resources in a more [9] J. Gozalvez y J. Dunlop, "System Performance and Adaptive
intelligent way as compared with RR but does not take the best Configuration of Link Adaptation Techniques in Packet-
advantage from good channel conditions as MaxCIR does. Switched Cellular Radio Networks", Computer Networks
In the differentiation of services there is a transfer in the user Journal (Elsevier), pp 404-426, 2006.
satisfaction from the less priority service to the more priority [10] W. Bang, K.I. Pedersen, T.E. Kolding, P.E. Mogensen,
service. By means of this differentiation it is possible to obtain “Performance of VoIP on HSDPA”, IEEE Proc. VTC,
better results in terms of H263 user satisfaction at the expense Stockholm, June 2005.
of a poorer WWW user satisfaction. The results show clearly [11] NEWCOM Project FP6-IST-507325-2002. Available at
http://newcom.ismb.it.
that the service differentiation becomes a good option to
[12] P. Barford and M. Crovella, “Generating representative web
maintain the H263 QoS. An improvement between 5 and 10 workloads for network and server performance evaluation”, Joint
percentage points in the H.263 user satisfaction is observed. It International Conference on Measurement and Modeling of
is worth noting that the H263 user satisfaction is in this Computer Systems, June 1998.
scenario worse than the observed in Fig. 1 since in this case [13] O. Lázaro, D. Girma and J. Dunlop, “H.263 video traffic
some additional interference is produced in the system due to modelling for low bit rate wireless communications”, IEEE
the transmission of web users. PIMRC, pp. 2124-2128, September 2004.

VI. CONCLUSIONS
In this paper it has been evaluated the effect of the channel
estimation inaccuracy on the performance of an HSDPA
system. As explained, this inaccuracy is due to the delays in
the acquisition and processing of the CQI reports. Generally if
the reporting period of the CQI increases the system
performance becomes worse. However, it has been
demonstrated that this degradation depends on the speed of UE
and in some situations on the load or on the resource allocating
scheme. There is a trade off between the performance
improvement produced by frequently reported CQIs and the
degradation in uplink interference produced by these reports.

808

Uwicore vtc07 effect of channel-quality indicator delay on hsdpa performance

More Related Content

What's hot (19)

Viewers also liked (9)

Similar to Uwicore vtc07 effect of channel-quality indicator delay on hsdpa performance (20)

Uwicore vtc07 effect of channel-quality indicator delay on hsdpa performance