Code Utilization in HSDPA

Tao Chen1, Huibin Lin2, Zhigang Yan3, Jing Liu3, Andreas Mller4
1

Nokia Research Center, 2Nokia Networks, 3China Mobile (CMCC), 4University of Stuttgart

{Tao.T.Chen, Huibin.Lin}@nokia.com, {Yanzhigang, Liujing}@chinamobile.com, [email protected] ABSTRACT In this paper, we study the impact of the code utilization in HSDPA on the system performance on both a qualitative and quantitative level. The extent of the HSDPA performance improvement, which can be achieved by reserving more codes for HSDPA, can be dependent on the traffic load, the channel conditions, the scenarios, the packet scheduling algorithms, etc. Furthermore, we draw some conclusions about HSDPA network planning and optimization for optimum code utilization. Index terms HSDPA, adaptive modulation and coding (AMC), performance evaluation I. INTRODUCTION High Speed Downlink Packet Access (HSDPA) has been introduced in Release 5 of the Universal Mobile Telecommunications System (UMTS) and can provide higher bit rates for packet data transmissions with peak data rates up to 10.8 Mbps. The key enabling technologies for reaching such high bit rates are link adaptation (LA) with adaptive modulation and coding (AMC), a fast hybrid automatic repeat request (HARQ) mechanism on the physical layer, a shortened transmission time interval (TTI) of 2 ms as well as fast scheduling algorithms in the Node-B, see for example [1], [2]. In the HSDPA network planning and optimization phase, a crucial question for the network operators is how to take fully advantage of HSDPA. AMC as one of the key technologies of HSDPA is responsible for the selection of an appropriate modulation and coding scheme (MCS), which generally consists of the code resource, the modulation scheme (QPSK or 16-QAM), and the coding rate. The number of codes that are available for the HS-DSCH affects the system performance such as the user throughput, the cell throughput and the call success rate. In general, it is the better for the HSDPA system performance with more reserved codes. However, more codes can not only increase the complexity of the UE decoding but also consume the limited downlink code resources. It might result in that fewer codes can be used for conventional dedicated channels anymore. Therefore, code utilization in HSDPA is important for the overall system performance. In this context, different schemes are recommended to apply for different scenarios in order to optimize the code efficiency from the system point of view. The remainder of this paper is structured as follows: In section II, we analyze the effects involved with code utilization in theory. Section III provides a short overview of the used simulation model and the actual simulation results are presented in section IV. Finally the conclusions are given in section V. II. CODE UTILIZATION IN HSDPA In this section, we shortly introduce the code allocation procedure involved with the code utilization in HSDPA and we analyze the factors influencing the selection of an appropriate MCS. A. The basic code allocation procedure Code allocation as one part of the MCS selection is governed by the AMC. The basic procedure of this mechanism is shown as below. At first, the UE measures the Ec/No on the common pilot channel (CPICH) and maps it to a suitable channel quality indicator (CQI), taking into account the general offset between the HS-PDSCH power and the CPICH power as described in [3]. Such a CQI value is usually periodically sent to the respective serving Node-B and represents a recommendation of the UE, which modulation scheme, transport block size and how many codes currently might be used to transmit data to the corresponding user. As there is always a certain delay between the time of the channel quality measurement in the UE and the time of actual data transmission. the channel quality at the scheduling instant generally deviates from the reported channel quality. The BS estimates the channel quality of each active HSDPA user based on the received individual CQI reports and other information, such as power control commands on the associated dedicated channels and HARQ feedback information (Ack/Nack). In addition, usually an outer loop link adaptation technique is applied as described in [4], [5], which might be realized by adding an offset value to all received CQI values. This offset is normally based on the ACKs/NACKs belonging to past transmissions and adjusted on a continuous basis. Scheduling and link adaptation are then conducted at a fast pace depending on the active scheduling algorithm and the user prioritisation scheme utilizing the compensated CQI. The scheduled UE decodes the HS-SCCH for the information of the MCS that is based on the compensated CQI. After two slots delay, the data carried on the HS-DSCH is decoded by the UE according to the corresponding MCS.

0-7803-9335-X/05/$20.00 2005 IEEE

389

B. Factors affecting the code utilization According to the code allocation procedure, it is quite obvious that the current channel conditions and the packet scheduling algorithms have a significant impact on the code utilization, which is an important part of the MCS. The current channel conditions are closely related to many factors, such as the terminal moving speed, the channel power delay profile and the cell scenarios. The reported CQI value, which reflects the current channel conditions in the downlink, is a major factor that influences the MCS selection. High movingspeeds generally lead to fast-changing channel conditions. The performance degrades deeply with speeds higher than 20km/h, as was shown in [7]. Different cannel power delay profiles, for example ITU Pedestrian-A and Vehicular-A, have different relative power strengths and time dispersion. Different scenarios such as the macro cell and the micro cell have different orthogonalities and interference levels. Therefore, the channel conditions obviously have a direct impact on the system performance. Packet scheduling algorithms also affect the code utilization. In the following study, we only consider the time-multiplexing of different users, i.e. different users are scheduled at different time slots. The scheduled UE may occupy the whole allocated code resources or just a selection of them, but doesn't have to share codes. The code utilization is indirectly dependent on the scheduled UE. If the scheduled UE observes relatively good channel conditions, the performance is relatively good, whereas poor channel conditions of the scheduled users generally lead to the performance degradation. The performance of different packet scheduling algorithms has been studied extensively in literature. Generally, the Proportional-Fair (P-FR), Fair-Throughput (F-TP), max C/I algorithms and Round-Robin (RR) are most often used. Most of them can be described by a generalized formulation of the scheduling metric as follows:

The simulation tool used for our studies is a dynamic system level simulator similar to the one described in [6]. We consider different scenarios, in particular a macro cell scenario as well as a mixed micro and indoor cell environment. Each user performs web browsing with a packet distribution. The minimum packet size is 100 bytes, the median 25 kB and the maximum 1 MB. We study the UE categories 1-6, 7-8 and 10, supporting a maximum number of 5, 10 and 15 codes respectively. The mapping of the possible CQI indices to the bit rates that can be achieved by utilizing the corresponding MCS are indicated for the different terminal categories in Fig. 1, which refers to [3].
14 12 Throughput (Mbits) 10 8 6 4
QPSK
5codes

15codes

10codes

C1-6 C7-8 C10

2 0 0 2 4 6

16-QAM

8 10 12 14 16 18 20 22 24 26 28 30 32 CQI Index

Fig. 1: CQI index tables for different UE categories The general parameter settings are shown in Table I. Table I. Simulation assumptions
Setting General Parameters Cell environment Macro/Micro Downlink frequency 2GHz Propagation model OKUMURA_HATA Pilot/Node-B power ratio 10% Power delay profile ITU Ped-A/Veh-A Traffic models Web browsing Call timeout period 5s HSDPA Related Parameters Packet scheduler Proportional-Fair(P-FR) /Round-Robin(RR) H-ARQ model Chase Combining (CC) Max. power for HS-DSCH 7W Max. power for HS-SCCH 0.5W 5/10/15 HSDPA code allocation Terminal class Category1-6/ Catergory7-8/Category10 (see [3] for details) Std. Dev. Of CQI 1.5dB meansurement Error Parameter

SchedulingMetric[n ] =
where

( rk [n ]) a ( Rk [n ]) b

rk [n ] is the instantaneous estimated data rate, Rk [n ] the average user throughput, k the current user

and n a discrete time index. For a=1 and b=0, the scheduling metric describes a maximum C/I or throughput packet scheduler (M-TP), whereas it represents a fair throughput (F-TP) scheduler for a=0 and b=1. In case of a RR scheduler, on the other hand, all terminals are allocated the same amount of time chips. III. SIMULATION MODELS

Some key performance indicators (KPI) include the average cell throughput, average user throughput and the call success rate. In this context, the average user throughput is the total number of received (data) bits per user divided by the time that user has been active. Besides, for evaluating the code efficiency from the system performance point of view, the cell throughput per code and the user throughput per code based on

0-7803-9335-X/05/$20.00 2005 IEEE

390

the used mean code number and the code efficiency rate are used separately from different aspects. The code efficiency rate is defined as the ratio between the average number of used codes and the number of reserved codes. IV. SIMULATION RESULTS Simulations were performed for different scenarios and different packet scheduling algorithms. The simulation results are shown and discussed here. There is no obvious improvement for the Veh-A scenario even if a higher number of codes are reserved for HSDPA. This holds for example for the cell throughput (refer Fig. 2), the user throughput (refer Fig. 3) and the call success rate (refer Fig. 4), independent of the underlying environment, be it a macro cell or a micro and indoor mixed cell scenario. On the contrast, the improvements for Ped-A channel profile with more reserved codes on the cell throughput, the user throughput and call success rate are observed clearly.
180% 170% 160% 150% 140% 130% 120% 110% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% -10% 0 10 20 active UEs/cell 30 40

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0 5 10 15 20 25 30 35 40
Veh+5 Veh+10

Veh+5
Veh+10

active UEs/cell

Fig. 4: The call success rate with P-FR in a micro and indoor mixed scenario With the Ped-A channel profile and same packet scheduler type such as P-FR, different scenarios such as the macro scenario or a micro and indoor mixed scenario will impact the code utilization. In Fig. 2 and Fig. 3 for P-FR algorithm and Ped-A environment, the gains in the micro and indoor mixed scenario for the cell throughput and user throughput are higher than those of the macro scenario due to relative good orthogonality and channel conditions. When the offered UE number reaches 18 per cell, the gain will hold for the micro and indoor mixed scenario but increase for the macro scenario. The limited code resource will restrain the performance of the micro and indoor mixed scenario more than that of the macro. Therefore, more reserved codes will be more beneficial for performance of the micro and indoor mixed scenario with the Ped-A environment. Besides, the P-FR scheduling gains will be different in different cell scenarios. In the micro&indoor environment, all terminals experienced channel qualities are almost in the same high level. So the gain is high but the change with different users are not much. While in the macro cell environment all terminals may experience totally different levels of the channel qualities, e.g. much more dynamic range, the multi-user diversity gain from P-FR will contribute more to the gain obtained from increasing the number of available codes.
180% 170% 160% 150% 140% 130% 120% 110% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% -10% 0 5 10 15 20 25 30 35 40 active UEs/cell

Macro+PF+Veh+10
Micro_Indoor+PF+Veh+10
Micro_Indoor+PF+Veh+15

Macro+PF+Ped+10
Micro_Indoor+PF+Ped+10
Micro_Indoor+PF+Ped+15

Fig. 2: Cell throughput gain relative to 5 codes. Normally, both the bad channel conditions and limited resources such as the limited available codes may lead to the unideal system performance. In the Veh-A environment, bad channel conditions are the dominant factor so that even obviously increasing codes can not improve the cell throughput. Conversely, due to the relative good cannel quality in a Ped-A environment, the code resource becomes the limiting factor for the system performance. Consequently, increasing the number of available codes can significantly enhance the whole system performance.
100% 90% 80% 70% 60%
Macro+PF+Veh+10

P-FR+10

P-FR+15
RR+10
RR+15

50% 40% 30% 20% 10% 0% -10% 0 10 20 active UEs/cell 30 40

Micro_Indoor+PF+Veh+10

Macro+PF+Ped+10

Micro_Indoor+PF+Ped+10

Fig. 5: The cell throughput gain relative to 5 codes in micro and indoor mixed cell with Ped-A

Fig. 3: The user throughput gain from 5 to 10 codes

0-7803-9335-X/05/$20.00 2005 IEEE

391

For different packet schedulers, P-FR and RR, the gains vary with the user number differently (refer Fig. 5). The Gain is more sensitive with the change of the UE number for P-FR than that for RR. This is because P-FR can obtain much more multi-user diversity gain compared to RR and will maximize utilizing the code resource. P-FR will sort users considering their current channel conditions and select the first one to schedule while RR will not consider the channel conditions and only allocate resource time-uniformly for each UE. Therefore, P-FR will require relative more reserved codes than RR. In summary, the system performance will be improved more or less with more reserved downlink spreading codes for HSDPA. The extent of the improvement will be mostly dependent on the channel conditions, scenarios, scheduling algorithms and the UE number. On the other hand, the multi-codes decoding required by UE will increase the complexity with more reserved codes. More reserved codes for HSDPA will also lead to less codes available for DCH users. Further, it will be a bottleneck for the downlink capacity due to the shortage of the spreading code. To best evaluate the utilization of codes, the code efficiency becomes a key indicator Table II. The code efficiency in micro&indoor cell environment with P-FR
Offered load(UEs/cell) Veh_5codes Veh_10codes Veh_15codes Ped_5codes Ped_10codes Ped_15codes 5 95.08% 49.58% 32.22% 93.23% 69.27% 53.43% 9 98.64% 51.19% 34.66% 98.82% 80.66% 71.04% 18 99.47% 53.60% 35.54% 99.78% 94.03% 79.73% 36 99.65% 55.17% 35.25% 99.91% 97.47% 81.19%

V. CONCLUSIONS From this study, it has shown that the HSDPA performance can be improved by reserving more downlink spreading codes for HSDPA. The performance gain is strongly related to the load situation of a cell, the current channel conditions, the cell scenarios and the adopted packet scheduling algorithms etc. In case of relatively bad channel conditions, such as for the Veh-A environment, no obvious improvement can be obtained even by reserving more codes for HSDPA. However, Higher gain can be achieved in the micro&indoor mixed scenario by reserving more codes due to better channel conditions. Besides, due to the limited downlink spreading codes, the code efficiency is also a key indicator to evaluate code reservation scheme. Different code strategies for different cell scenarios should be applied in the planning of HSDPA service. Only good code reservation strategies based on the actual situations can give the better performance and the higher code efficiency for the network. REFERENCES [1]High Speed Downlink Packet Access (HSDPA): Overall description;stage 2, 3GPP Technical Specification 25.308, version 5.6.0, Sept. 2004. [2]Harri Holma and Antti Toskala, Eds., WCDMA for UMTS, John Wiley &Sons, 2nd edition, 2002. [3]Physical Layer Procedures (FDD), 3GPP Technical Specification 25.214, ver5.9.0, June 2004. [4]David W. Paranchych and Mehmet Yavuz, A Method for Outer Loop Rate Control in High Data Rate Wireless Networks, in Proceedings of the 56th IEEE Vehicular Technology Conference, Vol. 3, Sept. 2002, pp. 1701 1705. [5]Michiharu Nakamura, Yassin Awad, and Sunil Vadgama, Adaptive Control of Link Adaptation for High Speed Downlink Packet Access (HSDPA) in WCDMA, in Proceedings of the 5th International Symposiumon Wireless Personal Multimedia Communications (WPMC), Oct. 2002,pp. 382 386. [6]Seppo Hmlinen, Harri Holma, and Kari Sipil, Advanced WCDMA Radio Network Simulator, in Proceedings of the IEEE International Symposium on Personal, Indoor and Mobile Radio Conference (PIMRC99), Sept. 1999, pp. 951 955. [7]Andreas Mller and Tao Chen, "On the effects of different user speeds on the performance of high speed downlink packet access", accepted for publication at the IST Mobile and Wireless Communications summit, June 2005.

Here, the code efficiency is defined as a ratio of the mean used codes by UE to the reserved codes for HSDPA. Obviously, the code efficiency will degrade with more reserved codes (refer Table II). And for the bad channel condition such as Veh-A, the efficiency will be lower since there are fewer possibilities for terminals to utilize more reserved codes. For example as 9 UEs/cell and 15 reserved codes, the efficiency with Veh-A is only 34.66%, which is only half of that in Ped-A. Besides, the code efficiency is also closely related with the UE number especially when P-FR is applied. Due to multi-user diversity gain, possibilities adopting high MCS can increase. Therefore, more codes can be used as a part of MCS. It will lead to the improvement of the code efficiency with more users. Therefore, there can be a balance between the system performance and code efficiency for the number of the reserved downlink spreading codes. More reserved codes can improve the system performance but will reduce the code efficiency.

0-7803-9335-X/05/$20.00 2005 IEEE

392