The Effect of Code-Multiplexing on the High Speed Downlink Packet Access (HSDPA) in a WCDMA Network Raymond Kwan, Peter H. J. Chong 2, Eeva Poutiainen, Mika Rinne Nokia Research Center, P.O. Box 47, FIN-45 Nokia Group, Finland 2 Nanyang Technological University, School of EEE, Nanyang Avenue, Singapore, 639798 ehjchong@ntu.edu.sg Abstract - The High Speed Downlink Packet Access (HSDPA) is currently an important research topic to enhance the downlink performance of a WCDMA network. It is a set of schemes built on top of the Downlink Shared Channel (DSCH), resulting an extremely high bit rate for a single user in the downlink. When the channel condition is good, a high utilization of the code resource for a single user can be effectively achieved. However, when the channel condition is bad, a fraction of the code resource would be wasted because only a user can use a small fraction of the code resource. Thus, code multiplexing offers a solution to utilize the limited code resource more effectively. In this paper, the effect of code multiplexing on the HSDPA performance is studied in a simulator platform. It means that the code resource is shared among a few users. Multicode transmission is also considered in this study. Two, semi-static and code sharing, multiplexing cases are studied. The number, n, of the code multiplexing users is one of the study parameters. The simulation results show that code-multiplexing can increase the throughput in a WCDMA network by improving code utilization while also decreasing the transfer delay between the base station and the terminals. Keywords Code-multiplexing; HSDPA; WCDMA; I. INTRODUCTION The Wideband Code Division Multiple Access (WCDMA) technology is standardized in the 3 rd generation partnership project (3GPP) []. The High Speed Downlink Packet Access (HSDPA) is currently an important research topic to enhance the downlink performance of a WCDMA network. The current WCDMA specification allows to support bit rates up to 2 Mb/s in indoor environment. The target of HSDPA is to allow WCDMA to support downlink bit rates of Mb/s. In WCDMA, there are three types of transport channels that can be used to transmit packet data. They are the common, the dedicated, and the shared transport channels [2],[3]. The transport channel switching for packet data is typically selected based on the resource requirements of the bearers, the amount of data to transmit, the load portion of the common and shared channels, the received interference, and the radio access capability of different transport channels. Downlink shared channel (DSCH) is one of the transport channels, which is intended to carry high bit rates and high load of packet traffic from the base station (Node B) to the terminals i.e. the User Equipment (UE). The DSCH is a time-shared channel with frame of ms, with effective power control and fast packet scheduling but without soft handover. Benefits of using the DSCH have been reported in [4]-[6]. In 3GPP, the HSDPA [7] is introduced as a concept, which will be applicable on the High Speed Downlink Shared Channel (HS-DSCH). It is a new transport channel shared by a set of terminals in a time division manner per a shorter frame length of a TTI of 2 ms. In order to increase the bit rate, multicode transmission is introduced. The multicode scheme increases the bit rate by splitting the bits of a Transmission Block to more than one channelization code at a TTI time. In this scheme, high rate data stream is divided into a number of lower rate data substreams. All of these sub-streams are transmitted in parallel synchronous multicode channels, so that there is no time delay between each other. As a result, the interference observed by one channel due to the other channels is avoided. The effect and performance of using multicode transmission for HS- DSCH in the HSDPA scheme without code multiplexing have been studied in which one HS-DSCH user is present per cell at a time [9]-[]. With a single HS-DSCH user per cell, all the code resource can be allocated to this user, and a very high bit rate can be achieved. However, when the channel of the allocated user is unfavourable, this user might not be able to utilize all the code resources available. Thus, there could be additional gains, if the code resource were shared among a few users per TTI, so that the use of code resource can be diversified, and better code utilization achieved. The purpose of this paper is to study the effect of codemultiplexing on the HSDPA in a simulated WCDMA network area, in which the users can share the code resource given to a cell. Two, semi-static and code sharing, multiplexing cases are studied. It is expected that the cell performance improve. Since several users are allowed to share the code resource, the use of code resource can be diversified based on the estimated channel conditions at the receivers. Thus, better code utilization can be achieved. The sharing of code resource in the code tree is shown in Fig.. -783-77-/3/$7. (C) 23 IEEE 728
In this paper, a brief introduction of HSDPA is given in Section II. In Section III, the simulation model is described, which is followed by the evaluation criteria in Section IV. Finally, the simulation assumptions and the results are shown in Section V and are followed by the conclusions in Section VI. SF = SF = 2 SF = 4 SF = 8 SF = 6 User User 2 User 3 User 4 Code-tree section reserved for the HS-DSCH at SF 6 Figure. Illustration of the code allocation with four code multiplexed users on the HSDPA. II. HIGH SPEED DOWNLINK PACKET ACCESS Some proposed schemes for the HSDPA [7] are Adaptive Modulation and Coding (AMC), Hybrid ARQ (HARQ), Fast Cell Selection (FCS) and Multiple Input Multiple Output antenna processing (MIMO). In this paper, the scope is in the first two schemes because they will be specified in release 5 whereas the FCS and the MIMO are study items for release 6. A. Adaptive Modulation and Coding It is well known that adapting the transmission parameters to the time-variant channel conditions can greatly enhance the link performance. One good example is the fast power control in a WCDMA network, in which the transmission power is adjusted based on the fast feedback from the receiver. Another form of adaptation is Adaptive Modulation and Coding (AMC), which is used for the HS-DSCH instead of the power control. The AMC changes the modulation and coding from a defined set in accordance with the fast variation of the channel conditions. The channel conditions are measured at the receiver and signaled to the base station as a proposal for a Modulation and Coding Scheme (MCS) for the next TTI. If the signal connection between the base station and the terminal is good, a higher order modulation with a higher coding rate can be selected. In worse channel conditions, a lower modulation order and/or lower coding rate (increased redundancy) is selected [7]. B. Hybrid ARQ redundancy by retransmissions. HARQ is insensitive to the measurement accuracy, but it requires reliable signalling for the retransmission requests [7]. Performance of different HARQ techniques has been recently investigated in [8]. III. SYSTEM SIMULATION MODEL HSDPA models have been implemented in details to the radio network simulator presented in [2]. The HSDPA is simulated with a Transmission Time Interval of 3 slots i.e. duration of 2 ms. The propagation, mobility and packet traffic models are based on UMTS 3.3 [3]. A. Adaptive Modulation and Coding In the network simulations, the radio link errors are generated using an error performance table available from a link simulator [2]. The peak bit rates used for the MCS in the link results are shown in TABLE I, which can be achieved using a single code from the Orthogonal Variable Spreading Factor (OVSF) code-tree with the specified spreading factor of 6. The system simulation assumptions are shown in Table II and the link simulation assumptions in table III. Higher bit rate per HS-DSCH connection can be achieved when multiple, N, OVSF codes are used in parallel as a multicode transmission. Hence, the bit rates can be increased by N times. The frame error rate (FER) performance as a function of received energyper-bit over noise (Eb/No) is shown in Fig. 2 for the example of transmission with four MCS and up to three multicodes. Note that Fig. 2 does not show all the combinations of multicode and MCS because some of the combinations do not provide a higher data rate. TABLE I. MODULATION AND CODING SCHEMES WITH THE ACHIEVABLE BIT RATES ON A SINGLE CODE. Modulation and Coding Scheme QPSK ½ 24 QPSK ¾ 36 6 QAM ½ 48 6 QAM ¾ 72 Bit rate [kbps] TABLE II. SYSTEM SIMULATION ASSUMPTIONS. Cell radius (3-sector antennas) Terminal speed Max. number of retransmissions Std. Deviation of slow fading 933 m 3 kmph 8 db Number of subscribers 5 While the AMC expects feedback from the receiver and adapts the modulation and coding based on the measured channel conditions, the HARQ adaptation increases Packet session arrival rate Packet scheduler.389/sec Round Robin 729
TABLE III. LINK SIMULATION ASSUMPTIONS. Turbo Coding: Max-Log-MAP, 8 iterations SDU is an IP packet, whose maximum size is limited to,5 bytes. Interleaver TTI four-block interleaver 2 ms Average HS-DSCH bit rate per user connection [kbps/user] is the bit rate experienced by a single user connection over the radio channel. It is defined as Channel Environment Terminal speed Single path Rayleigh, classic Doppler spectrum 3 kmph R DSCH, user = N N T b DSCH, i i= DSCH, collection, i, (2) Spreading Factor 6 In the network simulation, the assumption is that the link level behavior does not change with the multi-code transmission given the same modulation and coding is used. Appropriate MCS is selected based on the measurements made by the terminal. The MCS selection is done in every TTI, which provides a high rate of adaptation to the fast changing channel conditions. B. Hybrid ARQ The HARQ is modelled in a simplified way so that when a packet is sent in the first time, the SIR of each slot in a TTI is calculated. During the retransmissions, the SIR of each slot for the first (original) transmission and for the consequent retransmissions are calculated and summed to result the Chase combining gain. After that the mean SIR is calculated over the TTI, and the FER is obtained by using the look-up table for the link performance. Finally, a random frame error is generated based on the FER value. IV. EVALUATION CRITERIA The evaluation criteria are selected for three output measures for the network performance and throughput. Average HS-DSCH bit rate [kbps/cell/mhz] indicates the network throughput and is given as b R =, () k T B where b is the total number of correctly transmitted bits on the HS-DSCH from all the base stations in the simulated system over the whole simulated time, k is the number of cells in the simulation, T is the simulated time and B is the bandwidth [5 MHz]. Transfer delay [ms] is the 95 th percentile of the transmission times of the individual packets within a packet call of a document. The transmission time is the measured delay and defined as the correctly received data from time of arrival at the RNC buffer to the correct reception at the terminal. This corresponds to the Service Data Unit (SDU) delay, where the where the index i is the i th connection during the simulation and N is the total number of connections. b DSCH is the number of DSCH bits successfully transmitted and received over the i th connection during the data collection period T DSCH, collection. The data collection period is defined for the discrete periods of time when the RNC buffer is not empty and the HS-DSCH is active. V. SIMULATIONS A. Semi-static and Dynamic Code Sharing Simulations were used to evaluate the performance of the HS- DSCH in a network with the HSDPA, which includes the AMC and the HARQ. In all simulations, base station power of 4 W is reserved for the HS-DSCH, and it is equally shared among the code multiplexed users in each cell. A cell has the maximum of M max = 5 multicodes, which can be shared among users in that cell. In the semi-static simulation case, the maximum number of multicodes that each HS-DSCH user can have is fixed, and is determined by the number, n, of code multiplexed users in that cell. For example, with 5 multicodes and 5 code-multiplexed users in the cell, each user can have maximum 3 multicodes. The actual number of multicodes allocated to a user depends on the channel conditions of that particular user. Even if the channel condition for a particular user is good, maximum 3 multicodes can be used. On the other hand, in a bad channel condition, fewer multicodes would be allocated to that user. For n =, it implies that code multiplexing is not allowed at all. This simulation case is described as semi-static, because the maximum number of codes per user is fixed depending on the allowed number of code multiplexed users in a cell, while the actual number of allocated multicodes for that user depends on the channel conditions of that user. In the code-shared simulation case, the user i in a cell with n code multiplexed users can have maximum available M i multicodes, where M i i = M max M j, i n, (3) j= and M max is the maximum number of multicodes per cell and M j is the number of mutlicode allocated to user j. For example, 73
let M max for the first user be 5. Due to a particular channel condition, 8 codes can be allocated for this user. As a result, the second user in the queue would have maximum 7 codes available, and so on. Compared to the first set of simulations, the second set provides higher code allocation flexibility and the codes are shared ally. B. Simulation Results Fig. 3 shows the HS-DSCH throughput as a function of the number, n, of code-multiplexed users per cell. It can be seen that as n increases, the HS-DSCH throughput increases. With no code multiplexing, a single user will use all the code resource if the channel condition is good. However, if the channel condition is bad, that user can use few of the codes, resulting in wasting the rest of the codes. With code multiplexing, most or all of the code resource can be used even in a bad channel condition because the code resource is shared among the users. This observation is in line with the reasoning that by allowing more code-multiplexed users on the HS- DSCH, the code resource utilization increases, giving rise to an increase in the WCDMA network throughput. While it is true that code multiplexing can increase the efficiency in code resource allocation, giving a full flexibility of the code resource can cause more codes to be allocated for the first few users in the queue. Since a full flexibility in code allocation is given, the first few users would use more codes if their channel conditions are good, leaving insufficient code resources for the other users in the queue. For the simulation case, with maximum multicode ally shared, a small drop in the throughput is observed as n increases from to 5 per cell. In this study, the transmission power reserved for each code multiplexed user is equal. Thus, when there are more code multiplexed users, less power is given to each one. The end result is a reduction of power per code channel used. As a result, a possible under utilization of power could be resulted. With smaller allocated power, it is also possible that some users might not be able to use any or as many of the codes as possible, resulting a slight under utilization of codes. This explains why the throughput decreases when the number of code multiplexed users becomes too high. With maximum multicode fixed, the simulation case is invalid for n =. Hence, no comparison can be made from n = to 5. It can be seen that the HS-DSCH throughput is significantly better, when the code resource is ally shared among code multiplexed users. In both cases, if there were not enough multiplexed users, some power and code resource would have to be wasted. Fig. 4 shows the transfer delay as a function of n. It can be seen that the transfer delay decreases with n, except for the simulation case with n from to 5, due to increasing throughput. Similarly, the transfer delay for the simulation case is lower than that of the semi-static simulation case. Fig. 5 shows cumulative distributions of the mean HS-DSCH bit rate per connection per user for, 3, 5 and 5 code multiplexed users respectively. As expected, when more code multiplexed users are allowed, the bit rate given to each user is reduced because each user would have a smaller amount of the transmission power. Thus, each user can be allocated a less number of codes. A higher mean bit rate can be obtained for code sharing because the code resource is not thresholded by a hard limit as in the semi-static case. Thus, the case of code allocation gives more flexibility. Fig. 6 shows the mean HS-DSCH user bit rate as a function of n for both the semi-static and the code sharing case. It is shown that the HS-DSCH user bit rates are better when the code-resources are ally shared. The HS- DSCH user bit rate decreases with n because each user would have less transmission power available and, thus, less code resources. VI. CONCLUSIONS This paper presents the simulation results of code multiplexing HS-DSCH users in a WCDMA network, in which the HSDPA scheme with multicodes, AMC and HARQ are available at every base station. Two, semi-static and code sharing, multiplexing cases are studied. The results show that code multiplexing can improve the code resource utilization, and, subsequently, improve the network throughput and delay. In addition, when the code resource is shared ally among the HS-DSCH users in a cell, even higher network performance is observed. It can be seen that there is an advantage to allow a few code multiplexed users to share the HS-DSCH channel per TTI due to more transmission power. Allowing an increasing number of code multiplexed users to share the HS-DSCH channel per TTI does not give further gains and may even degrade the performance. ACKNOWLEDGEMENTS The authors would like to thank Ph. Lic. Mika Kolehmainen for his valuable help with the radio network simulator, Dr. Troels Emil Kolding for his expertise in link simulations, and Dr. Esa Malkamäki for helpful discussions. REFERENCES [] http://www.3gpp.org [2] H. Holma, A. Toskala et al. WCDMA for UMTS Radio Access for Third Generation Mobile Communications, John Wiley & Sons, 2. [3] 3 rd Generation Partnership Project; Technical Specification Group Radio Access Network; Physical channels and mapping of transport channels onto physical channels (FDD), TS 25.2 V3.. (999-2). [4] A. Ghosth, M. Cudak, and K. Felix, "Shared Channels for Packet Data Transmission in W-CDMA", in Proc. of IEEE Vehicular Technology Conference (VTC), pp. 943-947, 999. [5] R. Kwan, M. Rinne, Performance analysis of the Downlink shared channel, in Proc. of International Conference of Telecommunication (ICT), 2. [6] A. Mate, C. Caldera, M. Rinne, Performance of the Packet Traffic on the Downlink Shared Channel in a WCDMA Cell, in Proc. of International Conference of Telecommunication (ICT), 2. [7] "Physical Layer Aspects of UTRA High Speed Downlink Packet Access", 3G TR25.848, V.. (2). 73
[8] E. Malkamäki, D. Mathew, S. Hämäläinen, Performance of Hybrid ARQ Techniques for WCDMA High Data Rates, in Proc. of IEEE VTC2 Spring, Rhodes, 2. [9] E. Poutiainen, R. Kwan, S. Hämäläinen, P. Chong, "Performance Study of the High Speed Downlink Packet Access (HSDPA) in a WCDMA Network", in Proc. of CDMA International Conference (CIC ), Korea, Nov.2, 2. [] R. Kwan, E. Poutiainen, P. Chong, S. Hamalainen, A Throughput Study for the High Speed Downlink Packet Access (HSDPA) with Different Multicode Selections in a WCDMA Network, in Proc. of Finnish Wireless Communication Workshop (FWCW), Finland, Oct. 2. [] R. Kwan, P. Chong, M. Rinne, Analysis of the Adaptive Modulation and Coding Algorithm with the Multicode Transmission, in Proc. Of IEEE VTC 2, Vancouver, BC, Canada, Sept. 22. [2] S. Hämäläinen, H. Holma, K. Sipilä, "Advanced WCDMA Radio Network Simulator", in Proc. of IEEE PIMRC 99, 999. [3] TR 2 (UMTS 3.3): "Universal Mobile Telecommunications System (UMTS); Selection procedures for the choice of radio transmission technologies of the UMTS". mean transfer delay [seconds].5.95.9 5.75.7 5 maximum multicode fixed maximum multicode ally shared.55 5 5 Figure 4. The transfer delay as a function of the allowed number of codemultiplexed HS-DSCH users per cell. Max 3 multicodes per user cmux user 3 cmux user 2 QPSK /2 code QPSK /2 2 codes QPSK 3/4 6 QAM 3/4.4 mean: 4.293 5 5.4 :.6568 : 2.8943 5 5 FER 3 4 QPSK /2 6 QAM /2.4 5 cmux user :.468 : 2.244.4 5 cmux user :.55748 :.344 5 2 2 4 6 8 2 4 6 8 ρ = E / N b 5 5 5 5 Figure 2. Frame error rate performance for the example of four MCS with multicodes. Figure 5. The cumulative distributions of mean HS-DSCH connection bit rate as for (a) one, (b) three, (c) five and (d) 5 code multiplexed HS-DSCH users per cell. 5 maximum multicode fixed maximum multicode ally shared 4.5 maximum multicode fixed, mean maximum multicode ally shared, mean 45 4 DSCH throughput [kbps/cell/mhz] 4 35 3 DSCH bitrate per connection [Mbps] 3.5 3 2.5 2.5 25 5 5 Figure 3. The simulated HS-DSCH throughput as a function of the number of code multiplexed HS-DSCH users per cell..5 5 5 Figure 6. The mean HS-DSCH user bit rates as a function of the allowed number of code-multiplexed HS-DSCH users per cell. 732