Performance of the Medium Access Control Protocol for the High Speed Downlink Packet Access Gregory Manuel Nokia Research Center P.O. Box 407, 00045 NOKIA GROUP, Finland ABSTRACT In this paper, the Medium Access Control (MAC) protocol for the High Speed Downlink Packet Access (HSDPA) of the Wideband CDMA (WCDMA) radio transmission is studied. In a packet radio network, the delay performance is of special interest. HSDPA is a new concept, which enhances downlink performance of the WCDMA air interface to lower packet delays and higher throughput. The HSDPA MAC protocol is shortly introduced and its impact to the buffering requirements is studied. The retransmission probability and Transport Format selection probability are presented as distributions. Finally, the performance results are analysed from the simulated measures of packet delay distributions and window occupancy distributions. KEY WORDS WCDMA, High Speed Downlink Packet Access, MAC. 1 Introduction The Wideband Code Division Multiple Access (WCDMA) air interface [1,2,3,4,5] has defined common, dedicated and shared transport channels [6]. Particular interest in this paper is the analysis of WCDMA cells with the HSDPA concept. The proposed techniques of this concept include short Transmission Time Interval (TTI) with fast feedback signalling, packet scheduling at the base station and radio link adaptation techniques [7,8]. The radio link adaptation techniques include multicode transmission, Fast Hybrid Automatic Repeat reQuest (FHARQ) and Adaptive Modulation and Coding (AMC) [7,8,9,10]. The HSDPA concept is applicable on a new type of transport channel, which is intended to carry interactive and background packet traffic and optionally streaming. The traffic on the active bearers is available as Transport Blocks transmitted on the High Speed Physical Downlink Shared Channel (HS-PDSCH) [6,7,8,9]. The HSDPA operation requires control channels as the High Speed Downlink Shared Control Channel (HSSCCH) in the downlink and High Speed Dedicated Physical Control Channel (HS-DPCCH) in the uplink [6,7,8,9]. Section 2 shortly gives the reference to the HSDPA concept, section 3 describes the simulator and the simulation assumptions. In section 4, the window occupancy and packet delay distributions are analysed, the results are concluded in section 5. Mika Rinne Nokia Research Center P.O. Box 407, 00045 NOKIA GROUP, Finland 2 High Speed Downlink Packet Access In the WCDMA downlink, the limiting resources are the transmission power of the base station sector (i.e. the cell) and the channelization codes of the Orthogonal Variable Spreading Factor (OVSF) code-tree [5,6]. It is expected that the throughput will increase and packet delays decrease, if the code-tree resource is shared effectively. As many terminals in the cell may share the HSDPA codetree, it is favourable to adapt the Transport Format (TF) [6,7,8] according to the measured channel conditions at the receiver. During good channel conditions at a receiver terminal, a larger Transport Block (TB) [8,9] with more multicodes, higher modulation and less channel coding can be created, whereas during worse channel conditions a smaller Transport Block has to be created. The reverse feedback is specified as the Channel Quality Indication (CQI) report and FHARQ acknowledgements (A or N) [5,6,7,8,9]. As the transmissions of the Transport Blocks are very temporal i.e. 2 ms (three slots), it is possible to efficiently use the channel dynamics and operate the channel at very low Es/No still achieving the target bit error probability by retransmissions of the Transport Blocks without large delays. The physical channel structure and the transmission protocol are depicted in Figure 1. Here, N-channel processes are active for one base station (NodeB) and for two terminals (User Equipment, UE). The performance of HSDPA was analysed earlier in [11,12,13,14], HARQ transmission techniques in [15,16,17,18,19,20,21,22,23], incremental redundancy in [24, 25,26] and AMC e.g. in [27,28]. Figure 1. The physical channel structure and transmission protocol for the High Speed Downlink Packet Access with six FHARQ channels. 2.1 Air interface protocols for the HSDPA concept The Radio Link Control (RLC) protocol [29] operates equally for the Dedicated Channel (DCH) and for the High Speed Downlink Shared Channel (HS-DSCH) transport [6]. The Medium Access Control protocol [8,30] in the Radio Network Controller (RNC) will decide on which transport channel the Protocol Data Units (MAC-d PDU) will be transmitted. For the HS-DSCH transport, the MAC is further split to the base station MAC called the High Speed MAC (MAC-hs), which enables fast radio resource allocation [8,30]. The MAC-hs takes care of the Transport Block scheduling, FHARQ channel allocation and Transport Format selection. The forward signalling on the HS-SCCH serves the MAC-hs at the receiver and contains beside terminal identification parameters for Transport Format and Resource Indication (number and indexes of the channelization codes and modulation), FHARQ process identifier, TB size, redundancy version and new data indicator [9]. The reverse signalling on the HS-DPCCH contains the CQI-report and the FHARQ acknowledgements [9]. In order to perform effectively, the MAC-hs has a buffer and a transmission window for TBs. Each TB includes a Transmission Sequence Number (TSN) in its header to allow in-sequence reception [30]. At the transmitter, every TB in the window has to wait for a positive acknowledgement from the receiver and the window starts from the first non-acknowledged TB. At the receiver, the MAC-hs window exists as well, because the TBs received from the several FHARQ processes need to be reordered for (nearly) in-sequence delivery to the RLC protocol [29]. The mechanisms to read the reordering window are timer based (Timer T1) and limited-window based [30]. Timer (T1) is always running for a correctly received TB with TSN higher than the next expected TSN and the window is moved only if the next expected TSN was received and correctly decoded, if T1 triggers for time-out or if the window limit was reached [30]. Thus, when T1 triggers for a given next expected TSN, all time-out TSNs will be dropped from the window, or if the window gets full, the receiver will follow the upper edge of the window and the TSN at the bottom of the window is always dropped, when a new TB with a higher TSN is received. At the transmitter, the MAC-hs buffer is loaded by new data from the RLC buffers over the Iub-interface. 3 A protocol simulator for the WCDMA radio interface This paper presents simulation studies of the HSDPA protocol in a WCDMA cell. The simulator contains selected features of the Radio Resource Control (RRC) the Radio Link Control (RLC) and the Medium Access Control (MAC) protocols [29,30,31]. The simulator operates with WCDMA parameters [5,6], and the physical performance is available as instantaneous BLock Error Rate (BLER) probability as a function of signal to interference ratio from the lookup tables [4,5,6,32] and from Jakes model [33]. Traffic is generated here by a (non-real time) packet traffic model [32]. This model generates packets by the Poisson process, where the size of each packet is randomly selected from the Pareto distribution having a specified cut-off value [32]. The traffic source parameters are shown in Table 1. The statistics are collected over all packet calls in all sessions during a long simulation. TABLE I. Summary of the traffic source parameters. Traffic source parameters Settings Session arrival process Poisson Session inter-arrival rate High Mean number of packet 500 calls in a session Packet source rate 1,500 kbps Reading time 2 s Mean number of packets in 25 a packet call Packet inter-arrival time 1.536 ms Datagram size distribution Pareto Mean packet size 288 byte Maximum packet size 1,500 byte a of Pareto distribution 1.1 k of Pareto distribution 81.5 TABLE II. Summary of the RLC/MAC parameters. RLC/MAC parameters Settings RLC mode Acknowledged mode RLC transmissions Limited {10} RLC PDU size 440 bit RLC window size 256 PDU Polling scheme Periodic timer (Every k frames) Poll timer 10 frames Poll prohibit timer 10 frames MAC-hs window size {4,16,32} TB Timer T1 {38,50,76,90} ms The RRC protocol model includes simple procedures to setup the radio bearers and simple algorithms to schedule packets. The RLC protocol takes care of processing the PDUs, i.e. polling schemes, retransmissions and selective acknowledgements. The MAC-d schedules the logical channels and selects the transport channel. The MAChs schedules the HSDPA transmissions, selects the Transport Format and creates the Transport Block for each TTI. Fast L1 signalling is modelled on the HS-SCCH channel in the downlink and on the HS-DPCCH channel in the uplink with randomly distributed signalling errors2 . The RLC/MAC parameters are summarised in Table 2. The HSDPA transport is modelled with multicode transmission, FHARQ processes and AMC. The data is available in the 2 This is justified as these channels have fast power control, whereas HS-PDSCH has not. TABLE III. Summary of the transport channel parameters. TrCH parameters Settings RNC scheduler Round Robin Iub delay Distributed [0,30] ms1 DCH bitrate Variable Associated DCH 30 kbps HS-DSCH parameters TTI 2 ms (3 slot) NodeB scheduler Round Robin Spreading factor 16 (fixed) Number of codes, Nc Max {Nc{5,10,15}} L1 transmissions Limited {2,4,6,8} FHARQ channels 6 FHARQ processes {1...6} per UE Transport Format Channelization codes {1...max{Nc}} Channel code rate {1/2,3/4} Modulation {QPSK,16QAM} Channel parameters HS-PDSCH BLER 50% for first transmission HS-SCCH BLER 1% HS-DPCCH p{N|A} 1% HS-DPCCH p{A|N} 1e-4 UE velocity 3.6 kmph, Jakes Simulation time 1000 s base station buffers as MAC-d PDUs, which are assembled to a TB after the TF is selected for each TTI. Thus, exactly one Transport Block per TTI is triggered for transmission. The HS-SCCH and HS-DPCCH control information are formed respectively, and each TB delivery is attached appropriately with the TSN. In the receiver, Chase combining is implemented and the retransmitted symbols are soft-combined to the already received symbol energy to increase the probability of correct decoding for the TB [24]. The transport channel parameters are given in Table 3. 4 Analysis of the packet delay and MAC-hs window occupancy In this section, the impact of MAC-hs operation and parameters to the observed packet delay distributions are analysed. The MAC-hs window occupancy (memory size) is also compared for each case. The packet delay here is the peer-to-peer delay for a network SDU (e.g. a TCP segment) from the time, when the SDU was created at the traffic source in the network till it was received, correctly decoded and reassembled as an identical SDU at the terminal. Any core network delays are not included in the presented delay statistics. Figure 2 shows the distributions of the physical layer (L1) and RLC transmissions. It can be seen that four L1 transmissions was typically enough and the probability of RLC retransmissions was 2 Iub delay was set to 10 ms for the reverse link RLC STATUS PDU. about 5%, which is much lower than e.g. on the DCH operating at 20% BLER. Increasing the number of L1 transmissions further, may make the probability of RLC retransmissions even lower. However, the distribution of RLC transmissions seem to have a long tail, meaning that when the channel dynamics is slow and the transmissions occur during worse channel conditions, it takes a long time and many retransmissions before high enough Es/No is available for the correct decoding of the Transport Block. If even more L1 transmissions are allowed, the FHARQchannel is unnecessarily occupied for a longer time, still not decreasing the RLC retransmission probability notably. These properties of channel dynamics would also argument gains for other than Round Robin types of schedulers. Figure 2. The RLC transmission probability distribution for the HSDPA with channel operation point at about BLER 50% with number of L1 transmissions limited to (a) two and (c) four. The RLC transmission probability shown in (b) and (d) respectively. Figure 3 shows the probability of selecting a given Multicode, Modulation and Coding Scheme (MCS) for a Transport Format during a TTI [9]. Here, the MCS is an ordered set of multicode, modulation and channel coding choice arranged in the order of increasing Es/No requirement for increasing TB payload. The MCS index points to a unique selection in this set. The terminal is proposing the highest MCS, which meets the BLER target in reference to the most recently measured pilot channel. The statistics of the proposed MCS (on the left) shows that in these channel conditions, the terminal was proposing more often lower MCS. With higher power allocation and better average channel conditions, the probability of proposing higher MCS could of course increase. The statistics of the selected MCS (on the right) was even more weighting lower MCS, either because there was not enough data available in the buffer at the moment of transmission or because for the window-limited MAC-hs in Figure 7 respectively. The delay distributions show that when T1 time-out value was increased, the MAC-hs window memory requirement gets higher and the packet delays are allowed to increase. For a given T1 value, the MAC-hs window size does not largely depend on the given MCS set. However, buffering larger TBs will take more memory both in the transmitter and in the receiver. The packet delay distributions do not show big differences for different T1 settings. For a given MCS with 5 multicodes, the probability of getting higher packet delay is higher than for the MCS with 10 or 15 multicodes, which have no large mutual difference. Figure 3. Probability of a given MCS proposed by the terminal (left) and the MCS selected by the base station (right) with different MCS set available up to (a and b) 5, (c and d) 10 and (e and f) 15 multicodes respectively. Figure 4. The cumulative distribution of the MAC-hs window size occupied by the number of TBs for different Timer T1 setting and MCS set. a retransmission occurred3 . With a more strictly limited MCS, the probability of selecting a high MCS in the set is larger in comparable channel conditions, as shown in Figure 3 (a) and (b) for up to 5 multicodes. The MCS distributions in Figure 3 (e) and (f) present the ultimate case of MCS selection up to 15 multicodes, which makes the probability of selecting a high MCS in the set lower. Figure 4 shows the result of the MAC-hs window occupancy in number of TBs as a function of T1 setting. Figure 5 shows the result of the MAC-hs window occupancy in number of TBs as a function of limited window size. The MCS set was limited to maximum of 5, 10 and 15 multicodes respectively. In both cases, the TBs in the MAC-hs window may be of different size, because of the MCS selection algorithm. The packet delay distributions are shown for the Timer T1 limited MAC-hs in Figure 6 and 3 The MCS was not allowed to change during the retransmissions in these simulations. Figure 5. The cumulative distribution of the MAC-hs window size occupied by the number of TBs for different MAC-hs window size limit and MCS set. Next, the MAC-hs window size was limited to 4, 16 and 32 respectively. With a small window, the memory requirements are low and the forced packet delays are small. Here, a high MCS set with 15 multicodes makes a difference in packet delay, as larger TBs can be formed larger and not that strictly limiting L1 retransmissions, the packet delay distributions are not drastically different for any of the MCS sets. 5 Conclusions Figure 6. The cumulative distribution of packet delay for different Timer T1 setting and MCS set. Performance of the High Speed Downlink Packet Access protocol was analysed in the WCDMA air interface with the multicode transmission, N-channel FHARQ and AMC. The MAC-hs parameters have an impact to the performance by allowing a high physical channel BLER operation point e.g. of order 30-50% and still making the RLC protocol BLER very low e.g. of order 5%. The results are shown for selected channel conditions, packet traffic model and Iub delay distributions as typical conditions and the results were analysed as a function of the HSDPA MAC protocol parameters. The packet delay was observed to depend on the Transport Format set available and on the MAC-hs parameter settings as window timers and window limits. References Figure 7. The cumulative distribution of packet delay for different MAC-hs window size limit and MCS set. during good channel conditions. With the window limit set to 16, the window was rarely full and having higher MCS yields lower packet delays. With the window limit set to 32, the window was not observed to be full and it would already show the maximum requirement for the window memory in the terminal. With a lower MCS set with 5 multicodes, higher window size is requested in terms of number of TBs in the window. However, the size distribution of the TBs in the window may be smaller here than in the case of 10 or 15 multicodes, which allow larger TBs, if the channel conditions are very good. In T1 limited case (see Figure 6), the probability of having larger packet delays is larger for MCS up to 5 multicodes, but the mutual difference for MCS with 10 and 15 multicodes is very small. In the window-limited case (see Figure 7), where the window is set to the smallest value (four), the MCS with 15 multicodes gains in lower packet delay compared to MCS with 5 and 10 multicodes. When the window is [1] Ojanperä T., Prasad R., An overview of Air Interface Multiple Access for IMT-2000/UMTS., IEEE Communications Magazine, pp. 82-95, Sept. 1998. [2] Dahlman E., Gudmundson B., Nilsson M., Sköld J., UMTS/IMT-2000 Based on Wideband CDMA., IEEE Communications Magazine, pp. 70-80, Sept. 1998. [3] Adachi F., Sawahashi M., Suda H., Wideband DSCDMA for Next-Generation Mobile Communications Systems.,IEEE Communications Magazine, pp. 56-69, Sept. 1998. [4] Viterbi, Andrew J., CDMA Principle of Spread Spectrum Communication, Addison-Wesley Wireless Communication, 1995. [5] Holma Harri, Toskala Antti et al., WCDMA for UMTS Radio Access for Third Generation Mobile Communications, Second Edition, pp. 279-304, John Wiley & Sons, 2002. [6] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Physical channels and mapping of transport channels onto physical channels (FDD) release 5, TS 25.211 V5.3.0 (200212), pp. 8, 20-21, 40-42. [7] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Physical layer aspects of UTRA High Speed Downlink Packet Access (Release 4), TR 25.848 V4.0.0 (2001-03). [8] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network;; High Speed Downlink Packet Access (HSDPA) Overall Description; Stage 2, TS 25.308 V5.3.0 (2002-12). [9] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network;Services provided by the physical layer (Release 5), TS 25.302 V5.3.0 (2002-12). [10] Mogensen, P., High speed downlink packet access (HSDPA)-the path towards 3.5G, IEEE Workshop on Signal Processing Systems 2001, p. 344, Antwerp Belgium, Sep. 26-28 2001. [11] Moulsley, T.J., Throughput of high speed downlink packet access for UMTS, Second International Conference on 3G Mobile Communication Technologies 2001, pp. 363-367, London UK, March 2001. [12] Love, R., Ghosh, A., Nikides, R., Jalloul, L., Cudak, M., Classon, B., High speed downlink packet access performance, IEEE Vehicular Technology Conference (VTC) Spring 2001, vol.3. pp. 2234-2238, Rhodes Greece, May 6-9 2001. [13] Kolding, T.E., Frederiksen, F., Mogensen, P.E., Performance aspects of wcdma systems with high speed downlink packet access (hsdpa), IEEE Vehicular Technology Conference (VTC) Fall 2002, Vol.1 pp. 477481, Vancouver (BC) Canada, Sep. 24-28 2002. [14] Manuel, G.; Rinne, M.,Analysis of Transmission Window for the delay Performance of High Speed Downlink Packet Access, submitted to IEEE Conference on Software, Telecommunications and Computer Networks (SoftCOM), Dubrovnik, Venice, Ancona, Oct. 7-10, 2003. [15] Aridhi, S., Despins, C.L, Performance analysis of type-I and type-II hybrid ARQ protocols using concatenated codes in a DS-CDMA Rayleigh fading channel, Canadian Conference on Electrical and Computer Engineering 1995, vol.2. pp. 648-652, Sep. 5-8. 1995. [16] Cam, R.; Leung, C., Throughput analysis of some ARQ protocols in the presence of feedback errors, IEEE Transactions on Communications, vol. 45 Issue: 1 pp.35-44, Jan. 1997. [17] Cam, R.; Leung, C., Multiplexed ARQ for timevarying channels. I. System model and throughput analysis, IEEE Transactions on Communications, vol 46 Issue: 1 pp. 41-51, Jan. 1998. [18] Hamorsky, J., Hanzo, L., Performance of the turbo hybrid automatic repeat request system type II, Information Theory and Networking Workshop 1999, p. 51, Metsovo Greece, Jun.27-Jul.1 1999 . [19] Tingfang Ji; Stark, W.E., Turbo-coded ARQ schemes for DS-CDMA data networks over fading and shadowing channels: throughput, delay, and energy efficiency, IEEE Journal on Selected Areas in Communications, Vol. 18 Issue:8, pp. 1355-1364 , Aug. 2000. [20] Banerjee, A., Costello, D.J., Jr, Fuja, T.E., Comparison of different retransmission strategies for bandwidth efficient hybrid ARQ schemes using turbo codes, IEEE International Conference on Personal Wireless Communications (PWC) 2000, pp. 548 -552, Gdansk Poland, Sep. 14-15 2000. [21] Malkamäki, E., Mathew, D., Hämäläinen, S., Performance of hybrid ARQ techniques for WCDMA high data rates, Vehicular Technology Conference (VTC) Spring 2001, Vol.4. pp.2720-2724, Rhodes Greece May 6-9 2001. [22] Das, A., Khan, F., Sampath, A., Hsuan-Jung Su, Performance of hybrid ARQ for high speed downlink packet access in UMTS, IEEE Vehicular Technology Conference (VTC) Fall 2001, vol.4. pp. 2133-2137, Atlantic City (NJ) USA, October 7-11 2001. [23] Frederiksen, F., Kolding, T.E.,Performance and modeling of wcdma/hsdpa transmission/h-arq schemes, IEEE Vehicular Technology Conference (VTC) Fall 2002, Vol.1 pp.472-476, Vancouver (BC) Canada, Sep. 24-28 2002. [24] Frenger, P.; Parkvall, S.; Dahlman, E., Performance comparison of HARQ with Chase combining and incremental redundancy for HSDPA, IEEE Vehicular Technology Conference, (VTC) Fall 2001, vol.3 pp 18291833, Atlantic City (NJ) USA, October 7-11 2001. [25] Love, R., Classon, B., Ghosh, A., Cudak, M., Incremental redundancy for evolutions of 3G CDMA systems, IEEE Vehicular Technology Conference (VTC) Spring 2002, vol.1. pp.454-458 Birmingham (AL) USA, May. 6-9 2002. [26] Ghosh, A., Stewart, K., Ralasuk, R., Buckley, E., Bachu, R., Incremental redundancy (IR) schemes for W-CDMA HS-DSCH, IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) 2002, vol.3 pp. 1078-1082, Lisboa Portugal, Sep. 15-18 2002. [27] Nakamura, M., Awad, Y., Vadgama, S.,Adaptive control of link adaptation for high speed downlink packet access (HSDPA) W-CDMA, The 5th International Symposium on Wireless Personal Multimedia Communications (WPMC) 2002, Vol.2. pp: 382 -386, Honolulu Hawaii, Oct. 27-30 2002. [28] Kwan, R., Chong, P., Rinne, M.,Analysis of the Adaptive Modulation and Coding Algorithm with the Multicode Transmission, IEEE Vehicular Technology Conference (VTC) fall 2002, Vancouver Canada, Sep. 2428 2002. [29] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Radio Link Control (RLC) Protocol Specification (Release 5), TS 25.322 V5.3.0 (2002-12), pp. 13-16, 25-26, 47-49. [30] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network;Medium Access Control (MAC) Protocol Specification (Release 5), TS 25.321 V5.3.0 (2002-12), pp. 12-13, 16-17, 26, 31, 32, 47-51. [31] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network;Radio Resource Control (RRC) Protocol Specification (Release 5), TS 25.331 V5.4.0 (2003-03), pp. 107, 112-117. [32] UMTS 30.03, Selection procedures for the choice of radio transmission technologies of the UMTS., ETSI TR101 112 (1997-11). [33] Jakes, W.,Microwave Mobile Communications, Wiley & Sons, 1974.