System Performance Gain by Interference Cancellation in WCDMA Dedicated and High-Speed Downlink Channels

Transcription

1 System Performance Gain by Interference Cancellation in WCDMA Dedicated and High-Speed Downlink Channels Hans D. Schotten Research Mobile Communications Ericsson Eurolab Germany Neumeyerstr. 5, 94 Nuremberg, Germany Phone: , Fax: Jürgen F. Rößler Chair of Information Transmission University Erlangen-Nuremberg Cauerstrasse 7/NT, D-958 Erlangen, Germany Phone: , Fax: Abstract A large number of the existing and planned G networks rely on the GPP WCDMA standard. Since the downlink of WCDMA systems is usually interference limited, interference cancellation algorithms that reduce the interference on the desired signal can significantly improve the system capacity. In this paper, the system-level performance gain that can be achieved by interference cancellation on the terminal side is investigated. Two downlink channels are considered: the dedicated downlink channel, which is already part of the WCDMA standard, and the new high-speed downlink channel for packet-data support, which will be part of Release 5 of the WCDMA standard. In the analysis, four receiver classes are compared: the standard Rake receiver, a Rake receiver with pilot subtraction, a linear MMSE equalizer, and a non-linear iterative successive cancellation algorithm. The resulting performance gains and the necessary implementation complexity are discussed. I. INTRODUCTION This year, a large number of commercial G systems will start operation. Many of them will rely on the WCDMA standard as specified by the rd Generation Partnership Program GPP. Since the downlink of these systems is usually interference limited, terminal receivers that are able to reduce the effective interference on the desired signal can significantly enhance the system performance. In this paper, potential interference mitigation algorithms are investigated for two downlink channels: the dedicated downlink channel introduced in WCDMA Release 99 and the new high-speed downlink channel that is introduced into the WCDMA standard as part of Release 5. II. WCDMA DEDICATED AND HIGH-SPEED DOWNLINK CHANNELS The dedicated downlink channels are code-multiplexed using orthogonal channelization codes of variable spreading factor (OVSF-codes). A fast SIR based power control operating on slot level (5 Hz) is used for link adaptation. In principle, the power control guarantees that the dedicated channel is always assigned the transmit power necessary to achieve a certain demodulation performance at the terminal receiver. On the other hand, this link adaptation scheme shall avoid that more transmit power than necessary is allocated to a terminal since the overall transmit power of the base station is a capacity limiting quantity. Interference cancellation in the terminal will reduce the experienced interference and, via the power control, reduce the transmit power consumption for this terminal. So, interference cancellation will finally result in a larger system capacity. The new WCDMA high-speed downlink channel (High- Speed Downlink Shared CHannel, HS-DSCH) has been optimized to enhance the support of high-speed packet-data transmissions by increasing the supported peak data-rates up to rates exceeding 8Mbit/s and by significantly reducing the round-trip delays. So, the HS-DSCH extends the WCDMA packet-data capabilities far beyond the IMT-2 requirements. The technologies on which this new channel relies comprise fast link adaptation by means of adaptive modulation and coding, fast hybrid ARQ and fast scheduling. The HS-DSCH uses adaptive modulation and coding for fast link adaptation. A terminal experiencing good link conditions will be served with a higher data-rate than a terminal in a less favorable situation. To support the different data-rates, a wide range of channel coding rates and different modulation formats (QPSK and 6QAM) are supported. Fast power control is not used for the HS-DSCH. A fast hybrid ARQ scheme is introduced for the HS-DSCH. It is implemented on Node-B level and will enhance the packetdata characteristics by reducing the round-trip delay. The fast scheduling algorithm assigns the high rate HS- DSCH to one or a small group of users per time slot depending on the link performance they are instantaneously experiencing. This mechanism guarantees that no transmit resources are allocated for terminals in fading dips, but that the terminals experiencing good channel conditions, which can thus be served with high data-rates, are assigned the HS-DSCH with priority. The new high-speed channel - similar to the current Downlink Shared Channel - shares a part of the channelization code resource primarily in the time-domain among several users. The spreading factor of the assigned channelization codes is fixed to 6. Up to 5 out of 6 orthogonal codes can be allocated for the HS-DSCH. In order to allow these fast algorithms to work fast, the

2 Transmission Time Interval (TTI) - the time interval for which the modulation, coding and spreading format is fixed - is reduced to 2 ms. A more detailed description of the new WCDMA high-speed channel can be found in the GPP technical reports (e.g., TR25.95, TR25.848, and TR25.855) and in [],[2]. Applying interference cancellation in terminals receiving the HS-DSCH will allow them to receive transmissions with higher data-rates. This will increase the system throughput and, depending on the scheduler strategy, enhance the end-to-end performance. So, the benefit of advanced terminals can be seen from a system as well as an user point of view. III. INTERFERENCE CANCELLATION STRATEGIES FOR WCDMA TERMINALS Four receiver structures for WCDMA terminals have been investigated: The standard Rake receiver, which will be used as reference. A Rake receiver with a simple subtraction algorithm that subtracts known channels (pilot channel, synchronization channel) received from the connected base station and from a limited number of interfering base stations (e.g., the base stations in soft handover). This approach results in low additional receiver complexity since much information on these channels is already present at the terminal. A linear MMSE equalizer on symbol level with and without subtraction of known channels. Here, not only the inter-path interference of the signal of the connected base stations but also the noise covariance matrix of the signals of interfering base stations is considered. The used MMSE equalizer on symbol level is derived according to [], but with a sliding window approach to cope with inter-symbol interference. An iterative soft decision interference cancellation (IS- DIC) algorithm with a matched filter (MF) front-end. This receiver implementation represents a high-end solution. A detailed description of the MF ISDIC can be found in [4] where it is derived for synchronous CDMA and BPSK with random spreading sequences. To calculate soft decisions at the MF ISDIC for higher-order modulation a soft decision scheme corresponding to [5] is used. The utilized ISDIC operates as follows: All symbols of a certain block are successively estimated with a matched filter from the received signal whereby interference is cancelled using the latest soft estimates of interfering symbols. From a matched filter output a new soft estimate is calculated which is used to cancel interference when a new symbol of the block is estimated. Having calculated new estimates for all symbols the channel decoding unit is exploited to calculate a posteriori probabilities for all encoded bits which are then used to calculate new soft estimates for all symbols. The described iteration of the algorithm is repeated until the algorithm converges. In all cases, the receiver structures have been adapted to the characteristics of the investigated channel. To simplify the analysis, optimistic assumptions concerning channel estimation and other necessary measurements had to be used. Thus, the results presented in this paper shall only be used as an indicator describing the principal impact of these algorithms. Implemented solutions will probably suffer from measurement and estimation errors. IV. DEFINITION OF INTERFERENCE SCENARIOS AND LINK-LEVEL RESULTS Since the interference cancellation capability of the above described algorithms critically depends on the structure of the experienced interference, i.e., the characteristics of the dominant interference sources, the link-level simulations on which the system analysis is based have to reflect the same interference scenarios as modeled in the system-level simulation. For this reason, based on the analysis of a cellular layout, representative interference scenarios have been identified. These scenarios are characterized by the set of ratios of the powers of the strongest received interfering base station signals with respect to the received power of the signal of the connected base station. In order to illustrate this approach and the impact of the interference scenario, some examples for link-level results are presented in this section. Figure presents a typical cellular layout with omnidirectional antennas. Let us assume that the considered terminal is located on the black arrow from Base Station (BS ) to Base Station 2 (BS 2). The terminal is always connected to BS. In order to characterize the interference experienced by the terminal, we define the so-called geometry : where Îor is the power of the signal received from BS and is the power of the signals received from all other base stations. In Figure, the geometrical cell border (hexagonal layout) and the contour of the geometry G is plotted. In our link-level simulations, in addition to the signal from BS, the signals from BS 2, BS, and BS 4 are modeled explicitly, i.e., these signals and the corresponding fading channels are generated chip-true. The powers of these interfering signals are characterized with respect to the power of the desired signal from BS. Let I BS2, I BS, and I BS4 be the received powers of the signals from BS2, BS, and BS4, respectively. I AWGN is the received power of all other base stations that are modeled by AWGN in our link-level simulations. With these definitions, we get = I BS2 + I BS + I BS4 + I AWGN and =

3 ( Îor I BS2 ) + ( Îor I BS ) + ( Îor I BS4 ) + ( I AW GN ). For four interference scenarios, the sets of characterizing power ratios are summarized in Table. They are calculated for a propagation loss exponent of.45 and equal transmit power for all base stations. For the cellular layout depicted in Figure, the scenarios in Table describe terminal positions on the black arrow. A geometry of 2 db corresponds to a position of the terminal close to the connected base station. Here, the terminal will experience good channel conditions. Intracell interference due to multipath propagation will often be the dominant interference part. A geometry of 9 db corresponds to a more typical position where depending on the propagation conditions, the intercell interference can already be dominant. With a geometry of db and below we are usually entering the soft handover region close to the cell border. As an example for the impact of the interference scenario on the achievable link performance, Figure 2 and present simulation results for a simplified HS-DSCH scenario. Here, orthogonal codes of spreading factor 6 are used for the HS- DSCH. These codes are assigned 85% of the total base station output power. 5% of the base station output power are assigned to other channels including the pilot channel. Turbocoding is assumed with rate R c = /2 in Figure 2 and rate R c = / in Figure. A Vehicular A channel model is used BS 7 BS 4 BS 5 BS 2 6 A - BS 2 geometrical cell border - BS 6 BS Fig.. Multi cell scenario In Figure 2 and, Rake denotes the standard Rake receiver, Rake pc denotes the Rake receiver with pilot cancellocation location 2 location location 4 2 db 9 db db db I BS2 26, 9 db, 9 db 6, db, db I BS 28, 2 db 7, 2 db, 8 db 7, 6 db I BS4 28, 2 db 7, 2 db, 8 db 7, 6 db I AW GN 2, db, 2 db 8, 5 db 4, 5 db TABLE I TYPICAL POWER RATIOS AT MULTI CELL SCENARIO lation (other known channels are subtracted as well), MMSE denotes an MMSE equalizer with additional pilot cancellation, and MF ISDIC denotes the above described ISDIC receiver. Single user throughput simulation results in Mbit/s are presented for 4QAM (solid lines) and 6QAM (dashed lines). As can be seen from these figures, significant enhancements of the throughput can be achieved by advanced receiver techniques. For the Rake receiver, the use of 6QAM often results in a lower throughput than the use of 4QAM (even with different coding rates). For more advanced receiver algorithms and for good channel conditions, a benefit from the use of 6QAM can be observed. In general, the ISDIC offers the best results and in some cases (under the idealized assumptions used in these simulations) comes close to the optimum. However, a good compromise between complexity and performance is the MMSE equalizer that shows only small degradation compared to the ISDIC for medium and high geometry values. Note that these losses are larger for channel models with less multipath interference. V. SYSTEM-LEVEL ANALYSIS METHODOLOGY For the system-level analysis, a certain number of the above described interference scenarios have been identified to represent the different interference scenarios a terminal will typically experience in a WCDMA network. These scenarios are characterized by the relative received powers of the four strongest base stations as defined in the last section. Based on the above described multi-cell simulation set-up, link-level simulations are performed. For each scenario and each channel model, a parameter set (orthogonality factor, cancellation efficiency for each interfering signal, noise enhancement due to mismatched filtering) is derived that describes the link-level performance and allows a simple and efficient calculation of an equivalent SNR at the decoder input in the system-level simulation. Based on this equivalent SNR, the link performance in terms of the block error rate can be calculated. Using this pre-calculated characterization of the link performance, the system-level simulation is carried out. Here, for

4 Throughput [Mbit/s] Throughput [Mbit/s] Rake 4QAM Rake 6QAM Rake pc 4QAM Rake pc 6QAM MMSE 4QAM MMSE 6QAM MF ISDIC 4QAM MF ISDIC 6QAM Upper bound for 4QAM Upper bound for 6 QAM 2 9 < G [db] Fig. 2. Throughput at multi cell scenario for R c = /2 2 9 < G [db] Rake 4QAM Rake 6QAM Rake pc 4QAM Rake pc 6QAM MMSE 4QAM MMSE 6QAM MF ISDIC 4QAM MF ISDIC 6QAM Upper bound for 4QAM Upper bound for 6 QAM Fig.. Throughput at multi cell scenario for R c = / each terminal the received power levels of the four strongest received base stations are calculated and the closest of the pre-defined interference scenarios is determined. The parameters of this scenario are used to calculate the performance of the terminal. VI. RESULTS AND COMPARISON In this section, system-level results for a simple hexagonal cell layout with omnidirectional antennas are presented. The propagation loss factor is.45, the channel profiles are the Indoor A and the Vehicular A channel. In the system-level analysis typical radio network details are modeled. In order to avoid that the investigated impact of the receiver structures is masked by other performance limitations, only simplified traffic, radio resource control and mobility models are used. We assume that all cells are fully loaded and that 5% of the transmit power is allocated to orthogonal common channels including the pilot channel. The synchronization channel is not considered. For the dedicated channels, the system performance is measured in terms of the number of users which can be simultaneously supported. The number of users is step by step increased (with a simple admission control algorithm) until the maximum load is found. Hereafter, the capacity gains that can be achieved for the dedicated channels are compared to the results for the Rake receiver. With the subtraction algorithm, capacity gains of 2 5% were found. They depend on the channel profile (2.5% for the Vehicular A channel, 5% for the Indoor A channel) and of the number of signals subtracted. In our simulations, the known channels of the connected base stations, i.e., base stations in soft-handover, are always cancelled. The signals of the other base stations are cancelled if their received power does not fall below -6dB of the received power of the strongest connected base station for a period of ms. The cancellation accuracy was determined in the above described link-level simulations. For the linear MMSE filter approach, capacity gains of 8 % for indoor channels and of 7 2% for vehicular channels are found. This difference can be explained by the larger relative impact of the interpath interference in case of the vehicular channel. If the subtraction algorithm is applied in addition to the MMSE filtering, we get a capacity gain of 2 4.5% for the indoor channel and 2 2% for the vehicular channel. Obviously, the gains for the MMSE filtering and the subtraction algorithm can roughly be added. For the non-linear ISDIC, capacity gains of 4% for indoor channels and of 4 55% for vehicular channels can be observed. Here, especially the set of intercell channels that are included in the successive cancellation scheme determines the gain. Thus, the capacity gain which can be achieved can - to a large extent - be controlled by the additional computational efforts spent at the receiver. Compared to these results, three major differences have been observed for high-speed channels: For the HS-DSCH, the gain on system-level has to be described in terms of the system throughput since not all users are served with the same average data-rate. The achievable throughput gains critically depend on the implemented scheduler strategy. If terminals experiencing good link conditions are served with priority (e.g., max C/I scheduling), receiver structures which reduce the intracell interference due to multipath propagation show the largest gain since these terminals are often located closely to the connected base station

5 where multipath interference is dominant. If more balanced scheduler strategies are used or multipath interference is less critical, the other options show system throughput gains similar to the capacity gains found for dedicated channels. Note that for realistic traffic models the gain achieved by max C/I scheduling compared to a simple round robin scheduling approach is often reduced. For this case, a throughput gain of up to 5% is found for the vehicular channel. The additional gain achievable by pilot subtraction is slightly smaller (.5-.5% throughput gain) for the HS-DSCH scenario. For HS-DSCH, the ISDIC can work more efficiently since the HS-DSCH is usually allocated much more power than dedicated channels and more information on the structure of the HS-DSCH is known. This allows a more efficient and reliable demodulation of channels transmitted in neighbored cells and a higher accuracy of their cancellation. For this reason, this approach results in significantly larger gains for HS-DSCH than found for dedicated channels. System throughput gains of up to 6% for the indoor channel and up to 9% for the vehicular channel are found for the above described ISDIC receiver. REFERENCES [] Stefan Parkvall, Erik Dahlman, Pal Frenger, Ber Beming, Magnus Persson, The Evolution of WCDMA towards higher speed downlink packet data access, in Proc. of VTC 2 Spring, Rhodes, May 2. [2] Hans Schotten, Evolution of G radio access techniques, in Proc. of the International Symposium G Infrastructure and Services GIS, Athens, pages 6 65, 2. [] Sergio Verdu, Multiuser Detection, Cambridge University Press, first edition, 998. [4] Ralf R. Müller and Johannes B. Huber, Iterated soft decision interference cancellation for CDMA, in Digital Wireless Communications. 998, Springer. [5] Christian Sgraja, Werner G. Teich, Achim Engelhart, and Jürgen Lindner, Multiuser/multisubchannel detection based on recurrent neural network structures for linear modulation schemes with general complex valued symbol alphabet, Technical Report ITUU-TR-2/, COST Workshop 262, Ulm, Jan. 2. [6] J. F. Rößler, W. H. Gerstacker, A. Lampe, and J. B. Huber, Matchedfilter- and MMSE-based iterative equalization with soft feedback for QPSK transmission, in Proc. of the 22 Zurich Seminar (IZS 2), 22. VII. CONCLUSION In this paper, the system performance gain of a WCDMA network achievable by the introduction of advanced terminal receivers is investigated. WCDMA dedicated channels and the new WCDMA high-speed channels are considered. In order to evaluate the impact of advanced receivers that take the structure and colorness of interfering signals from other base stations into account, a new evaluation methodology has to be introduced. A possible approach is described in this paper. Both, the dedicated channels and the new high-speed channels will benefit from a future introduction of advanced terminal receivers. For the dedicated channels we will mainly see a system capacity gain, whereas for the HS-DSCH, a gain can be seen from the system as well as the user perspective. Significant differences in the achievable gain have been found depending on the channel model. Additional factors affecting the gains are the network topology, the distribution of the users in the cell and, for HS-DSCH scenarios, the scheduler policy. It will probably be difficult to decide which receiver algorithm is superior in general. The optimum choice (performance per costs) of a receiver algorithm will depend on the identified bottleneck scenarios characterized in terms of supported data-rate, position in the network etc. In principle, the presented linear MMSE filter approach in combination with a simple subtraction algorithms seems to be a reasonable compromise between complexity and performance. Based on single-user throughput results in can be concluded that the support of 6QAM will no necessarily result in a benefit for simple Rake receiver terminals. A clear gain by the introduction of 6QAM for HS-DSCH can only be found if advanced receivers are introduced.