A Flexible Air Interface for Integrated Broadband Mobile Systems

Transcription

1 A Flexible Air Interface for Integrated Broadband Mobile Systems André Noll Barreto*, Michael Mecking and Gerhard Fettweis* *Dresden University of Technology Mobile Communications Systems, 01062, Dresden, Germany, Tel: / Fax: {noll,fettweis}@ifn.et.tu-dresden.de Munich University of Technology Arcisstr. 21, D Muenchen, Germany, Tel: / Fax: Michael.Mecking@ei.tum.de Abstract - In this paper we present a flexible air interface for an Integrated Broadband Mobile System, which utilises adaptive antenna arrays in three configurations. These correspond to different classes of users which support different data rates and Quality of Service requirements. Due to the complexity of smart antenna algorithms, a trade-off between data rate and mobility is required. The modulation and coding parameters of the various classes are specified and simulation results are shown for a space-time channel, considering a pre-rake and a space-time Rake receiver. 1 - Introduction First and second generation mobile communication systems have been successfully deployed worldwide, but new developments are necessary in order to satisfy the new requirements of higher capacity, higher data rates and service flexibility. One step in this direction has been achieved by the 3rd generation systems currently at the final stages of the standardisation process [1]. However, there are still limitations in terms of supported data rates and services integration. The Integrated Broadband Mobile System (IBMS) project is part of the German wireless ATM effort ATMmobil [2] and targets the integration of different mobile communications systems by means of a common signalling channel [3] and the development of several air interfaces which support high data rates of up to 34Mbits/s with particular emphasis on service flexibility. The following concepts are being developed under this project: an infrared system, an indoor 24GHz system and an outdoor 5GHz system. This paper is concerned with the description of the air interface for the latter [4], which is intended for high capacity microcell coverage. The air interface is based on the use of antenna arrays in order to provide high system capacity, good performance at high data rates and service flexibility. The system will be described in Section 2, with particular emphasis on a flexible coding scheme in order to support the required Quality of Service (QoS), and simulation results will be shown in Section System description The outdoor IBMS system relies on the use of smart antennas, which enables an increase in capacity, as well as an improvement in the link performance due to the increase in the signal-to-noise ratio and to the reduction of intersymbol interference. However, smart antennas are expected to be employed only for slow moving mobiles, due to the complexity of antenna tracking algorithms. Thus, a trade-off between data rate and mobility is required. Three different Traffic Channel Classes (TCC) with different antenna configurations and different QoS parameters (bit error rate, frame error rate, mobility, etc.) have been established. In TCC-A, only omnidirectional antennas are employed and a data rate of 64kbits/s is envisaged. In TCC-B a linear antenna array with 8 elements is employed at the base station to support data rates of up to 2Mbits/s. In TCC-C antenna arrays are to be employed in both mobile (4 elements) and base stations (8 elements), supporting up to 32 Mbits/s. A common signalling channel, the NACCH (Network Access and Connectivity Channel), is embedded in all the Traffic Channel Classes. The IBMS concept is depicted in Fig. 1. The system employs TDD as duplexing method, which is more appropriate for use with smart antennas, since, unlike FDD, the channel impulse response is almost the same at both transmission directions, provided the channel coherence time is large compared to the TDD frame period. This fact also allows the deployment of preprocessing techniques, such as the pre-rake [4]. The IBMS outdoor system has a TDD-frame of 1.5ms, with 10µs guard interval between uplink and downlink phases. These parameters were chosen with microcell coverage in mind, with cell radius of up to 300m and velocities of up to 80km/h. Data Rate 32 Mbits/s 2 Mbits/s 64 kbits/s TCC C NACCH TCC B TCC A movable portable high Mobility Fig. 1 - Trade-off between data rate and mobility BS

2 2.1 Air Interface For each Traffic Channel Class (TCC) different sets of physical layer parameters were defined, respective to the different requirements and configurations. This was done with the purpose of enabling a flexible system, in which seamless transitions between different classes are feasible. Hence, a single modem concept for all the different classes has been designed. This was achieved in having a single symbol rate of 18,1622 Mbauds (symbol period T s =55,1ns) with roll-off factor α=0,3 for every traffic class in the available channel bandwidth of 25 MHz. Using this concept, single carrier modulation could be used also for high data rates in a broadband wireless environment, instead of multi-carrier modulation schemes such as OFDM. This is made possible through the deployment of smart antennas, and the consequent reduction in the intersymbol interference, which mitigates or even eliminates the need for a complex equaliser. Considering the data rate requirements and available bandwidth, for the lower rate classes (TCC-A and -B) we employ direct sequence spread spectrum with OQPSK modulation, whereas in TCC-C 32-QAM coded modulation is employed. The multiple access methods are different for the various classes and related to the basic physical layer parameters, i.e., spreading factor and antenna configuration. Thus, in TCC-A CDMA is used and in TCC-B both CDMA and SDMA, due to the use of smart antennas. In TCC-C we have basically SDMA, but also TDMA in cases where the smart antennas alone cannot separate the signals from different users. TCC Data Rate Proc. Gain (symbols/bit) Modulation A 64 kbits/s 128 OQPSK B 2 Mbits/s 4 OQPSK C 34 Mbits/s - 32 QAM Some important physical layer parameters are listed in Table 1, and the single modem concept is shown in Fig. 2. The rates of the repetition codes in Fig. 2 do not correspond to the whole processing gain, since part of it is due to a convolutional or turbo code, as explained in Section 2.2. TCC-A/B coded bits CLASS SELECTOR TCC-C QAM symbols Table 1: Physical Layer Parameters Repeat 32x (TCC A) 4x (TCC B) c n QPSK mapper T s /2 Fig. 2 - Single Modem Concept H(f) cos(2πf c t) H(f) 90 Σ Each TCC has its own coding scheme, which may also vary according to the application. These schemes are described in detail in the following sub-section. 2.2 Channel Coding TCC - A This class was designed with applications like voice or low rate packet transmission in mind requiring low complexity and a modest BER performance or very reliable error detection. The maximum data rate in TCC- AisR b =64 kbits/s which corresponds to a transmission rate of R Tx =1/128 bits/channel use. The low spectral efficiency required allows the decoder to work reliably under severe noise and multi-access interference. Typically, the low transmission rate is achieved by concatenating a medium rate error correcting code with a direct-sequence spreading code, neglecting however the fact that spreading is equivalent to a repetition code which achieves no coding gains or increased diversity over time-invariant channels. The spreading is usually assumed to combat multi-access interference and enable synchronization while the errorcorrecting code is to handle the remaining noise. We propose the use of a low rate error correcting code to combat both multi-access interference and noise. As a compromise between coding and spreading, which still serves synchronization issues, we employ a binary maximum free distance convolutional code of rate R c =1/8 and 256 states (memory 8) followed by block interleaving, direct sequence spreading with processing gain 32, and OQPSK modulation. Decoding is performed via the Viterbi algorithm which is maximum likelihood and insensitive to scaling of input metrics provided by a Rake receiver. Error events of the decoder correspond to detours in the code trellis and give rise to burst errors in the decoded information sequence. In case of low rate packet transmission, a single bit error in the packet triggers a retransmission via an ARQ protocol, or the packet is discarded. Thus, outer algebraic block codes over higher order symbol alphabets are used to correct resilient bursty errors after Viterbi decoding and to provide packet integrity. Reed-Solomon codes of different dimension and block length over the Galois field GF(256) (8 bits = 1 symbol) as outer code have been implemented. These codes are maximum distance separable (MDS) in the sense that each two parity symbols allow the correction of 1 additional symbol error in the codeword, and this is the minimum required (singleton bound) [5]. Bounded distance decoding is applied, implemented via the Berlekamp- Massey algorithm. Thus, if a viable codeword is found within a designed distance (less than half of the minimal distance of the code) to the received codeword, the decision is made in favour of that word otherwise an erasure is declared. Hence, there is a trade-off between correctable errors and undetected codeword errors

3 resulting in erroneous packets. Table 2 depicts the parameters for the Reed-Solomon codes in different modes that support a trade-off between delay (packet length) and error correction capabilities. E.g., with the given transmission rate a 48 byte packet has to be transmitted on 4 slots resulting in a minimum transmission delay of 5.25 ms. mode A1 A2 A3 packet size 48 bytes 96 bytes 192 bytes block length 54 bytes 108 bytes 216 bytes correctable errors prob. of undetected error 5 1 if e> if e> if e>12 Table 2: Reed Solomon code parameters for TCC-A (e denotes the given number of symbol errors in a codeword) The very low probability of undetected errors ensures that (hardly) no erroneous packets are to be handed over to higher layers, and no further parity check codes are necessary. TCC - B This class was designed for low BER applications or packet data. In contrast to TCC-A, the maximum data rate transmittable over the channel is R b = 2 Mbits/s. This increase is mainly achieved with the application of smart antennas at the base station, which improves the channel characteristics and reduces delay spread as well as multiple access interference. To fully exploit capabilities of multi-receiving antennas at the base station accurate channel measurements of delay and directions of arrival are crucial. Therefore, TCC-B supports users confined to walking speeds in order to prevent non-trackable channel fluctuations. The given data rate of 2 Mbits/s is equivalent to a transmission rate of R Tx =1/4 bit/channel use, revealing no necessity for coded modulation schemes. In fact, an option for the implementation of this transmission rate is a re-use of the coding scheme developed for TCC-A without the additional spreading but merely by scrambling the output. Thus, coding combats both remaining multi-access interference after spatial filtering and noise. Due to higher data rates on the channel the transmission delay decreases and the longest packet of 192 bytes fits twice into a single slot of 0.75 ms. Additionally, to further improve spectral and power efficiency we devised another coding scheme based on parallel concatenated convolutional codes (PCCC, Turbo codes) which belong to the best known codes and operate close to the capacity limits for sufficiently long information block lengths. The designed PCCC consists of two 8-state systematic recursive convolutional codes (generator polynomial g(d) = 17/15 in octal) separated by a pseudo-random interleaver constructed with linear congruential sequences [5]. For low BER applications, a block of 384 bytes is encoded into 6272 bits by alternately puncturing the parity bits of the constituent codes. Then the codeword is block interleaved, spread with a modest processing gain 4, OQPSK modulated, and transmitted on a single slot of 0.75 ms. Via iterative decoding algorithms a BER of 10-6 to 10-5 is achieved close to capacity limit. Due to interleaving, the bit errors are spread within a block rather than being clustered. Thus, for packet data applications packets of 96 bytes (768 bits) are first encoded with a binary BCH code of block length 808 bits. The decoder is able to correct up to 4 bit errors within a codeword by bounded minimum distance decoding, thereby operating with very low probability of undetected errors due to erroneously decoded packets. 4 such encoded packets are combined and jointly encoded by the PCCC. To comply with the overall codeword length of 6272 bits the parity bits of the PCCC are slightly more punctured than without additional outer code. The overall code constructed has higher minimal distance than the PCCC alone and the flattening in the error rates observed with PCCC for high SNR is eliminated or at least significantly reduced. TCC - C TCC-C was designed to support high data rate packet transmission with a maximum data rate of R b =32 Mbits/s. This requires a coding scheme with a transmission rate of R Tx =4 bits/channel use. The high spectral efficiency indicates the necessity of a coded modulation scheme. The required channel improvement is achieved by utilising of multiple antenna concepts both at the receiver and the transmitter. To enable accurate channel measurements and antenna adaptation the user is restricted to low mobility. Furthermore, the use of smart antennas makes the the channel less dispersive, which effectively results in a close to AWGN channel behaviour. A spectral efficiency of 4 bits/channel use requires an increased signal constellation. By information theoretical arguments [5], an extension of the signalling alphabet to 32-QAM is sufficient. Most designs of coded modulation schemes favour a joint approach to coding and modulation by well known trellis-coded modulation or multi-level coding. These approaches work satisfactorily for the particular channel they were designed for, but in most cases lack robustness to channel variations. E.g., trelliscoded modulation schemes for AWGN channels maximize Euclidian distance while for fading channels Hamming distance is decisive. Recently, with bit interleaved coded modulation (BICM) a new approach has been presented deliberately separating coding and modulation [7]. The clue is a labelling of the signal points such that in a parallel bit channel model, one channel for each bit in the label, each channel has roughly same capacity. Then, instead of independently coding each level with a binary code as in multi-level coding, a single binary code of increased block length and interleaved transmission over the bit channels is equally appropriate. Surprisingly, a

4 simple Gray labelling results in nearly equal bit channel capacities. Due to the separation of coding and modulation BICM schemes are much more robust to channel variations, and we resorted to BICM for TCC-C. To achieve a transmission rate of 4 bits/channel use 32-QAM with (quasi)-gray labelling is concatenated with a simple block interleaver and a powerful PCCC of code rate 4/5 (see Fig. 3). PCCC encoder R = 4/5 Fig. 3 - Bit Interleaved Coded Modulation The design of the block interleaver is not critical. The PCCC encoder is equivalent to the one used for TCC-B except of additional puncturing of parity bits to increase code rate. Prior to decoding, probabilities for the labelling bits have to be estimated based upon the received symbol. Since we are interested in packet transmission with reasonable low packet losses, outer BCH encoding of packets of 96 bytes as described in the previous section is applied. Mind, that the BCH code corrects up to 4 independent bit errors in packets after PCCC decoding, or indicates all non-correctable error patterns. By our coding construction, a modem upgrade from TCC-B to TCC-C requires only a change in puncturing and signal mapping apart from multi-antenna processing. 2.3 Rake Processing bit interleaver 32-QAM (cross) quasi Gray mapping The use of spread spectrum in TCC-A and TCC-B implies that Rake processing should be employed to make use of the multipath channel diversity. Using multiple antennas upon reception, as in the uplink of TCC-B, provides a new spatial signal dimension. A simple implementation that makes use of this dimension is a beam former followed by a conventional Rake receiver, but in this case the paths that fall outside the main beam are not considered. The optimal receiver is a space-time Rake [9], in which a beam-former is implemented for each delayed path, or, alternatively, a conventional Rake receiver is implemented on each antenna element. Since TDD is employed, the channel estimations obtained in the uplink can be used to pre-process the signal with a pre-rake. In [8] it has been shown that the link performance can be significantly improved if a combination of a pre-rake at the transmitter and a Rake receiver, which we call a post-rake, is used. The post- Rake is basically a conventional Rake receiver, but it is matched to the combination of pre-rake and multi-path channel. The post-rake requires little extra complexity in relation to a conventional Rake receiver and provides a significant performance gain. The reason for the better performance with a pre- and a post-rake is that the pre-rake corresponds to a channelmatched pre-filtering. With this technique the signal power is concentrated at the frequency ranges where the channel is most favourable. The principle behind channel matched transmission is analogous to the concept of water-filling. However, ideal water-filling cannot be applied since the receiver noise is in general not known at the transmitter. 3 - Results The air interface presented in Section 2 was simulated and some of the results will be shown here to demonstrate the feasibility of the presented design concept. In order to take into account the effect of multiple antennas in the link performance we have considered a vector channel model, which is described in detail in [10]. In this model the multiple Rayleigh- or Rice-faded paths, besides having different time delays, also come from various directions, which are obtained from a geometrically based scatterer model [11]. We have used the same parameters for all traffic classes, with 6 paths (1st path with Ricean fading, K=10dB, the remaining with Rayleigh fading) and a maximum path delay of 500 ns. The results in Fig. 4 show the results for TCC-A (64 kbits/s) in the low bit error and in the low packet error modes (mode A3). We see in both cases that a good performance can be achieved with a reasonable signal-tonoise ratio. We also observe the significant performance improvement that can be obtained with a pre- and a post- Rake compared to a conventional Rake receiver. It should be reminded here that in this traffic class one single antenna is employed both at the base and at the mobile station. Bit Error Rate / Frame Error Rate 10 4 low bit error modus E /N (db) Fig. 4 - TCC-A (Downlink) Rake receiver pre /post Rake low packet error modus In Fig. 5 the results for TCC-B (2 Mbits/s) in the low bit error mode and with turbo code are shown. The results of the modulation and coding scheme with no antenna diversity and a conventional Rake receiver are compared with the ones obtained at the uplink with 8 receive antenna elements at the base station and space-time Rake processing. It can be seen that with the latter technique the whole antenna gain (a factor of 8, corresponding to 9dB)

5 can be realised, which is required for service to be supported at a reasonably low signal-to-noise ratio. Bit Error Rate Fig. 5 - TCC - B (Uplink) In Fig. 6 the uplink packet error rate in TCC-C is shown. The results for an AWGN channel without multiple antennas and for a system with conventional phased array beamforming at both the base (8 antenna elements) and the mobile stations (4 antenna elements) are shown. The beam is centred at the line-of-sight (LOS) path and no equalisation is performed at the receiver. The results are displayed for 5 different channel configurations, where in each configuration the multipaths come from different directions. We observe that the performance is strongly dependent on the current channel configuration. For configurations 1 to 3 a good performance can be achieved with the implemented system, since the non-los paths impinge outside the main lobe and are thus strongly attenuated. The performance gain in relation to an AWGN channel is due to the multiple antennas (a factor of 8 4, appr. 15dB). In channels 4 and 5 the non-los paths were not eliminated by the beamformer and inter-symbol interference arises. In this case an equaliser or a smarter beamformer that tries to attenuate the non-los paths have to be employed. Ideally, in order to make use of the diversity provided by the several paths, a space-time equaliser should be employed. Nevertheless, we have shown that in most cases a channel with hardly any intersymbol interference is obtained with a simple beamformer and LOS propagation, and that an equaliser is not always necessary. Packet Error Rate E /N (db) s 0 Ch. 3 Ch. 2 Ch. 1 Ch. 4 and 5 beamforming Fig. 6 - TCC - C (Uplink) 1D Rake rec. Space Time Rake rec. AWGN E s /N 0 (db) 4 - Conclusion In this paper we have proposed an air interface for an Integrated Broadband Mobile System based on a single modem concept (Fig. 2) for all traffic channel classes. We have designed a flexible coding scheme to support a variety of services requiring low BER performance or low packet losses. Using low-rate codes to combat both multiaccess interference and noise for low rate transmission services. For services requiring higher transmission rates, coding is based upon suitably modified parallel concatenated convolutional codes ( Turbo -codes) in conjunction with bit-interleaved coded modulation principles for TCC-C. We have also demonstrated that appropriate Rake processing (pre-/post-rake, space-time Rake) can improve link performance significantly in spread spectrum transmission. 5 - References [1] Ojanpera T.and Prasad R., "An Overview of Air Interface Multiple Access for IMT-2000/UMTS", IEEE Comm. Mag., September 1998, pp [2] R. Keller et al., Wireless ATM for Broadband Multimedia Wireless Access: The ATMmobil Project, IEEE Personal Communications, October 1999, pp [3] M. Bronzel et al., Integrated Broadband Mobile System (IBMS) featuring Wireless ATM, Proc. of the ACTS Mobile Communication Summit 97, Aalborg, Denmark, 7-10th October, 1997, pp [4] D. Hunold, A.N. Barreto, M. Bronzel, G. Fettweis, "Investigations on Capacity in the Integrated Broadband Mobile System (IBMS) Using a Wireless Network Simulator", Proc. of the MoMuC 98, Berlin, October 1998, pp [5] G.C. Clark, J.B. Cain, Error-Correction Coding for Digital Communications, Plenum Press, New York, 1981 [6] G. Ungerboeck, Channel Coding with Multilevel/Phase Signals, IEEE Trans. on Inf. Theory, Jan. 1982, pp [7] G. Caire, G. Taricco, E. Biglieri, Bit-Interleaved Coded Modulation, IEEE Trans. on Inform. Theory Vol. 44, No. 3, May 1998 pp [8] A.N. Barreto and G. Fettweis, Performance Improvement on DS-Spread Spectrum Systems using a pre- and a post-rake, Proc. of the Int l Zurich Seminar 2000, Zürich, Switzerland, Feb [9] A. Paulraj, B.C. Ng, Space-Time Modems for Wireless Personal Communications, IEEE Personal Comm. Mag., Feb. 1998, pp [10]J. Jelitto, M. Stege, M. Löhning, M. Bronzel and G. Fettweis, A Vector Channel Model with Stochastic Fading Simulation, PIMRC 99, Tokio, 1999 [11]J.C. Liberti and T.S. Rappaport, A geometrically based model for line-of-sight multipath radio channels, Proc. of VTC 96, pp