Blind iterative multiuser detection for error coded CDMA systems.

While COMA is proving to be a reliable communication technology, it is still susceptible to the effects of proximity problem and multiple access interference. A history and overview of existing COM standards is also provided. The need for multi-user detection is explained and a description and comparison of different detection methods that have appeared in the literature is given.

LIST OF TABLES

ACRONYMS

CHA PTERl INTRODUCTION

Motivation

A common blind detector uses the minimum output energy (MOE) criterion, where output variance is minimized with respect to components orthogonal to the desired user's dispersion waveform. The aim of this research is to develop a detector that has improved resistance to the effects of interference, multipaths and the 'near-far' problem. The proposed solution to these problems is to combine the blind MOE and iterative techniques, resulting in a blind iterative detector that requires as little time as possible. amount of information for multi-user detection, and still provide error correction capabilities to further improve performance.

Organisation of the Thesis

The two main iterative decoding techniques, maximum a posteriori and soft-output Viterbi algorithms, are explained, as these techniques are fundamental to the operation of COMA iterative detectors. The results show an improved performance due to the iterative structure for different operating conditions.

DS-CDMA

Multi-Access Communications
Spread Spectrum Communications
DS-CDMA Model

Since the entire bandwidth is used for each signal, the bandwidth of the transmitted signal IS. The user's signal is decoded by mixing the received signals with orthogonal signat ure waveforms. The auto-co rre lat ion will allow the desired wave to pass; but if the signature waveforms are orthogonal, the cross corre latio's component cancels out from many other users.

Figure 2.1: Compa riso n of bandwidth usag e for TD I\IA. FDI\IA and CDI\IA systems

Pseudo-Noise Sequences

The chip codes that determine the users' characteristic waveform are pseudo-noise (PN) sequences, which can be generated by an array of feedback shift registers. The basic PN sequence is the maximum length sequence (ML). The generated sequence has a length L = 2N-1, where N is the number of shift registers.

Figure 2.4: Th e data . ch ip and spread (pro duct) waveforms

Existing CDMA Systems

IS-95 (CDMA One)
CDMA 2000

This is a dual-mode standard with an AMPS analog mode. The IS-95B standard was finalized in 1998. The access channel is used by the mobile station to initiate a call, respond to a paging channel message, and for location updates. The access channel only works with a fixed data rate of 4.8 Kbps. Each access channel is associated with a paging channel, so there can be up to seven access channels.

Table 2.3 [5] lists the parameters for the C DMA 2000 protocol , and the following paragraphs will giv e an overview of the channel structure

Conclusion

MULTIUSER DETECTION

Conven tional Detector
Th e Multiu ser Channel

The Near-Far Problem

Optimal MUD
D ecorrelating Receiver
MMSE MUD

Blind Adaptive Algorithm
Detector Performance
Parallel Interference Cancellation

Conclusion

Due to the effect of MAl, when the number of users in the channel increases, the performance of the system deteriorates. For n multipath components, each interfering user will generate n interfering signals, so the detector will have to remove n(K-l) signals. The desired user will also have n components that will need to be merged. Since the MMSE receiver takes background noise into account, the noise amplification problem of the decorrelation detector is circumvented. The BER performance is thus better than that of the decorrelation receiver, and when the background noise is reduced to zero, the MMSE detector approaches the decorrelation receiver.

A measure of performance is the signal-to-interference ratio (SIR) at the output of the linear transform. The projection of the MOE output energy gradient (x l) on the orthogonal linear subspace must be found, so that equation (3.12) satisfies the orthogonality condition at each step of the algorithm. The steepest line of descent along the subspace in SI is the gradient projection into that subspace. The more complex the recursions, the higher the convergence speed will be. The blind detector can be implemented in asynchronous channels.

Figure 3.2: The effect of channel population on performance, SN R = IOd8.

ERROR CODING

Blo ck Codes

Genera l Block Code Structure and Performa nce
Coding and Decoding of Block Codes
Examples of Codes

Burst Error Correction

Block Interleaving
Convolutional Interleavin g
Reed-Solomon Codes

Convolutional Codes .1 Encoding

Decoding

Turbo Codes

Encoder
Decoding
Effects on Turbo Code Performance

Conclusion

The syndrome S for an error in the jth element represents the jth column of the matrix H. With the shift registers in place, each of the rows will have a delay of (I - I)s, so the output sequence will still be the same as the input sequence, except that d ( k) delayed by (1-l) with respect to tod(k). The number of bits stored is the limit range, k. The flip-flops (or shift register elements) are connected to v modulo-2 adders (EXCLUSIVE-OR logic flights), and the commutator samples the outputs of the flights. .

The path with the fewest (or no) errors is the 'correct' path. The three-digit value is the output, VIV2V3. From this the input bit can be calculated. It relies on the fact that paths entering a node in a lattice will be identical from that point on. For Viterbi decoding, the metric is calculated for the single path entering each state of the encoder. If multiple paths connect to the same node, discard the one with the largest cumulative mismatch (errors in Figure 4.6), the surviving paths are saved, and this process continues until the end of the message. The same k information sequences are interlaced and fed into the second encoder; the (n-k) control sequences generated by the second encoder are also transmitted. The code rate is then kI(2n-k) [25].

Thus, the metric should be defined such that maximizing the metric will maximize P(Yi';Ps:). The metric should also be easy to compute in a recursive manner from the (k-l)th step to the kth step of the trellis. The length of the frames has a big impact on performance. The original paper on turbo codes by Berrou, Glavieux and Thitimajshima [28], shows results close to the Shannon limit for large frame lengths. There is a disadvantage to using such large frame lengths in that it results in large delays. But smaller frame lengths still result in good performance [24].Dolinaret al.

CHAPTERS BLIND ITERATIVE MUD WITH ERROR CODING

Iterative MUD .1 General

Soft-Output CDMA Decodin g
Selected Iterative Detector Structures

A general description of the operation of the iterative detector will be given, as shown in Figure 5.1. It accepts soft decisions from the COMA decoder and executes its own soft decisions, which are interleaved to be processed for the next iteration. After the required number of iterations are completed, the hard decision d is made. Iterative decoders converge to their final value, varying between 3 iterations [48] and 10 iterations [41]. The receiver structure proposed in [47] is similar to that shown in Figure 5.1, where the SISO block is a MAP decoder, and the Soft Outp ut COMA block is a 'COMA MAP' decoder.

The difference between the two is that a MAP decoder outputs message symbols that accepts encoded symbols, while a COMA MAP decoder accepts spread symbols as initial input - it must be able to take into account the COMA channel. Here the 'APP' blocks represent the sent MAP decoders, and they, along with the linker and delinker, work as described above. The 'tanh' functions generate soft decisions with which the canceled received signal can be calculated. Each individual decoder processes the received signal and generates soft decision values of the transmitted coded symbols dk[J]. A soft decision decoder), in the general form of. Where is the filter to return to mine?

Figure 5.2 shows the block diagram of the propo sed rece iver in [45]. Here the ' APP' blocks repre sent MAP decoders, and they, along with the interleavers and de-interleavers, function as described above

Conclusion

PERFORMANCE OF THE PROPOSED BLIND ITERATIVE DETECTOR

The Effect of Channel Parameters .1 Channel Population

The effect of the number of users on performance will be discussed in more detail in Section 6.2.1. Figures 6.6 and 6.7 show the effect of channel population on the iterative decoder compared to that of the blind MOE and analytical results respectively. It is clear that with an increase in the number of users in the channel, there is a decrease in performance. Again, the superior performance of the iterative decoder compared to the blind MOE detector can be seen.

The performance of the iterative detector will again converge to that of the blind detector, for the same reasons as above. The simulated and analytical results for the proposed blind iterative detector were presented in this chapter and the performance of the proposed detector was compared with the performance of the blind MOE detector. From this comparison, it can be seen that the proposed detector has superior performance to that of the blind MOE receiver.

Figure 6.1: I'erforman ce afte r 3 iterations for (a) 2 users. (b) 5 user s. a nd (c) 10 users

CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK

Research Conclusions

The advantage of this detector is that it has the iterative structure, but because it is blind, it only requires knowledge of the spreading order of the desired user. The initial detection is performed by a blind MOE receiver, which provides the initial soft inputs and estimates of the channel to calculate the priors. The proposed blind iterative structure was simulated using a (2, 1, 1) convolutional code. The simulated and analytical graphs showing the performance of the proposed blind iterative detector were compared with the performance graphs of the blind MOE detector. A modification can be made to the iterative blind adaptive algorithm proposed in [50] to incorporate error coding into the algorithm implementation.

The convolutional code in the transmitter can be replaced by a block code, since iterative decoding of block codes has been proven to be possible [59]. The general structure of the iterative blind decoders proposed in this report and in [54] can be modified to allow the implementation of other blind algorithms. The blind MOE detector provides an estimate of the channel parameters and soft initial decision input so that priorities can be calculated. The performance of the proposed system decreases with an increase in the channel population or the number of multipaths.

Garello and G.Montorsi," A search for good convolutional codes to be used in the construction of turbocodes," IEEE Trans. 42] Hesham El Gamal and Evaggelos Geraniotis, "Iterative Multiuser Detection for Coded CDMA Signals in AWGN and Fading Channels," IEEE JSAC, vol. 44] Slavica Marinkovic, Branka Vucetic, and Akihisa Ushirokawa, "Space-Time Iterative and Multi-stage Receiver Structures for CDMA Mobile Communication Systems" IEEE JSAC, vol.

53] Ejaz Khan og Dirk Slock, "Blind Iterative Receiver for Multiuser Space-Time Coding Systems,"Proc.61hBaiona Workshop on Signal Processingin Communications, 8-10 Sept. 54] Teng Joon Lim, Tao Zhu og Mehul Motani, " Blind Iterative Decision-Feedback Multiuser Detection ofFEC-Coded CDMA Signals,"IEEE Commun.Lett.,vol. 59] Joachim Hagenauer, Elke Offer og Lutz Papke,"Iterative Decoding of Binary Block and Convolutional Codes,"IEEE Trans.

PUBLICATIONS FROM THE THESIS