A HYBRID DIGITAL SIGNAL PROCESSING-ANNEALED NEURAL NETWORK DETECTOR FOR CODE-DIVISON-MULTIPLE-ACCESS MULTIUSER DETECTION

Abstract

Due to the nature of CDMA, multiple access interference (MAI), caused by spreading codes not being orthogonal to one another and worsen further by near/far problem, are unavoidable. Conventional matched filter has very bad bit error rate (BER) performance, even with the use of power control (which limits transmitting power of stronger users) to combat near/far problem. Power control technique is also self-defeating, reducing the capability of the CDMA system. Multiuser detection has been proposed to overcome the shortcoming of conventional detectors. The Optimal Multiuser Detector (OMD), proposed by Verdu [3], is too complicated to implement; its complexity increases exponentially with increasing number of users. Research is focused on the search for a sub-optimal solution that has good BER performance and lesser hardware complexity than that of OMD. Several sub-optimal detectors have been proposed but they are not practical in actual implementation, suffering from either increased complexity or performance degradation with more users in the system. In this project, we proposed a new multiuser detector which we believe will have very good performance and acceptable hardware complexity. The proposed detector is a 2-stage multiuser detector for DS/CDMA. The first stage uses a reduced detector to simplify the optimisation process. This is carried out by a procedure which iteratively partitions the optimisation search space into manageable proportions. Subsequently, the second stage employs an annealed neural network to extract the multiple signals from different users. The 2-stage system draws on a combination of earlier works [10,11] and the result is a system that performs better than the former separate systems. Simulations, performed using Matlab 5 .0, have shown that the performance of the proposed detector is superior to the other sub-optimal detectors as well. It is able to achieve low BER and combat near/far problem effectively. These improvements are demonstrated with comprehensive analyses and simulations for the A WGN channel. Synchronous and asynchronous DS/CDMA systems with various operating conditions are used in the simulations. Simulations are performed for both synchronous and asynchronous systems. In these systems, we vary parameters such as spreading codes, number of users in the system, power ratio etc. to give a broader view to the proposed detector's performance. For example, the spreading codes perform differently from one another; having very different auto-correlation and cross-correlation properties. It is interesting to see the BER performance of these sub-optimal detectors under different operating conditions. Our proposed detector is also flexible; parameters of the detector can be changed to suit different environments. Neural networks offer an exciting new dimension to our present day communications problems. Developments in these areas of research will greatly enhance our understanding and forge a new direction for the communication industry. This thesis is mainly concerned with combating near/far problem in Additive White Gaussian Noise (AWGN) channel. The thesis can be extended to include the study of Multipath fading on the proposed multiuser detector.