Abstract:
This thesis investigates error control coding and multi-user detection for a single and multi-user communication system using methods of artificial intelligence. The main motivation of the research is to improve on the drawbacks and the constraints of the existing decoding/detection algorithms by proposing alternative methods based on the principle of soft computing. This thesis contains two main parts: the first part investigates the decoding of convolutional codes using a recurrent neural network and a support vector machine on a classical single-user communication system, and the second part studies a multi-user detector for a CDMA system based on a support vector machine. In particular, this thesis investigates a chaos-based CDMA system and compares it with other conventional systems. The theoretical analysis of the proposed methods are studied in detail by mathematical modeling and numerical examples, where all relevant design parameters and issues are considered. A quantitative approach is used to measure and compare the performance of the system by a series of Monte-Carlo simulations. The regular methods for convolutional decoding such as the Viterbi and the Turbo algorithms are reviewed and compared with the proposed methods. It is shown that the recurrent neural network decoder has a similar performance to the conventional Viterbi decoder, while the complexity reduces from an exponential to a polynomial function with respect to the encoder size. The inherent parallel processing capability of this decoder makes it suitable for high data-rate applications. On the other hand, the support vector machine decoder can learn and adapt to the surrounding environment, hence it achieves an extra coding gain of 2dB over the Viterbi decoder under a Rayleigh fading channel. Furthermore, the semi-blind support vector machine detector has a comparable performance with the well-known MMSE detector, and it is suitable for the forward link. The complexity of the detector is made even simpler than a matched filter receiver once feature extraction is incorporated. The results from the thesis suggest that these proposed methodologies can effectively make the radio links smarter and more flexible for future wireless systems.