Adaptive control in digital communication systems

This dissertation represents the results of research endeavors on the investigation and development of efficient adaptive control algorithms and system architectures with which the performance of the decoding of convolutional codes employed in forward error correcting digital communication systems can be improved.;The researches were concentrate on the development of various adaptive control algorithms and thus different adaptive coprocessors that may be used to adaptively adjust the inputs to a digital soft-decision maximum likelihood decoder (a Viterbi decoder).;Three different algorithms and the corresponding system architectures of the coprocessors were studied for different digital communication channel conditions, such as the channel with pure Additive White Gaussian Noise (AWGN), the channel with Rayleigh fading noise, and the channel employing the Code Division Multiple Access (CDMA) cellular system.;Two different neural network coprocessors, trained with reinforcement learning algorithm or backpropagation learning algorithm, were developed to adaptively adjust the soft decision thresholds of a quantizer which is used to convert the analog demodulator outputs into three bits digital signals. Between these two algorithms, due to the fully utilization of the information about the distribution probability of the demodulator output signals, the coprocessor trained with the backpropagation learning algorithm provides better and consistent performance improvement.;The third algorithm is based upon the inherent property of the convolutional encoder and, instead of adjusting the thresholds of the soft-decision quantizer, the developed coprocessor is employed to change the soft decision bits directly. Since the algorithm does not count the "soft" information contained in the demodulator outputs, the performance of this approach is poor.;As compared to a system without a developed coprocessor, smaller Bit Error Rate (BER) can be obtained from the system with a developed neural network coprocessor.