AbstractTurbo codes is a new class of error-correcting codes that can approach the Shannon bound by a decibel or less with iteratively processed maxiinum-a-posteriori decoders. Now, turbo codes is considered as one of the most exciting and potentially important developments in coding theory in many years. Its invention has changed the conventional design principles of the coding scheme from the attempt to increasing the minimum Hamming distance of the code to the goal of reducing the multiplicity of codewords with low Hamming weights.This thesis investigates some problems of the theory and practice of turbo codes with emphasis on the design of the constituent code and the interleaver, continuous decoding algorithms and their performance on Rayleigh fading channels. The main results are:I.The developments in channel coding theory and the Shannon bound for one and two-dimension signals are surveied;2.The principles of turbo codes are described, and the simulation results of demonstrating its performance for the BCJR, Max-Log-MAP and M-BCJR soft-output decoding algorithms are presented.3.The performance of the turbo codes employing 16, 8 or 4 states constituent codes are compared with each other. The effect of a block interleaver or a random interleaver on turbo codes is analyzed, and the concept of 揺ffective length of codewords~ is presented.4.Two novel type of soft-output decoding algorithms for turbo codes with a convolutional interleaver is proposed, based on a decision rule of symbol-by- symbol maximum a posteriori (MAP) probability. Each of them requires only a forward recursion and has lower decoding delay. Two algorithms can be used in continuous decoding for both recursive and non-recursive encoders.5.The simulation method and the performance of turbo codes on a Rayleigh fading channel are described, and an average upper bound for the maximum likelihood decoding strategy of turbo codes is derived.