TRELLIS ENCODING FOR SOURCES AND CHANNELS (DATA COMPRESSION, QUANTIZATION, SPEECH CODING, JOINT SOURCE AND CHANNEL CODING)

Dynamic programming principles have recently been used to design source encoding systems with applications to speech and image coding. These systems encode large blocks of input data with relatively low complexity and high fidelity of reproduction.; Performance improvement is still possible, however, by exploiting the correlation of the input samples via prediction. In this thesis, novel search and design algorithms for predictive trellis encoders are presented. Simulation results are compared for trellis and predictive trellis codes on first and third order autoregressive sources and sampled speech.; For noisy channels, the distortion measure used by the encoder is modified to design joint source and channel trellis encoding systems. These systems provide protection against channel errors without employing explicit error control coding, thus avoiding complexity (for table-driven systems) and bandwidth increase. Experimental results are provided for independent and autoregressive Gaussian sources, binary symmetric channels, and absolute error and squared error distortion measures. The performance of the joint codes is compared with the performance of the tandem combinations of trellis source and channel codes. Performance improvement via a predictive joint source and channel trellis code is demonstrated. The noisy channel centroids are derived for various vector quantization schemes.; In the context of distributed noisy source coding, simulation results for low rate vector trellis waveform coding of random waveform sources corrupted by additive and statistically independent noise are presented. The code design algorithm is modified for the design of quantizers cooperating to compress a random source in a distributed context. Comparisons are made with an equivalent complexity, optimal, centralized quantizer of similar structure with access to the same observations.