9303868 Fischer The goal of the project is to develop effective trellis source codes and structured vector quantizers. The trellis codes are constructed to use trellis coded quantization to achieve significant granular gain, and trellis boundary shaping to achieve significant boundary (entropy) gain. The main objectives of the research are to develop a theory for analyzing the performance of the trellis codes, and develop efficient algorithms for the code design. The method will also be extended to develop trellis codes for sources with memory. Two classes of structured vector quantizers are developed for low-delay source coding applications. The first class is a two-stage vector quantizer/lattice vector quantizer, with soft-decision first-stage encoding. Preliminary results indicate that this vector quantizer structure, while simple, provides performance (for memoryless sources) that is competitive with, or superior, to all other existing techniques. The second class of vector quantizers is lattice-based, with an elliptical boundary region. The main objective is to develop efficient algorithms defining the vector quantizer so that it can be made adaptive. The elliptical vector quantizer will be evaluated in a speech coding system. ***
