ENHANCED SEARCH FOR MULTISTAGE VECTOR QUANTIZATION

Abstract

The purpose of this dissertation is to describe a technique to enhance the search used for location of the codevectors during the process of Multistage Vector Quantization (MSVQ). Faster location of the codevectors in MSVQ is done by devising a search methodology based on codebooks which are designed by incorporating number of smaller codebooks in a structured manner at every stage. Binary tree is used for making this structure. Tree structure applied at each stage is independent of each other and is developed separately .We try to reduce the computational complexity which is the measured in terms of number of distance computations carried to reach the final quantized vector. For designing the stage codebook sequential design algorithm for MSVQ is used. Various parameters like dimension of vector, codebook size, number of stages, number of tree codebooks is kept into account while designing the required structure. To apply the search procedure an indexing of tree nodes is required for which an index assignment procedure is followed. Different stage codebook sizes, tree codebook parameters are tried for reaching the appropriate results. Comparison with multipath sequential search is also shown. This is comparatively raw area with ample scope of further research. This work provides scope for trying different techniques with varying combinations in forming alternative trees for further reducing the complexity.