A novel adaptive algorithm applied to a class of redundant representation vector quantizers for waveform and model based coding

Abstract

Recently, novel vector quantization techniques in multiple nonorthogonal domains for both waveform and Linear Prediction (LP) model based, signal characterization have been reported. This approach gives an improved signal coding performance as compared to vector quantization in a single domain. In these techniques, each vector, formed either directly from the signal waveform or from the LP model coefficients extracted from the signal, is encoded in the domain that best represents the vector. An iterative algorithm for codebook accuracy enhancement, applicable to both waveform and LP model based vector quantization in nonorthogonal domains is developed and presented in this paper. In this algorithm, in the learning mode, each set of codebooks is retrained by those training vectors that selected that particular set of codebooks in the most recent iteration. Sample results are provided which clearly demonstrate the improved performance for the same bitrate.