Index assignment for transmitting vector quantized LSF parameters over binary Markov channels
R. Iordache, I. Tabus · 2003
This paper introduces a new index assignment (IA) technique for transmitting vector quantized line spectrum frequencies (LSF) over channels with memory. The original performance criterion, the channel distortion D/sub C/, is replaced with one approximation, the generalized linearity index, having the advantage of being faster to evaluate. We then present a suboptimal algorithm for maximizing the generalized linearity index, and illustrate its performance in vector quantization of a Gaussian source over a Markov channel (the specific Markov model being known as "contagion urn model"). We also apply the algorithm to the transmission of LSF parameters quantized as in G.729 standard. The results encourage the use of the generalized linearity index for assigning the indices for quantized LSF parameters transmission over channels with memory, showing consistency of high generalized linearity index with low spectral distortion of LSF parameters.