Vector quantization with size-constrained codebooks

Abstract

An algorithm for calculating lower bounds to the performance of all size-constrained vector quantizers has been derived earlier. Applications to the AR(1) source have shown that these lower bounds are very close to the actual rate-distortion curves, even for small sizes of the output alphabet. In this algorithm, however, the codebook changes with rate. In a number of applications it is desirable to operate with the same codebook. The authors modify the above algorithm to calculate lower bounds to the performance of all such multirate coders. An important conclusion is that constraining the codebook vectors does not degrade performance as compared to the size-only constrained case. The authors have proposed a practical alphabet and entropy constrained vector quantizer (AECVQ). This coder is obtained by modifying the ECVQ algorithm so that the codebook remains the same for all rates.<>