Tandem Transcoding Without Distortion Accumulation for Vector Quantizers.

The problem of distortion accumulation due to tandem transcodings in a network arises when high-quality, low-rate quantizers (codecs) are introduced into the network. An important class of codecs is vector quantization, which is gaining popularity as a high-performance quantization technique with possible applications in speech and image coding. In this paper, we provide novel transcoding algorithms that eliminate distortion accumulation due to tandem transcodings for memoryless, finite- state, and predictive vector quantization. The algorithms can be implemented via table lookups for the first two, whereas the latter requires online calculations. For 16 kbps predictive vector quantization and 48 kbps PCM systems operating over four tandem connected links, the new transcoding algorithm provides 7 dB improvement relative to conventional tandem encodings. The technique is also applicable to adaptive and adaptive predictive vector quantization.