Symbol-MAP-based trellis vector quantization

We exploit the similarity between source compression and channel decoding to develop a new encoding algorithm for trellis vector quantization (TVQ). We start by drawing the analogy between TVQ and the process of sequence-ML channel decoding. Then, the new search algorithm is derived based on the symbol-MAP decoding algorithm, which is used in soft-output channel decoding applications. Given a block of source output vectors, the new algorithm delivers a set of probabilities that describe the reliability of the different symbols at the encoder output for each time instant, in the minimum distortion sense. The performance of both the new algorithm and the Viterbi algorithm is compared using memoryless Gaussian and Gauss-Markov sources. The two algorithms provide expected similar distortion-rate results. This behavior is due to the fact that sequence-ML decoding is equivalent to symbol-MAP decoding of independent and identically distributed data symbols. Although the new algorithm is approximately 4 times more complex than the Viterbi (1974) algorithm, it provides distortion-dependent reliability information that can be used to improve the quality of compression.