Algorithms and architectures for dynamic programming on Markov chains

Algorithms and architectures are developed for dynamic programming on finite-state Markov chains. The results are applied to phase tracking of a phase-jitter carrier, Viterbi decoding of convolutional codes, and isolated-word recognition using hidden Markov models and dynamic time warping. It is argued that a computing ring, which is really a cyclic systolic array of cells, provides a good tradeoff between performance and complexity for implementation in VLSI. The computing ring may be pipelined with a preprocessor and a memory management unit to achieve a modular design.<
>