Low power Viterbi search architecture using inverse hidden Markov model

Presents a new VLSI algorithm and architecture for a low-power Viterbi search. The Viterbi search engine reduces the computational complexity of exhaustive matching between the training sequence and the observation sequence (or reference sequence). However, the search engine consumes much computation time and hardware resources for finding the maximum likelihood in a hidden Markov model (HMM). In this paper, we propose a fast Viterbi search engine using an inverse HMM (IHMM). The conventional Viterbi search engine finds the maximum-likelihood score in a path matrix of a HMM, whereas our search engine finds the minimum-dislikelihood score in a path matrix of a HMM. A benefit of this method is that it can remove redundant path matrix computations. The power consumption and the computation time are reduced by 68.6% for a 72.4% increase in the number of gates.