作者: S. Sivaprakasam , K.S. Shanmugan
DOI: 10.1109/26.380185
关键词: Markov property 、 Markov model 、 Markov chain 、 Markov kernel 、 Mathematics 、 Matrix analytic method 、 Combinatorics 、 Markov process 、 Continuous-time Markov chain 、 Algorithm 、 Variable-order Markov model
摘要: A hidden Markov model for burst errors is specified by a probability transition matrix P, an initial vector p, and the state dependent of error B. Several procedures are available estimating p B from given (observation) sequence. However, even with some restrictions on structure underlying models, estimation computationally intensive particularly when observation sequence contains long strings identical symbols. We show that, under mild assumptions, arbitrary P equivalent to unique "block diagonal" /spl Lambda/. also present very efficient algorithm Lambda/ set using modified Baum-Welch (1972) algorithm. >