HMM Modeling of Burst Error Channels by Particle Swarm Optimization of the Likelihood Function

This paper proposes a new strategy for fitting Hidden Markov Models to error processes of channels with memory. Our approach consists of obtaining the analytical expression of the likelihood function of the model parameters and applying particle swarm optimization (PSO) to obtain their maximum likelihood (ML) estimates. In particular, this approach is here applied to the well known single error-state (simplified) Fritchman models, which have been recognized as a very useful tool for modeling error process of several communications systems over the last decades. The paper also addresses the mathematical analysis of several statistics of burst errors produced by these models. Some numerical examples are given in order to illustrate the effectiveness of the approach here proposed.