Genetic Algorithm Based Finite State Markov Channel Modeling

Abstract

Statistical properties of the error sequences produced by fading channels with memory have a strong influence over the performance of high layer protocols and error control codes. Finite State Markov Channel (FSMC) models can represent the temporal correlations of these sequences efficiently and accurately. This paper proposes a simple genetic algorithm (GA) based search for the optimum state transition matrix for a block diagonal Markov model. The burst error statistics of the GA based FSMC model with respect to Autocorrelation Function and error free interval distribution of the original error sequence are presented to validate the proposed method. The superiority of the GA approach over the semi-hidden Markov model (SHMM) based Fritchman model is exhibited in significant improvement of closeness of match and in the usage of shorter length of error sequences. Another Baum-Welch algorithm (BWA) based GA search method has been proposed and compared with the BWA and SHMM methods for the same error sequence. Again the superiority of GA approaches is recognized, especially for the smaller error lengths.

VQ-based model for binary error process

“…Less commonly used types of models include the chaosbased models (eg [15][16][17] and a more complex approach by Kopke et al in [18]), the discrete process-based generative model (DPBGM) based on the principle of Rice's sum of sinusoids (extensively described in [19] and [20], both papers are validating the model proposed in [21] on a real EGPRS channel trace), Stochastic context free grammars (SCFG), fractal models, multi-fractal wavelet model [22] and improvements of existing models by new concepts, such as genetic algorithms (eg [23]). …”