GA Optimized HMM Error Model for OFDM

Abstract

Finite-state HMM error models are an established powerful tool for capturing temporal characteristics of fading channels. They make network simulations much more practical and fast while developing and testing higher layer wireless system protocols and designing interleaving and FEC schemes. This paper has introduced two new Genetic Algorithm (GA) based approaches, namely, GA-S and GA-B as an alternative to HMM error models, for the accurate learning of error statistics of autocorrelation function (ACF) and error-run distributions. Validity comparisons of fit obtained show that the GA method is better than both the conventional Baum–Welch algorithm (BWA) and Semi-hidden Markov model (SHMM) methods, even when the target sequences are as short as 1000 in length. The independent elements of the [A] and [B] parameters of BWA used as chromosomes of the population space are used for error generation within the Genetic Algorithm. Mean square error of statistical properties of the error sequences is used to determine fitness of the chromosomes unlike other works using average log-likelihood ratio. Applicability has been tested by numerical simulations using error sequences of different lengths as well as target sequences of fixed length from an OFDM transceiver system under different fading rates. A T _s spaced time-delay model of the propagation channel with a fixed power delay profile has been used. For slow faded channels or when long sequences of error free intervals poses difficulties for BWA training, the GA method proves to be an excellent alternative for fast and accurate modeling of the error bursts. Unlike the computationally cumbersome BWA or the simplified SHMM approach, the GA model is capable of arriving at desired levels of accuracy with two to three states, in contrast to the Markov models needing a much higher number of states.