Blind channel identification using evolutionary programming

The problem of blind channel identification involves estimation of the channel coefficients based on the received noisy signal. The coefficients are estimated by using higher order cumulant fitting of the received signal. The optimization of the cumulant-fitting cost function is a multimodal problem, and conventional approaches using gradient algorithms often involve local optima in the absence of a good initial estimate. In this paper, we use evolutionary algorithms which evolve towards better regions of search space by means of randomized processes of selection and variation, to optimize the cost function. The effectiveness of genetic algorithms as well as evolutionary programming using self-adaptive mutation as stochastic optimization techniques is studied, and the results presented for the blind channel identification problem.