Towards hardware implementation of genetic algorithms for adaptive filtering applications

Genetic algorithms (GAs) have been successfully applied to resolve adaptive filtering problems. The main advantage of using such algorithms over conventional adaptive filtering techniques, is their ability to deal with nonlinear systems. However, intensive computations are needed to achieve proper performances, which can be very critical when limited-resource devices are considered for hardware implementation. This work proposes a low computation load genetic algorithm for adaptive filtering applications. Very low bit-wordlength fixed-point arithmetic is used in all operations to minimize the algorithm footprint. We compare the proposed method with the least mean square (LMS) and theatrical performances in identifying the parameters of an auto regressive moving average (ARMA) model. Simulation results show the high performances of the algorithm to deal with linear and nonlinear environments where only 6-bit wordlength is used.