Geometric Feature Equalizers Based on Hybrid Genetic Algorithm

A nonlinear geometric feature equalizer adopting minimum bit error rate principle is proposed in this paper for the filtering of noise and interference whose frequency band overlaps with the desired signal in communications, and a novel hybrid genetic algorithm, namely hybrid genetic algorithm-stochastic gradient, is also proposed for training the equalization model. Considering that the noise and the interference have different stochastic character, the desired information is recovered by neural network based on minimum bit error rate principle. Simulation results show that when extended binary phase shifting keying signal is contaminated by the mix of white Gaussian noise and relatively strong interference signals of amplitude modulation and frequency modulation, the performance of matched filter or linear equalizer degenerates rapidly, but geometric feature equalizer provides very low bit error rate. Furthermore, performance of hybrid genetic algorithm is superior to that of stochastic gradient.