Incremental design of high complexity FIR filters by genetic algorithms

A genetic algorithm (GA) methodology is developed to design linear phase finite impulse response (FIR) filters, incrementally. Traditional GAs initialize the population with random generated individuals and evolve that population. When this method is applied to design medium to high complexity FIR filters, GA can not find the global minimum, easily. In order to overcome this problem, this paper proposes an incremental evolution strategy. The method starts by evolving a filter with a small number of taps. After convergence, the number of taps is increased and the population is initialized with the result of the previous evolution. This process is continued until a predefined error level is reached. The method outperforms the well known FIR filter design methods: window based design, Parks and McClellan equiripple algorithm, and least squares design