Abstract: This paper proposes an improved genetic algorithm for determining the optimal structure of the sparse Volterra filter. The algorithm defines chromosomes as the components of possible solutions instead of possible solutions in the classic genetic algorithm. The chromosomes are encoded with integer, which results in high efficiency. The genetic evolution process starts from the lower-order and short memory nonlinear model, and optimal setting is obtained at the end of the evolution process. Thus, the order and memory length of the nonlinear model needn’t be determined in advance, which avoids the booming expansion of the searching space for the high-order and long memory nonlinear model. The algorithm estimates the sparse kernel vector by the LMS algorithm, and therefore avoids repetitive estimation of the same vector and reduces computation. The simulation is finally carried out by applying the algorithm to nonlinear equalization for the data relay satellite channel, and the results show that the size of seeds group is largely reduced and that the genetic evolution converges faster.

Key words: adaptive Volterra filtering, genetic algorithm, sparse filter, channel equalization