Abstract: For traditional finite-impulse-response (FIR) filter, designers usually default to that the coefficient should be adjacent. In this paper, we propose a new model of the FIR filter based on the L0-norm, which improves the performance of FIR filter by increasing few delays and keeps the linear phase at the same time. Genetic algorithm is proposed to solve the proposed non-convex optimization model. An approximate solution is derived based on the result of genetic algorithm to reduce the computational burden. Computer simulation results demonstrate its lower approximation error compared with traditional FIR filters with the same filter order.