A Least-squares Design of Low Delay FIR Filters with Sparse Coefficients Using Accelerated Proximal Gradient Method

Abstract

We treat an optimization problem for low delay FIR digital filters with sparse coefficients in the least squares sense. Sparse filters indicate that the coefficients include not only real value but also 0 coefficients. In this study, we consider l1 regularization scheme to solve the design problem for sparse filters. Here, we employ an algorithm for design of sparse FIR filters based on an accelerated proximal gradient method. Finally, we show examples to demonstrate the effectiveness of the proposed method by comparing with equivalent non-sparse filters and filters designed by FOBOS.