Support Vector Regression for Basis Selection in Laplacian Noise Environment

Abstract

We demonstrate that the objective function of a basis selection problem in Laplacian noise environment falls into the framework of support vector regression (SVR), and, by iteratively solving a convex quadratic programming (QP) problem that guarantees a globally optimal solution, the sparse solution to the inverse problem can be found. The effectiveness of the proposed algorithm is verified via the application to direction-of-arrival (DOA) estimation. Different from the existing DOA estimation method based on SVR, the proposed algorithm is applicable with single snapshot and does not have to know the number of the sources. Meanwhile, the method does not require a large number of training sets, which in turn decreases the computational complexity.