Partial search vector selection for sparse signal representation

Abstract

In this paper a new algorithm for vector selection in signal representation problems is proposed, we call it space is searched. Partial Search (PS). The vector selection problem is described, and one group of algorithms for solving this problem, the Matching Pursuit (MP) algorithms, is reviewed. The proposed algorithm is based on the Order Recursive Matching Pursuit (ORMP) algorithm,it extends ORMP by searching a larger part of the solution space in an e®ective way. The PS algorithm tests up to a given number of possible solutions and returns the best, while ORMP is a greedy algorithm testing and returning only one possible solution. A detailed description of PS is given. In the end some examples of its performance are given, showing that PS performs very well, that the representation error is considerable reduced and that the probability of finding the optimal solution is increased compared to ORMP, even when only a small part of the solution space is searched.