A New Algorithm to Estimate Mixing-Matrix of Underdetermined Blind Signal Separation

Abstract

The paper puts forward a new algorithm to estimate mixing-matrix according to the underdetermined blind signal separation of 3 observed signals and 4 sources. According to the geometric meaning of the SCA model, the paper analyzes the numerical feature of the observed signal and proves that the inner product under Euclidean space can be used to classify the observed signal in the situation. Besides, the paper gives a method for determining the number of source signal and introduces an estimation algorithm for mixing-matrix using inner products in the Euclidean space combined with the density of interval point. The algorithm can effectively identify the number of source signals and can realize the estimation of mixing-matrix. The experimental results show the algorithm is feasible.