Ant Colony Clustering Algorithm for Underdetermined BSS

In underdetermined blind source separation, in order to quickly and accurately estimate the mixing matrix and the source signals, this paper presents a new algorithm based on ant colony clustering. The basic ideal of the algorithm is that we utilize the linear clustering characteristic of sparse signals to estimate the number of sources and the column vector of the mixing matrix by estimating the directions of the straight lines. In the preprocessing step, the observed signals in time domain are transformed to sparse signals in frequency domain. Through normalizing the observed data the linearity clustering is translated to compact clustering, and then using ant colony clustering to get the number of source signals and the mixing matrix. Finally, based on the estimation of the mixing matrix, the source signals are recovered by linear programming method. The simulation results illustrate the availability and accuracy of the proposed algorithm.