A Novel Approach for Recovering 2-valued Independent Sparse Components from Whitened Data in Noisy Environments
Mahmoud Keshavarzi, Siavash Bayat, Sahar Keshavarzi · 2016
Some sources transmit periodic and quasi periodic sparse pulse trains in the environment and a number of sensors might receive them through a single channel simultaneously. It is usually our interest to know which pulse belongs to which source. This identification process has wide applications in communications, radar system, medical applications, and neural systems. Blind source separation (BSS) is one solution for this problem. This paper proposed a geometrical approach to solve BSS problem when observations are whitened data and are obtained from the linear mixtures of 2-valued sparse signals (such as sparse pulse trains). In other words, the proposed approach aims to estimate a rotation matrix, and then to recover the independent sparse components from the whitened data. There are two assumptions in this work: First, all components of the source vector are non-negative 2-valued sparse signals. Second, all entries of the mixing matrix are positive. We have also provided some numerical simulations to illustrate the proposed method's performance. The results show that the performance is affected by the SNR value in sensors.