Detection and separation of the sources from underdetermined instantaneous mixtures without estimating the inverse mixing matrix

In BSS (Blind Source Separation) the knowledge of mixing parameters and the nature of source signals are not available instead only the mixtures of sources are available at the sensor from which the individual source signal has to be detected and separated. Separation of sources from an instantaneous mixture of source signals requires determining number of sources and locating single source region. Most of the BSS algorithms require an inverse mixing matrix to be estimated in order to separate the sources which is difficult especially for the case of underdetermined mixture where number of source signals is more than the number of sensors leading to a non-square mixing matrix. In this paper a method to detect and separate sources from a linear instantaneous mixture of source signals is presented. This method involves detection of number of sources in the mixture by using time frequency ratio of mixed signals. The key idea used here is the sparseness of sources in the time frequency domain when compared to time domain. Once the number of source signals are determined the region of single source signals are identified to separate the sources. The sources are separated without estimating the inverse mixing matrix thereby eliminating most of the assumptions in literature which is usually done on mixing conditions.