An Improved Estimation Algorithm of the Source Number with Fewer Sensors Than Sources

The estimation of the source number is the foundation and the prerequisite for blind source separation. Most source number estimation algorithms assume that the number of the observed signals is more than that of the source signals. However, these methods are always failed when there are more source signals than observed signals. Traditional blind signal estimation methods for this problem do not have a good accuracy and a strong anti-noise property. This paper considers the problem of mixing matrix estimation in underdetermined blind source separation (UBSS). The application background of the algorithm is the underdetermined speech signal. We propose an improved estimation algorithm for source number based on the potential function clustering. First, short-time Fourier transform is performed to obtain sparser signals. It can remove some noises and redundancy information through the method. Then, we use the algorithm to detect the time-frequency (TF) points occupied by a single source for each source. Finally, we estimate the number of the source signals by designing an improved potential function clustering algorithm to obtain a better accuracy. It estimates the number of the source signals through estimating the local maximum. And the local maximum is estimated by clustering algorithm. Simulation results show that this paper can accurately estimate the source number when there are more source signals than observed signals. And the accuracy can improve evidently under the low signal-to-noise ratio (SNR). Further, the improved potential function clustering algorithm enhances the anti-noise performance and the stability to a certain extent and owns future prospects.