Research on Source Number Estimation based on Geschgorin Disk Estimator Theorem and Minimum Description Length Criterion

This paper presents a method which combines Geschgorin Disk Estimator theorem (GDE) with Minimum Description Length criterion (MDL) for source number estimation in blind source separation. GDE-MDL method sets the advantages of GDE and MDL, which can accurately estimate the source number in color noise and improve the consistence of the source number estimation. This method based on the covariance matrix of the unitary transformation to suppress noise, can improve the sensitivity of plausibility function and have good applicability in low noise signal ratio and color noise. The simulation results show that the GDE-MDL method has good stability, and can achieve a better estimation of the number of sources in the case of white noise and color noise.