Research On Source Number Detection In Colored Noise

The over- and under-estimation of source number and their effects to the MUSICalgorithm are firstly analyzed in this paper. Simulation experiments show that in theover-estimation case, false peaks appear in the MUSIC spectrum, and some peaksdisappear in the under-estimation case. Then a modified MUSIC algorithm isconsidered. This algorithm does not need the knowledge of source number and performas good as the MUSIC algorithm provided with the true source number. Some sourcenumber detection algorithms are introduced. The information criteria and the predictedeigen-threshold approach are analyzed, followed by the analysis of the effect of colorednoise to the eigenvalues of the covariance matrix. Then a method to eliminate theinequality of the eigenvalues caused by colored noise is presented. Using the correctedeigenvalues, the information criterion and the predicted eigen-threshold approach canestimate the number of sources in colored noise. Computer simulation and theprocessing results of real array data show that the AIC criteria and the predictedeigen-threshold approach, using the corrected eigenvalues, outperform the Gerschgorindisk method, while the MDL criteria underperform the Gerschgorin disk method.