Abstract: Correctly estimating the number of sources is a necessary condition for the majority of high-resolution spatial spectrum estimation algorithm. Source number estimations such as the criteria based on Akaike information theory (AIC), minimum description length (MDL) criterion, and Gerschgorin disk criterion (GDE) may lead to performance degradation, or even cannot correctly work in the low SNR. A novel source number estimation method based on joint probability density function of cosine of the angle between sample eigenvectors is proposed. Firstly, in noise subspace the cosine values between an eigenvector and other eigenvectors are obtained by the sample covariance matrix decomposition. Then the joint probability density function value of these vector angle cosine is determined. At last, two divided adjacent values of the density function is compared with a threshold value for determining the source number. Numerical simulation and tank experimental verification show that the performance of the proposed method is far better than that of the conventional algorithms mentioned above.

Key words: array signal processing, source number estimation, eigenvector angle cosine, joint probability density function