Estimating the number of signals observed by multiple sensors

Abstract

Inferring the presence of signal sources plays an important role in statistical signal processing and wireless communications networks. In particular, knowing the number of signal sources embedded in noise is of great interest in cognitive radio. We propose a new algorithm for estimating the number of dominant sources observed by multiple sensors in the presence of multipath and corrupted by additive Gaussian noise. Our method is based on the exact distribution of the eigenvalues of the sample covariance matrix for multivariate Gaussian variables. Numerical results show that the new method has excellent performance, and is particularly important for situations with small sample size.