Detection and localization of multiple sources via Bayesian predictive densities

A new approach based on a Bayesian inference scheme and unitary subspace decomposition is proposed to detect and estimate coherent and noncoherent signals. The authors assume that the signal vectors are random Gaussian vectors with zero mean and unknown covariance matrix and the prior of the direction-of-arrivals is a uniform distribution. Under these assumptions, the Bayesian estimator for the directional parameters coincides with the maximum likelihood estimator. In the detection part, the proposed detection criterion outperforms the minimum description length (MDL) principle and Akaike´s information criterion (AIC) particularly for a small number of sensors and/or snapshots, and/or low SNR. This is achieved without additional computational complexity. Simulation results that demonstrate the performance of the proposed solution are included.