A numerical Bayesian approach for DOA and frequency estimation of exponential signals in Gaussian and non-Gaussian noise

We present a Bayesian approach for DOA and frequency estimation of narrowband signals in additive complex Gaussian and non-Gaussian noise. Using Bayesian techniques, the a posteriori probability densities for DOA and frequency parameters are derived from the signal and noise models. These posterior probabilities are then used in the self-targeting Metropolis-Hastings algorithm to derive the samples for the DOA and frequency parameters. The mean square errors (MSE) of the parameters are compared with the Cramer-Rao lower bound (CRLB) and with various subspace-based methods. Unlike the conventional subspace-based methods such as MUSIC, ESPRIT etc., this new algorithm can be used with a significantly lower number of samples to estimate the parameters with acceptable MSE