Wideband Array Signal Processing Using MCMC Methods

To appear in IEEE Transactions on Signal Processing, 2004.
William Ng, J.P. Reilly, T. Kirubarajan, and J.-R. Larocque

Abstract   This paper proposes a novel wideband structure for array signal processing. A new interpolation model is formed where the observations are linear functions of the source amplitudes, but nonlinear in the direction of arrival (DOA) parameters. The interpolation model also applies to the narrowband case. The proposed method lends itself well to a Bayesian approach for jointly estimating the model order and the DOAs through a reversible jump Markov chain Monte Carlo (MCMC) procedure. The source amplitudes are estimated through a maximum a posteriori (MAP) process. Advantages of the proposed method include joint detection of model order and estimation of the DOA parameters, the fact that reliable performance can be obtained using significantly fewer observations than previous wideband methods, and that only real arithmetic is required. The DOA estimation performance of the proposed method is compared with the theoretical Cramer Rao lower bound (CRLB) for this problem. Simulation results demonstrate the effectiveness and robustness of the method.