Comparison of performance bounds for DOA estimation
Harry L. Van Trees, Kristine L. Bell · 2009
For a single signal, numerical comparison of the Bayesian Cramer-Rao lower bound, the Chazan-Ziv-Zakai lower bound and the Weiss-Weinstein lower bound on DOA estimation MSE shows that the CZZLB is the tightest lower bound and can accurately predict the threshold SNR, which is a critical system design parameter. The analysis is extended to two signals where the multiple parameter WWLB is applicable but appears to be a weak lower bound. Simulation results show that below the threshold SNR, a maximum likelihood DOA estimation procedure such as the Expectation Maximization(EM) algorithm seeded by the Alternating Projection Maximization (APM) algorithm should be used to provide the best DOA estimation performance. Above the threshold SNR, the use of the statistically efficient MUSIC algorithm is adequate.