Abstract: Approximated maximum likelihood estimator (AML) has been shown to be the best performance in short sampling wideband sources DOA estimation. However, the computation burden of AML is very large. In order to lighten computation burden, Markov Monte Carlo (MCMC) methods are combined with approximated maximum likelihood DOA estimator. A novel approximated maximum likelihood DOA estimator based on perfect sampling (PAML) is proposed. PAML regards the power of the AML spectrum function as the target distribution up to a constant proportionality, and uses perfect sampler to sample from it. The performance of AML and genetic algorithm (GA) are compared with PAML through computer simulations. The simulations show that on the condition of two sources, the PAML provides similar performance to those achieved by the AML and GA in low and mid SNR ranges, but its computational cost is only 1/24 of AML or 1/3 of GA. With the number of sources rising, the advantage of computational reduction of PAML is even more prominent.

Key words: wideband source, short sampling, approximated maximum likelihood estimator, perfect sampling, computational complexity, DOA estimation