A Comparative Analysis of ML-based DOA Estimators

Abstract

This paper compares different directions of arrival estimation techniques relying on the maximum likelihood (ML) principle. The considered algorithms are well-known, but to the best of authors' knowledge a thorough comparison has not be performed in the open literature. This is especially true for the unconditional maximum likelihood (UML) and the asymptotic maximum likelihood (AML) algorithms. The UML is the ML estimator when source signals are sample functions drawn from zero-mean complex Gaussian processes and the noise is a white Gaussian process; on the other hand, the AML is an approximated ML that tends to the UML as the number of snapshots $K$ increases towards infinity. In [1], it is stated that for a uniform linear array with $N$ elements the performance of the two algorithms are essentially the same as $K\geq 10N$. However, we will show that the two algorithms may have different performance for lower sample supports.