Comparative performance of eigenvector rotation and MUSIC algorithms for angle-of-arrival estimation

A simulation study of the statistical properties of two eigenvector-based methods for angle-of-arrival estimation is presented. The two methods investigated are the multiple signal classification (MUSIC) method developed by R.O. Schmidt (1986) and the eigenvector rotation (ER) method developed by G. Vezzosi (1982) and further pursued by Farrier and Jeffries. The results demonstrate (a) that for large sample size both MUSIC and ER estimates approach the normal distribution with zero mean, (b) that the ER method generally performs better than MUSIC with respect to bias and variance at low signal-to-noise ratios and/or small sample size, and (c) that ER (or augmented ER, where pertinent) is more robust than MUSIC with regard to sensor position errors and source correlations