Performance analysis of MUSIC method for DOA estimation

Direction-of-arrival estimation plays an essential role in many signal processing applications and MUSIC is one of the most popular techniques for DOA estimation. A crucial problem in the MUSIC method is to estimate the eigensystem of the covariance matrix of observed signals from the sensor array. This paper presents the theoretical description of the behavior of the eigen-structure of the estimated covariance matrix in a noisy environment, especially, in the presence of an "outlier". An analytical approach to the perturbation of the estimated signal subspace on the final estimates of DOA is developed in this paper which exploits the high dependency of the DOA estimates on the signal eigenvectors. The resulting conclusion that the MUSIC method is highly sensitive to outliers demonstrates the importance of robust approach for the MUSIC method. A new robust singular value decomposition (RSVD) algorithm is incorporated into the MUSIC method to reduce the sensitivity of the final DOA estimates to outlying observations. Finally, numerical simulation results are presented to illustrate progress made in performance of the MUSIC by the new robust approach.