Asymptotic Eigenvalue Density of Noise Covariance Matrices
Ravishankar Menon, Peter Gerstoft, William S. Hodgkiss · IEEE Transactions on Signal Processing · 2012
The asymptotic eigenvalues are derived for the true noise covariance matrix (CM) and the noise sample covariance matrix (SCM) for a line array with equidistant sensors in an isotropic noise field. In this case, the CM in the frequency domain is a symmetric Toeplitz sinc matrix which has at most two distinct eigenvalues in the asymptotic limit of an infinite number of sensors. Interestingly, for line arrays with interelement spacing less than half a wavelength, the CM turns out to be rank deficient. The asymptotic eigenvalue density of the SCM is derived using random matrix theory (RMT) for all ratios of the interelement spacing to the wavelength. When the CM has two distinct eigenvalues, the eigenvalue density of the SCM separates into two distinct lobes as the number of snapshots is increased. These lobes are centered at the two distinct eigenvalues of the CM. The asymptotic results agree well with analytic solutions and simulations for arrays with a small number of sensors.