Spectrum estimation by iterative minimization of the I-divergence measure

A method for spectrum estimation (either narrowband or broadband) based on minimization of Csiszar's (1975) I-divergence measure is introduced. The blurring effect of the observation window (or antenna array) is minimized by applying a nonlinear deconvolution procedure. This algorithm has subsequently been shown to minimize the I-divergence between the spectrum and its estimate subject to a non-negativity constraint. Here the method is applied to the spectrum estimation problem for stationary processes. A reblurring procedure is used to regularize the method. Simulations show that the method offers error performance comparable to that of MUSIC, although the resolution performance is inferior. The method is iterative, allowing a tradeoff between resolution and error performance, and can be implemented using fast FFT hardware.<
>