Robust Auto-Regressive Spectrum using a Reiterative Median Cascaded Canceller

Abstract

Auto-regressive (AR) models are used to form temporal and/or spatial super-resolution spectra for source signal detection and estimation. An AR spectrum is considered a super-resolution technique that can distinguish signal frequencies or angular locations with higher resolution, and often using many fewer data samples, as compared to Fourier spectral techniques. This paper presents a novel method to form a robust AR spectrum by exploiting the reiterative median cascaded canceller (RMCC) algorithm. The result is a robust estimate of a linear prediction weight vector and its corresponding AR spectrum. In addition, by utilizing the spectral estimates for each iteration of the RMCC, the non stationary spurious peaks typical of AR spectra are reduced significantly. We note that no additional training data is required to form the multiple spectral estimates in this technique.