High resolution 2-D spectral analysis at low SNR

Abstract

This paper presents a fully two-dimensional high-resolution spectral analysis technique. The method consists of extrapolating observed data beyond the observation window by means of 2-D least squares prediction filters. High-resolution spectral analysis is then obtained by a discrete Fourier transform of the extrapolated data. Previously, methods of obtaining a high-resolution spectrum analysis on 2-D data have applied 1-D high-resolution analyses sequentially to each dimension. The fully 2-D method presented here permits the use of nearly the square of the number of extrapolation coefficients that can be used by the sequential 1-D analyses. The advantage of the new procedure is that more coefficients permit extrapolating more 2-D sinusoids in the data and also provide reduced sensitivity to noise, thereby allowing operation at lower SNR's. A second technique for improving the performance of high-resolution spectral analysis in a chosen region of the transformed data is also presented. The technique consists of filtering the data with a 2-D FIR bandpass filter prior to the high-resolution transform. This method is also effective in reducing the sensitivity of the high-resolution analysis to noise and other sinusoids.