Applications of Modern Spectral Estimation Techniques to Radar Data

摘要

The Modern Spectral Analysis (MSA) techniques involving linear prediction theory are reviewed and applied to radar signal processing. Specifically, the maximum entropy or forward-backward linear prediction method as implemented with Andersen's Burg algorithm is compared with the least-squares method as implemented with Marple's algorithm using as test signals autoregressive (AR) processes of 2nd and 4th orders plus single and dual sinusoids in Gaussian white noise. It is shown that Marple's indicators for terminating the AR model order iteration perform better than the more commonly employed Akike or Parzen techniques for both AR processes and noisy sinusoids. The concept is examined for using MSA to predict the AR coefficients of a clutter-dominated radar return, and in turn employing these coefficients as a FIR digital filter to suppress the clutter. Recent work on adaptive clutter filtering is reviewed. The ability of these two algorithms to resolve two closely-spaced sinusoids in a high noise environment is studied using Tranter's test signal. It is shown that model-order size rather than signal-to-noise (SNR) seems to be the dominant factor for SNR in the range of 10 to 30 dB. (Author)