Qualitative Assessment of Certain Spectral Analysis Methods

Abstract

Parametric methods of spectral analysis play an important role when solving the problems of primary radar data processing. The paper mainly focuses on comparing methods to obtain a spectrum by means of two criteria, i.e. computational complexity and resolution. The latter represents the most important characteristic of the algorithms used to process a spectrum of the reflected radar signal. The paper considers mathematical foundations of parametric methods, which use white noise as an excitation sequence, including the models: an autoregressive, moving average model and an autoregressive-moving average one. Reveals their differences from the classical approach to the spectral analysis represented by periodogram and correlogram methods. Emphasizes the greatest popularity of the autoregressive model due to linearity of dependencies used in it. Shows the main advantages and disadvantages of a number of autoregressive methods. Considers non-classical methods based on the analysis of eigenvalues. An experimental study of the spectral analysis methods in the MATLAB environment has been carried out for to compare them in terms of computational complexity and resolution. The simulation results with the input signal, represented by the superposition of harmonics on white noise, are shown. The character of the spectra peaks obtained and the order of each model, desirable for their resolution, are estimated. An analysis of the results obtained has shown that among the methods considered, those of based on the analysis of eigenvalues possess the best resolution and the greatest computational complexity. The autoregressive parametric methods have revealed similar properties. Classical methods are characterized by the least computational complexity and the least accuracy. A conclusion has been that the spectrum quality depends on the order of the model used and on the input signal noisiness. The results obtained can be used when solving the problems of primary radar data processing to provide a reasonable choice of spectral analysis methods based on resolution and computational complexity.