SPECTRAL ESTIMATION: FROM CONVENTIONAL METHODS TO HIGH-RESOLUTION MODELING METHODS.

Spectral analysis forms the basis of a major part of signal processing, typically for distinguishing and tracking signals of interest, and for extracting information from the relevant data. Given a finite number of noisy measurements of a discrete-time stochastic process, or its first few covariance lags, the classical spectral estimation problem was that of estimating the shape of its continuous power spectrum. For arrays, the spectrum of interest is a line spectrum, and the modern spectral estimation problem is that of estimating the locations of these spectral lines. In this chapter we first discuss the conventional methods for power spectrum estimation. Then in the later sections we focus our attention on the modern methods, with special emphasis on those modeling methods that are suited for resolving narrowband signals.