High-Resolution Frequency Measurement by Linear Prediction

Abstract

The efficacy of linear prediction spectrum analysis (LPSA) (also called autoregressive or maximum entrophy spectrum analysis) for problems concerning narrowband signals in noise is examined by means of Monte Carlo simulation. For the case of pure tones (sinusoids) in white noise, both frequency estimation accuracy and resolution of two closely spaced tones with the linear prediction approach are discussed. In addition, estimation of the carrier frequency of a narrowband random (fading) Gaussian signal in noise is considered. The results are compared with standard fast Fourier transform (FFT) methods. These comparisons indicate that there is no significant advantage to using the linear prediction approach with real (bandpass) data. It is concluded that these experiments should be repeated with complex (baseband) data before a complete assessment of the LPS approach can be made.