Robust high resolution spectral estimation: a combined non-parametric–parametric approach

A robust high resolution algorithm is proposed for estimating the frequency locations of narrow-band signals from a short-time measurement data with low signal to noise ratio (SNR). The algorithm is based on combining the parametric and non-parametric spectral estimation approaches. The measurement fast fourier transform (FFT) spectrum is segmented into components based on the presence of maxima. The separated portions of the spectrum are then used to construct signal spectral estimates of FFT through a low order optimal linear prediction process. Signal components are generated from these spectral estimates using inverse FFT (IFFT) and a modified singular value decomposition (SVD) based linear predictive coding algorithm (LPCA) is used to estimate their ARMA models. The desired high resolution signal spectrum is computed using the estimated ARMA model parameters. The proposed method is evaluated in terms of the spectral resolution and speed of computation when the SNR is low and the signal modes are closely spaced.