An Efficient CS-Based Spectral Peak Search Method

Abstract

Spectral peak search is an essential part of the frequency domain parametric method. In this paper, a spectral peak search algorithm employing the principle of compressed sensing (CS) is proposed to rapidly estimate the spectral peaks. The algorithm adopts fast Fourier transform (FFT) with a few points to obtain the coarsely estimated spectral peak positions, and then only three small-scale inner products are iteratively calculated by increasing the input sequence length to rapidly refine the estimated positions. Compared with the conventional methods, this algorithm can directly capture the exact locations of spectral peaks without acquiring the entire spectrum. In addition, the proposed algorithm can be easily integrated into the existing frequency domain interpolation methods to accurately determine the spectral peak positions, and if so, only 30% of inner product operations of the original algorithms are required. Theoretical analysis and numerical results show that this algorithm yields accurate results with low complexity for analyzing both one-dimensional and two-dimensional signals.