A robust frequency domain subspace algorithm for multi-component harmonic retrieval

Multi-component harmonic retrieval finds applications in many fields, including radar, sonar and wireless communications. In previous years, several high-resolution subspace-based techniques have been proposed to overcome the spectral leakage problems inherent in the conventional periodogram. However, these methods require high SNR and/or white noise, and are highly computationally intensive in scenarios with a large number of spectral components. We propose a frequency domain subspace algorithm which overcomes these limitations. The method is computationally efficient in providing estimates of the frequencies within a user-selected sub-band. The estimates are exact in the noise-free case, and exhibit low sensitivity to out-of-band signals and colored noise.