DFT preprocessing for high-resolution frequency estimation

This work considers the use of the DFT of the autocorrelation (AC) matrix (DFT-of-AC) for extracting the signal and noise subspaces. It is shown that when DFT preprocessed data is incorporated within the frameworks of minimum-norm (MNM) or Prony's methods, improved high-resolution estimates are obtained. Furthermore, if the signal-subspace part of the DFT-of-AC vectors are used in place of eigenvectors in MNM, the high-resolution performance is further enhanced. Theoretical perturbation analysis of the DFT-based MNM (D-MNM) shows that the estimates are unbiased and, furthermore the theoretical mean-squared error results indicate improved high-resolution performance, especially at low SNR.<>