Speed comparison among methods for restoring signals with missing high-frequency components using two different low-pass-filter matrix dimensions

An arithmetic operations (multiplications and divisions) count is presented for three recent algorithms that restore signals with missing high-frequency components. The cases when the discrete Fourier-transform (DFT) low-pass-filter matrices had dimensions (a) N by N and (b) N by L were studied, where N is the whole signal length and L is the length of its known part. We show that when N is large all these algorithms are 2 orders of magnitude slower for case (a) than for case (b). We also suggest a method for calculating the rank of the DFT low-pass-filter matrix for case (b) that gives more-accurate results than those previously published.Peer reviewed: YesNRC publication: Ye