I. Introduction
Under frequency deviation, oscillation, and ramp conditions, frequencies of different harmonics will have different deviations from the nominal values. For off-nominal frequencies, the finite-impulse response (FIR) filter notches related to the nominal frequency harmonics no longer correspond exactly to the unwanted frequencies in the mixed signal. Therefore, the ability of the FIR filter to reject the harmonic contamination reduces as the frequency diverges from nominal. As a result, measurements of different order harmonics need estimators with different widths of passband and/or stopband. However, most of the well-known standard methods, for example, discrete Fourier transform (DFT), do not have such ability. It should be noted that some techniques, such as the discrete Taylor–Fourier transform (TFT) [1] and its multiple-resonator (MR)-based filtering implementation [2], [3], were proposed as an extension of the full DFT. Multiple zeros provide reinforcing of the required attenuations at the harmonic components and more overall attenuation in the stopband. However, the attenuation bands around the harmonic frequencies are still with equal widths for all harmonic orders, which is not convenient for the harmonic analyses of the signals with off-nominal frequency. An issue of the off-nominal-frequency harmonics estimation is a subject of researches for the last decades. As an example, in [4], performances of the dynamic harmonic synchrophasor estimator on the basis of the sin function interpolation have been improved by selecting different values of the prescribed input parameters, for different harmonics. This way, with the increase in the harmonic order, the corresponding estimator’s passband (or stopband) width also increases.