FFT-based algorithms oriented to measurements on multifrequency signals

This paper reports original windows to be used in spectral analyses which require high levels of accuracy in harmonic amplitude estimation of multiple-tone signals. The proposed technique allows the user to avoid time-consuming interpolation algorithms after computing the tapered DFT of an observed stretch of time-series. This aim is pursued by using specialized cosine flat-top data windows, which perform tapering with a very low amplitude degradation (less than 0.01 dB) owing to their main-lobe flatness across a frequency range equal to the DFT bin width. The selection of this kind of data windows requires the preventive estimation of the noise floor as the wider bandwidths do not allow their use at low SNRs (). The global accuracy attained at higher SNRs is far better than with non-interpolated DFT, tapered by classical cosine windows, as bias due to leakage, which is negligible after flat-top windowing, contributes to r.m.s. error more than variance with increasing of SNRs. This may compensate for their reduced selectivity performance and makes them more suitable for calibration purposes.