The Effects of Window and Delay Selection on the Spectrogram of a Signal.

The power spectral density of steady-state waveforms can be readily computed by the classical periodogram (perhaps smoothed) of a sampled waveform. In the case of quasistationary signals, many times the short-time Fourier transform is used whose magnitude is displayed in a time-frequency spectral density, referred to as spectrogram in speech processing. This paper is a study of how the parameters for digitally performing the short-time Fourier transform of several known signals affect the resolution in both the time and frequency domains. The parameters which are investigated included window type, window length, and incremental delay of the sliding window. A comparison between the resulting spectrograms and the actual spectral density leads to an understanding of the parameter selections and the trade-offs necessary to achieve desirable results. (Author)