Frequency-Undersampled Short-Time Fourier Transform

TL;DR

This paper introduces FUSTFT, a frequency-undersampled version of STFT that reduces the number of frequency components by half, aiming to improve signal compression and invertibility while maintaining analytical capabilities.

Contribution

The paper proposes FUSTFT, a novel variation of STFT that computes fewer frequency components, along with inversion methods that preserve signal properties.

Findings

FUSTFT effectively reduces spectrogram size.

Inversion methods maintain signal integrity.

Numerical examples confirm validity.

Abstract

The short-time Fourier transform (STFT) usually computes the same number of frequency components as the frame length while overlapping adjacent time frames by more than half. As a result, the number of components of a spectrogram matrix becomes more than twice the signal length, and hence STFT is hardly used for signal compression. In addition, even if we modify the spectrogram into a desired one by spectrogram-based signal processing, it is re-changed during the inversion as long as it is outside the range of STFT. In this paper, to reduce the number of components of a spectrogram while maintaining the analytical ability, we propose the frequency-undersampled STFT (FUSTFT), which computes only half the frequency components. We also present the inversions with and without the periodic condition, including their different properties. In simple numerical examples of audio signals, we confirm the validity of FUSTFT and the inversions.