A statistical approach for tuning the windowed fourier transform

Abstract

A time frequency analysis is used in many fields for studying signals with a time-varying spectral content. The windowed Fourier transform is one of the most used time-frequency representations. In order to use this technique several parameters must be defined, including the type, the length and the overlap of the windows. For tuning the windowed Fourier transform a new method based on the information theory is presented. Several tests with robotic signals illustratethe appropriateness of the proposed method.