In this paper, we propose to use the phase information of the Short Time Fourier Transform (STFT) to improve a time-frequency segmentation based on the statistical features of the STFT, and proposed by the authors in 2006. If the resolution of the STFT is too low, close components may be segmented in a single pattern. The idea is to add phase information provided by the reassignment principle in order to determine if there are more than one component in a pattern instead of two. Reassignment, originally proposed by Kodera et al. in 1976, is a non-linear method which creates a new time-frequency representation by moving the spectrogram values away from their computation place. Reassignment focuses energy components by moving each time-frequency location to its group delay and instantaneous frequency that represent more accurately the component energy. The obtained reassignment vector field associated to a given spectrogram describes how time-frequency locations are reassigned. We propose to use the reassignment vector field not to modify the time-frequency representation, but to give information on the signal structure. We compute local reassignment vectors on patterns segmented by the method mentioned above. Given that spectrogram tends to spread the time-frequency patterns, whereas reassignment method moves back energy to a pattern's point, reassignment vectors aim at the pattern. That leads to a pattern's boundary information, which is used to determine how many components are embedded in a single segmented pattern. Moreover, this information describes the boundaries of frequency modulations as well as wide band signal, and extends the use of the reassignment principle to wide band signals. This principle is finally applied to the shaft's vibrations of a three phase AC induction engine, in order to separate the different harmonics embedded in a single pattern.
In order to set up a list of libraries that you have access to,
you must first login
or sign up.
Then set up a personal list of libraries from your profile page by
clicking on your user name at the top right of any screen.