Use this URL to cite or link to this record in EThOS:
Title: Time-frequency analysis and filtering based on the short-time Fourier transform
Author: Hon, Tsz Kin
Awarding Body: King's College London (University of London)
Current Institution: King's College London (University of London)
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Access from Institution:
The joint time-frequency (TF) domain provides a convenient platform for signal analysis by involving the dimension of time in the frequency representation of a signal. A straightforward way to acquire localized knowledge about the frequency content of the signal at different times is to perform the Fourier transform over short-time intervals rather than processing the whole signal at once. The resulting TF representation is the short-time Fourier transform (STFT), which remains to date the most widely used method for the analysis of signals whose spectral content varies with time. Recent application examples of the STFT and its variants – e.g. the squared magnitude of the STFT known as the spectrogram – include signal denoising, instantaneous frequency estimation, and speech recognition. In this thesis, we first address the main limitation of the trade-off between time and frequency resolution for the TF analysis by proposing a novel adaptation procedure which properly adjusts the size of the analysis window over time. Our proposed approach achieves a high resolution TF representation, and can compare favorably with alternative time-adaptive spectrograms as well as with advanced quadratic representations. Second, we propose a new scheme for the time-frequency adaptation of the STFT in order to automatically determine the size and the phase of the analysis window at each time and frequency instant. This way, we can further improve the resolution of the conventional as well as the time-adaptive spectrograms. Finally, we focus on denoising non-stationary signals in the STFT domain. We introduced an optimized TF mask in the STFT domain, which is based on the concept of the multi-window spectrogram. Experimentation has shown that the introduced approach can effectively recover distorted signals based on a small set of representative examples of the noisy observation and the desired signal.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available