Time-frequency analysis based on split spectrum applied to audio and ultrasonic signals

Abstract

Signal processing is a large subject with applications integral to a number of technological fields such as communication, audio, Voice over IP (VoIP), pattern recognition, sonar, radar, ultrasound and medical imaging. Techniques exist for the analysis, modelling, extraction, recognition and synthesis of signals of interest. The focus of this thesis is signal processing for acoustics (both sonic and ultrasonic). In the applications examined, signals of interest are usually incomplete, distorted and/or noisy. Therefore, reconstructing the signal, noise reduction and removal of any distortion/interference are the main goals of the signal processing techniques presented. The primary aim is to study and develop an advanced time-frequency signal processing technique for acoustic applications to enhance the quality of the signals. In the first part of the thesis, a technique is presented that models and maintains the correlation between temporal and spectral parameters of audio signals. A novel Packet Loss Concealment (PLC) method is developed with applications to VoIP, audio broadcasting, and streaming. The problem of modelling the time-varying frequency spectrum in the context of PLC is addressed, and a novel solution is proposed for tracking and using the temporal motion of spectral flow to reconstruct the signal. The proposed method utilises a Time-Frequency Motion (TFM) matrix representation of the audio signal, where each frequency is tagged with a motion vector estimate that is assessed by cross-correlation of the movement of spectral energy within sub-bands across time frames. The missing packets are estimated using extrapolation or interpolation algorithms using a TFM matrix and then inverse transformed to the time-domain for reconstruction of the signal. The proposed method is compared with conventional approaches using objective Performance Evaluation of Speech Quality (PESQ), and subjective Mean Opinion Scores (MOS) in a range of packet loss from 5% to 20%. The evaluation results demonstrate that the proposed algorithm substantially improves performance by an average of 2.85% and 5.9% in terms of PESQ and MOS respectively. In the second part of the thesis, the proposed method is extended and modified to address challenges of excessive coherent noise arising from ultrasonic signals gathered during Guided Wave Testing (GWT). It is an advanced Non-destructive testing technique which is used over several branches of industry to inspect large structures for defects where the structural integrity is of concern. In such systems, signal interpretation can often be challenging due to the multi-modal and dispersive propagation of Ultrasonic Guided Waves (UGWs). The multi-modal and dispersive nature of the received signals hampers the ability to detect defects in a given structure. The Split-Spectrum Processing (SSP) method with application for such signal has been studied and reviewed quantitatively to measure the enhancement in terms of Signal-to-Noise Ratio (SNR) and spatial resolution. In this thesis, the influence of SSP filter bank parameters on these signals is studied and optimised to improve SNR and spatial resolution considerably. The proposed method is compared analytically and experimentally with conventional approaches. The proposed SSP algorithm substantially improves SNR by an average of 30dB. The conclusions reached in this thesis will contribute to the progression of the GWT technique through considerable improvement in defect detection capability.