Machine learning inspired ship-radiated noise modelling and cancellation for underwater acoustic communication systems

Achieving high data rate and reliable communication in shallow and harbour underwater acoustic (UA) environments can be a demanding task in the presence of ship-radiated noise. However, few research studies have examined the properties of ship-radiated noise in terms of its time-domain statistical characteristics and its negative effects on UA communication systems. From the observation of spectrograms and the temporal signals of various acoustic shipping noise recordings, high frequency and impulsive characteristics are visible. These impulsive agitations can be detrimental to the performance of multi-carrier UA communication systems, thus impulse noise cancellation methods are necessary to reduce errors. In this thesis, we investigate the impulsive and correlative interference generated due to nearby shipping activity and its effects on orthogonal frequency-division multiplexing (OFDM) systems. The research objectives are twofold: (1) model the time-domain stochastic characteristics of ship-radiated noise, and (2) achieve shipping noise cancellation for UA OFDM systems. We propose the use of unsupervised learning techniques to train generative models that capture the time-domain stochastic behaviours of ship-radiated noise using a publicly available database of long-term acoustic shipping noise recordings. These models can then be used for further analysis of ship-radiated noise and performance evaluation of UA OFDM systems in the presence of such interference. The results indicate a two component Gaussian mixture model serves as a better approximation for high frequency ship-radiated noise while generative adversarial networks produce improved realizations of shipping noise in lower frequencies. We offer sparsity and deep learning-based ship-radiated noise cancellation solutions that are constructed under a compressed sensing framework. Obtained results show that the sparsity-based estimation and cancellation algorithms demonstrate competitive mitigation capabilities for high frequency impulsive ship-radiated noise. The deep learning-based cancellation methods depict measurable shipping noise mitigation results to the sparsity-based techniques, but with superior run-time performance. In addition, the deep learning-based methods outperform the sparsity-based approaches in lower frequency ship-radiated noise due to the supplementary correlative structure. Furthermore, experimental results indicate the deep learning-based cancellation approaches scale better to new realizations of high frequency and low frequency shipping noise signals compared to the sparsity-based methods.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat