Acquiring accurate and reliable target information over large areas is crucial for various underwater applications. This can be achieved by deploying underwater wireless sensor networks (UWSN), in which spatially distributed sensor nodes detect targets and communicate the acquired information. Traditionally, sensing and communication are performed separately, requiring dedicated waveforms, hardware, and signal-processing resources. Given the limited resources available in underwater systems, this separation can be inefficient. To reduce this inefficiency, an integrated sensing and communication (ISAC) framework enables shared waveform and system resources to be used for both sensing and communication. However, sharing these resources does not remove the effects of the underwater propagation environment on the received signals.
Appropriate modeling of underwater acoustic propagation is also important for reliable target detection, localization, and tracking. In shallow-water active sonar systems, the received signal may contain the target-reflected signal, the direct-path signal, and reflections from the sea surface and seabed, along with ambient noise. These multipath components vary over time and can mask weak target reflections, motivating models that account for their structure and temporal variation. Therefore, this thesis investigates how modeling underwater acoustic propagation, and particularly the time-varying propagation channel, can support target localization, detection, and tracking.
First, a simplified channel model is used to investigate the feasibility of communication-centric signals for underwater target localization in a bistatic sonar setup. The localization performance is evaluated using the Cramér-Rao lower bound (CRLB), while the delay and Doppler resolution characteristics are assessed using the wideband ambiguity function. The results show that combining bistatic target-reflected information with passive bearing measurements improves the theoretical localization accuracy without requiring an additional sensing transmission. Under the considered conditions, the localization bounds are comparable to those obtained using a conventional linear frequency-modulated (LFM) waveform.
Second, a wideband Doppler linearization method is used to obtain a statistical model of the time-varying multipath channel. The approximate model enables recursive tracking of the structured multipath components using an extended Kalman filter. BELLHOP-based evaluations show that explicitly modeling and tracking the time-varying multipath channel improves detection performance.
Finally, the inferred channel statistics are used to approximate the raw sensor measurement likelihood, enabling computationally efficient target tracking within a Bernoulli track-before-detect framework. Evaluations using datasets representing both model-matched and model-mismatched conditions show improved target-confirmation and localization performance compared with constant-false-alarm-rate (CFAR)-based tracking approaches.