Dynamical State-Space Modeling And Processing For Networked Ocean Acoustic Tomography

There exist complex movements and variabilities in the ocean. It is of great significance to have a better understanding about the movement in the ocean. Hence, it is necessary to monitor the ocean with enough resolution as well coverage in both time and space. Effective measurement techniques for ocean temperature and current distribution have been of great interest in ocean observations. As the main information carrier for ocean remote-sensing, acoustic wave can be used to monitor the ocean dynamic state in a large scale, long du-ration and real time. The ocean environmental parameters are essentially dynamical given the underlying ocean physics, and can thus be tracked by exploring the sequential filtering concept. The key problem is how to construct an accurate and adaptive evolution model. In this thesis, we study the Bayesian tracking of two-dimensional ocean current field as an example, aiming to develop the theory and methods of dynamical state-space modeling and signal processing for ocean measurements, under the framework of distributed underwater acoustic network.Ocean Acoustic Tomography (OAT) is used to infer the ocean state from the pre-cise measurement of acoustic travel time or other acoustic properties, often with transmis-sion/receiving nodes on the peripheral of the observation region, providing the possibility to observe small to mesoscale processes of the ocean. As an application of OAT in the coastal region, Coastal Acoustic Tomography (CAT) was intended to monitor spatial structure of sound speed or current velocity in a semi-enclosed sea. Considering the correlation of the current velocity in space and time domain, a CAT tracking method based on the spatiotem-poral evolution model is proposed in the thesis. The current variation is modeled as an autoregressive (AR) process, and the AR coefficients are adaptively updated using the past and current estimated velocities. The results show that the proposed method is robust to the measurement jitter, and can handle the cases of current variation with different evolution rates.Ocean monitoring often has to face the trade-off between resolution and distance, spa-tiotemporal scale and resources. Underwater Acoustic Sensor Network (UASN) provides a new approach to handle those issues. A UASN consists of a large number of low-power, com-pact sensor nodes that are deployed in a specific area to perform a collaborative monitoring task, achieving resolution via node distribution density and coverage via distribution scope. The estimation of parameter field in UASN is regarded as a realization of OAT in the frame-work of distributed network. Hence, the UASN for acoustic tomography can be named as Distributed Acoustic Tomography Network (DATN). In the thesis, based on the triangular gridding, a spatiotemporal current tracking method is proposed. Specifically, the observa-tion region is divided into sub-triangle grids; current variation within each sub-triangle is modeled as a spatiotemporal evolution process. The results show that the proposed method improves the estimation performance. Moreover, it is robust against burst errors and link failures, which are common in the underwater sound channel.The limitation of underwater communication resources makes it quite challenging to fuse data in a DATN. For the decentralized processing, messages are exchanged between neighboring nodes and the estimation is executed at each node. Since the parameter esti-mation does not rely on any centre node, it is environmentally robust and scalable. In this thesis, a consensus-innovation based method is proposed to track the current variation in a decentralized manner. The proposed method is derived from the information theoretical form of Kalman filter. In the proposed method, the averaged pseudo-measurement is pro-cessed by consensus and innovation. Message exchange makes the estimates reach consensus, and local measurements provide the innovation which is used to update the prediction. To improve the convergence rate, a method of constructing the consensus weighting matrix is also proposed based on routing selection. Moreover, the unbiasness and the condition of convergence are proved. The results show that the proposed method has a faster conver-gence rate with a comparatively small amount of communication cost, which is suitable for underwater environments with large time delay.The performances of the proposed methods are validated with simulations and/or exper-iments. Through this thesis research, it is expected to enrich the OAT technique by applying the spatiotemporal correlation of the current field to improve the estimation accuracy, to develop the parameter estimation method using DATN, which provides a new way in sensing large-scale and high-resolution current field, and to make significant progresses on network sensing theory, which will form a solid foundation for future applications of the underwater acoustic sensing network in practical ocean environmental observations.