Calibrating and Beamforming Distributed Arrays in Passive Sonar Environments
This dissertation presents methods for calibrating and beamforming a distributed array for detecting and localizing sources of interest using passive sonar. Passive sonar is critical for underwater acoustic surveillance, marine life tracking, and environmental monitoring but is increasingly difficult with greater shipping traffic and other man-made noise sources. Large aperture hydrophone arrays are needed to suppress these sources of interference and find weak targets of interest. Traditionally, large hydrophone arrays are densely sampled uniform arrays which are expensive and time-consuming to deploy and maintain. There is growing interest instead in forming distributed arrays out of low-cost, individually small arrays which are coherently processed to achieve high gain and resolution. Conventional array processing methods are not well suited to this end and this dissertation develops new methods for array calibration and beamforming which ultimately enable high resolution passive sonar at low-cost. This work develops estimation methods for array parameters in uncalibrated, unsynchronized collections of acoustic sensors and also develops adaptive beamforming techniques on such arrays in complex and uncertain ocean environments.
Methods for estimating sampling rate offset (SRO) are developed using a single narrowband source of opportunity whose parameters need not be estimated. A search-free method which jointly estimates all SRO parameters in an acoustic sensor network is presented and shown to improve as the network size increases. A second SRO estimation method is developed for unsynchronized sub-arrays to enable SRO estimation with a source that has a bearing rate. This is of particular value in ocean environments where transiting cargo ships are the most prevalent calibration sources.
Next, a technique for continuously estimating multiple sub-array positions using a single, tonal moving source is presented. Identical, well-calibrated sub-arrays with unknown relative positions exhibit a rotational invariance in the signal structure which is exploited to blindly estimate the inter-array spatial wavefronts. These wavefront measurements are used in an Unscented Kalman Filter (UKF) to continuously improve sub-array position estimates.
Lastly, this work addresses adaptive beamforming in uncertain, complex propagation environments where the modeled wavefronts will inevitably not match the true wavefronts. Adaptive beamforming techniques are developed which maintain gain even with significant signal mismatch due to unknown or uncertain source wavefronts by estimating a target-free covariance matrix from the received data and using just a single gain constraint in the beamformer optimization. Target-free covariances are estimated using an eigendecomposition of the received data and assuming that modes which potentially contain sources of interest can be identified. This method is applied to a distributed array where only part of the array wavefront is explicitly modeled and shown to improve interference suppression and the output signal-to-interference-plus-noise ratio (SINR).
This idea is then extended to realistic environments and a method for finding potential target components is developed. Blind source separation (BSS) methods using second-order statistics are adopted for wideband source separation in shallow-water environments. BSS components are associated with either target or interference based on their received temporal spectra and are automatically labeled with a convolutional neural network (CNN). This method is applicable when sources have overlapping but distinct transmitted spectra, but also when the channel itself colors the received spectra due to range-dependent frequency-selective fading. Simulations in realistic shallow-water environments demonstrate the ability to blindly separate and label uncorrelated components based on frequency-selective fading patterns. These simulations then validate the robustness of the developed wavefront adaptive sensing (WAS) beamformer compared to a standard minimum variance distortionless response (MVDR) beamformer. Finally, this method is demonstrated using real shallow-water data from the SWellEx96 S59 experiment off the coast of Southern California. A simulated target is injected into this data and masked behind a loud towed source. It is shown that the WAS beamformer is able to suppress the towed source and achieve an target output SINR which is close to that of the optimal beamformer.
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License.
Rights for Collection: Duke Dissertations
Works are deposited here by their authors, and represent their research and opinions, not that of Duke University. Some materials and descriptions may include offensive content. More info