Blind deconvolution in multipath environments and extensions to remote source localization

In the ocean, the acoustic signal from a remote source recorded by an underwater hydrophone array is commonly distorted by multipath propagation. Blind deconvolution is the task of determining the source signal and the impulse response from array-recorded sounds when the source signal and the environment's impulse response are both unknown. Synthetic time reversal (STR) is a passive blind deconvolution technique that relies on generic features (rays or modes) of multipath sound propagation to accomplish two remote sensing tasks. 1) It can be used to estimate the original source signal and the source-to-array impulse responses, and 2) it can be used to localize the remote source when some information is available about the acoustic environment. The performance of STR for both tasks is considered in this thesis. For the first task, simulations and underwater experiments (CAPEx09) have shown STR to be successful for 50 millisecond chirp signals with a bandwidth of 1.5 to 4.0 kHz broadcast to source-array ranges of 100 m to 500 m in 60-m-deep water. Here STR is successful when the signal-to-noise ratio is high enough, and the receiving array has sufficient aperture and element density so that conventional delay-and-sum beamforming can be used to distinguish ray-path-arrival directions. Also, an unconventional beamforming technique (frequency-difference beamforming) that manufactures frequency differences from the recorded signals has been developed. It allows STR to be successful with sparse array measurements where conventional beamforming fails. Broadband simulations and experimental data from the focused acoustic field experiment (FAF06) have been used to determine the performance of STR when combined with frequency-difference beamforming when the array elements are nearly 40 signal-center-frequency wavelengths apart. The results are good; the cross-correlation coefficient between the source-broadcast and STR-reconstructed-signal waveforms for the simulations and experiments are 98% and 91-92%, respectively. In addition, the performance of frequency-difference beamforming and conventional beamforming has been simulated for random sparse arrays. These simulation results indicate that frequency-difference beamforming can determine the array-to-source direction when conventional beamforming cannot. However, extension of the frequency-difference concept to frequency-sum beamforming does not yield a robust beamforming technique. For the source localization task, the STR-estimated impulse responses may be combined with ray-based back-propagation simulations and the environmental characteristics at the array into a computationally efficient scheme that localizes the remote sound source. These localization results from STR are less ambiguous than those obtained from conventional broadband matched field processing in the same bandwidth. However, when the frequency of the recorded signals is sufficiently low and close to modal cutoff frequencies, STR-based source localization may fail because of dispersion in the environment. For such cases, an extension of mode-based STR has been developed for sound source ranging with a vertical array in a dispersive underwater sound channel using bowhead whale calls recorded with a 12-element vertical array (Arctic 2010) . Here the root-mean-square ranging error was found to be 0.31 km from 18 calls with acoustic path lengths of 6.5 to 24.5 km.