ALBERT

Description: Submitted in partial fulfillment of the requirements for the degree of Ocean Engineer at the Massachusetts Institute of Technology and Woods Hole Oceanographic Institution January 1994

Description: In 1991 the Heard Island Feasibility Test demonstrated that it is possible to transmit coded acoustic signals nearly half way around the world. One of the key issues in the feasibility test was to determine the spatial structure of the received transmissions. In this thesis, data from the Canadian Defense Research Establishment Pacific horizontal line array is used to form an estimate of the directional power spectrum. This spectrum determines if any horizontal multipath is detectable. The preliminary signal conditioning, including frequency spectrum estimation and demodulation required before beamforming is described. Conventional and adaptive beamforming methods are examined with synthetic data to demonstrate the limitations on the directional spectrum results. The principle result of this work is that no stable horizontal multipath is evident. The mean arrival angle for the five hours of data analyzed is 212° ± 1.5°.

Description: The Office of Naval Research provided funding for the author under the ONR fellowship program.

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and Woods Hole Oceanographic Institution September 1994

Description: Normal mode theory provides an efficient description of signals which propagate axially in the SOFAR channel and are detectable at long ranges. Mode amplitudes and their second order statistics are useful in studies of long-range acoustic propagation and for applications such as Matched Mode Processing (MMP) and Matched Field Tomography (MFT). The purpose of this research is to investigate techniques for estimating the average power in the modes of a signal given pressure measurements from a vertical line array. This thesis develops the problem of mode estimation within a general array processing framework which includes both deterministic and stochastic characterizations of the modal structure. A review of conventional modal beamforming indicates that these methods provide poor resolution in low signal-to-noise ratio environments. This is not surprising since standard estimation techniques rely on minimizing a squared error criterion without regard to the ambient noise. The primary contribution of this thesis is an adaptive estimator for coherent modes that is based on a method suggested by Ferrara and Parks for array processing using diversely-polarized antennas. Two formulations of the adaptive method are investigated using a combination of analytical techniques and numerical simulations. The performance evaluation considers the following issues: (i) power level of the noise, (ii) orthogonality of the sampled modeshapes, (iii) number of data snapshots, and (iv) coherence of the signal. The new approach is fundamentally different from other modal estimators such as those used in MMP because it is data-adaptive and maximizes the received power instead of minimizing the squared error. As a result, the new methods perform significantly better than least squares in high noise environments. Specifically, the Ferrara/Parks formulations are able to maintain nulls in the modal spectrum since they do not suffer the bias error that significantly affects the least squares processor. A second contribution of the thesis is an extension of the coherent estimator to facilitate estimation of phase-randomized modes. Although the results of this work are preliminary, the extended formulation appears to offer several advantages over least squares in certain cases.

Description: Funding for this work was provided by a Clare Boothe Luce Fellowship (1991-1992) and the University of California-Scripps ATOC Agreement, P0# 10037359.

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution September 1999

Description: An algorithm is developed for underwater acoustic signal processing with an array of hydrophones. With various acoustic signals coming from different directions, the maximum likelihood approach is used to estimate the source bearings and time series. Simulated annealing is used to implement the resulting time-domain beamformer. Broadband signals in spatially correlated noise are treated. Previous time-domain beamformers did not consider the correlation between random noise, and they did not use the concept of maximum likelihood, which is asymptotically optimal. We show that improved resolution can be achieved using this new method.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution June 1997

Description: This work is concerned with coherent communication by means of acoustic signals over underwater communication channels. The estimated scattering functions of real data ranging from the Arctic environment to tropical waters show that underwater communication channels can not be captured by a single, simple channel model. This thesis considers mainly a subset of underwater communication channels where the Doppler spread is more severe than the delay spread. An appropriate representation of the linear time- variant channel is introduced, and the wide sense stationary uncorrelated scattering (WSSUS) channel assumption enables characterization in terms of scattering functions. The concept of Doppler lines, which are frequency domain filters, is used in the derivation of a receiver for Doppler spread channels. The channel is simulated by means of a ray representation for the acoustic field and a time-variant FIR filter. The impact of physical ocean processes on the Doppler spread is demonstrated, and from this modeling explanations for the Doppler spread observed in real data are obtained. A decision feedback equalizer (DFE) adapted with recursive least squares (RLS) is analyzed, and its limit with respect to pure Doppler spread is found. By using the DFE with a phase locked loop (PLL) suboptimal system behavior is found, and this is verified on real data. In the case of a simple Doppler shift the cross-ambiguity function is used to estimate the shift, and the received signal is phase rotated to compensate this before it enters the receiver. A modified RLS called the time updated RLS (TU-RLS) is presented, and it is used in a new receiver. This receiver is initialized by means of the cross-ambiguity function and the performance is characterized by probability of decoding error vs delay spread, Doppler spread and SNR. The receiver uses Doppler lines to compensate both discrete and continuous Doppler spread. The receiver stability depends on the conditioning of the block diagonal correlation matrix propagated by the TU-RLS. The receiver is used to decode both real and simulated data, and some of these data are severely Doppler spread.

Description: The work in this thesis was funded in part by the Norwegian Research Counsil under a general PhD contract, the education office of Woods Hole Oceanographic Institution and the Office of Naval Research under contracts N00014-95-0153 and N00014-95-l-0322.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution May 1998

Description: Planktonic protozoan grazers have the potential to significantly affect the chemistry of particle-associated trace metals. This is due both to the importance of protists as consumers of bacterial-sized particles, and to the unique low-pH, enzyme-rich microenvironment of the grazer food vacuole. This thesis examines the role of protozoan grazers in the marine geochemistry of strongly hydrolyzed, particle-reactive trace metals, in particular Th and Fe. A series of tracer experiments was carried out in model systems in order to determine the effect of grazer-mediated transformations on the chemical speciation and partitioning of radioisotopes C9Fe, 234Th, 51Cr) associated with prey cells. Results indicate that protozoan grazers are equally able to mobilize intracellular and extracellular trace metals. In some cases, protozoan regeneration of trace metals appears to lead to the formation of metal-organic complexes. Protozoan grazing may generate colloidal material that can scavenge trace metals and, via aggregation, lead to an increase in the metal/organic carbon ratio of aggregated particles. Model system experiments were also conducted in order to determine the effect of grazers on mineral phases, specifically colloidal iron oxide (ferrihydrite). Several independent techniques were employed, including size fractionation ors9Fe-labeled colloids, competitive ligand exchange, and iron-limited diatoms as "probes" for bioavailable Fe. Experimental evidence strongly suggests that protozoan grazing can affect the surface chemistry and increase the dissolution rate of iron oxide phases through phagotrophic ingestion. In further work on protozoan-mediated dissolution of colloidal Fe oxides, a novel tracer technique was developed based on the synthesis of colloidal ferrihydrite impregnated with 133Ba as an inert tracer. This technique was shown to be a sensitive, quantitative indicator for the extent of ferrihydrite dissolution/alteration by a variety of mechanisms, including photochemical reduction and ligand-mediated dissolution. In field experiments using this technique, grazing by naturally occuring protistan assemblages was shown to significantly enhance the dissolution rate of colloidal ferrihydrite over that in non-grazing controls. Laboratory and field results indicate that, when integrated temporally over the entire euphotic zone, protozoan grazing may equal or exceed photoreduction as a pathway for the dissolution of iron oxides.

Description: This work was financially supported by a Department of Defense ONR-NDSEG Graduate Fellowship, Office ofNaval Research AASERT Award (N00014-94-1-0711), and the National Science Foundation EGB Program (OCE-9523910).

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution February 1998

Description: In practical applications with bathymetric sidescan sonars, the multipath reflections and other directional interferences are the key limiting factors for a better performance. This thesis proposes a new scheme to deal with the interferences using a multiple-row bathymetric sidescan sonar. Instead of smoothing the measurements over some time or angle intervals, which was previously widely investigated, we resolve the multipath interferences from the direct signal. Two approaches on signal direction-of-arrival DOA and amplitude estimation are developed, the correlated signal direction estimate CSDE for three-row systems and the ESPRIT-based method. These approaches are compared using different sonar data models, including a stochastic model from the statistical analysis on bottom scattering and a coherent model from the analysis on interference field; the simulations show the ESPRIT-based approach is quite robust at the angular separation of 100 between two sources and at the signal-to-noise ratio above 10dB except for highly coherent or temporally correlated signals, for which CSDE works very well. The computer simulation results and the discussions on practical algorithm implementation indicate the proposed scheme can be applied to a real multiple-row bathymetric sidescan sonar. With the capability to simultaneously resolve two or more directional signals, the new sonar model should work better for a wider variety of practical situations in shallow water with out significant increase of the system cost.

Description: Funding supporting my thesis research project was provided by the Office of Naval Research ONR.

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and Woods Hole Oceanographic Institution May 1994

Description: During the 1991 Heard Island Feasibility Test, a vertical hydrophone array deployed off Monterey, CA, recorded transmissions from a low-frequency acoustic source nearly 18,000 km away. By determining the modal structure of the received transmissions, it is possible to characterize the physics of such long range propagation. This thesis focuses on the determination of the modal, or vertical, structure of the signal. It was necessary to first develop a conditioning scheme to address several data quality issues, including very low signal levels (-15 dB SNR on a single channel), large transient spikes, and a limited set of operational channels. Very narrowband filtering was used to obtain a 25 dB increase in SNR. Doppler shifts for each transmission event were predicted from available parameters and were found to be within ±2 mHz of the measured shifts. The modal analysis employed two methods: comparing variations in signal energy with depth to the vertical extent of the modes, and fitting the data using a least squares modal decomposition. The least squares performance given a subsampled basis set of modes was studied and improved upon through the use of diagonal loading. Lack of array orientation data hindered the analysis, and least squares fitting was used to estimate the most likely orientation. The least squares analysis indicated the presence of modes at least up to mode 7, possibly higher. This is significant in that predictions prior to the experiment were that all but the lowest modes would be attenuated by boundary interactions along the path. Results from independent analyses of the same data also support the conclusion that the signal structure is quite complex.

Description: Funding for this work was provided by the Department of Energy Grant #DE-FG02- 91ER61100 (Vertical Arrays for the Heard Island Acoustic Feasibility Experiment on Monitoring Global Warming), and the University of California - Scripps ATOC Agreement, P0#10037359.

Description: Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution May 1996

Description: This thesis addresses the three major components of a texture classification system: texture image transform, feature extraction/selection, and classification. A unique theoretical investigation of texture analysis, drawing on an extensive survey of existing approaches, defines the interrelations among 11 types of texture analysis methods. A novel unification of the different methods defines a framework of transformation and representation in which three major classes of transform matrices capture texture information of increasing coherence length or correlation distance: the spatial domain method (co-occurrence method), the micro-structural method (run-length method), and the frequency multi-channel method (Fourier spectrum method). A more concise vector representation of a selected transform matrix is then needed for input to a classifier. Unlike traditional methods, which use various special functions to describe the properties of each transform matrix, a new approach directly applies a principle component analysis technique to the transform matrix. The Karhunen-Loeve Transfonn (KLT) extracts a vector of dominant features, optimally preserving texture information in the matrix. This approach is made possible by the introduction of a novel Multi-level Dominant Eigenvector Estimation (MDEE) algorithm, which reduces the computational complexity of the standard KLT by several orders of magnitude. The statistical Bhattacharyya distance measure is then used to rank dominant features according to their discrimination power. Experimental results of applying the new algorithm to the three transform matrix classes show a strong increase in performance by texture analysis methods traditionally considered to be least efficient For example, the power spectrum and run-length methods now rank among the best. Using the same MDEE algorithm, the three extracted feature vectors are then combined into a more complete description of texture images. The same approach is also used for a study of object recognition, where the combined vector also include granulometric; object-boundary, and moment-invariant features. In most classification experiments, a simple statistical Gaussian classifier is used. The plankton object recognition experiments use a Learning Vector Quantization (LVQ) neural-net classifier to achieve superior performance on the highly non-uniform plankton database. By introducing a new parallel LVQ learning scheme, the speed of network training is dramatically increased. Tests show a 95% classification accuracy on six plankton taxa taken from nearly 2,000 images. This result is comparable with what a trained biologist can accomplish by traditional manual techniques, making possible for the first time a fully automated, at-sea approach to real-time mapping of plankton populations.

Description: ONR and the WHOI joint program education office

Description: Submitted in partial fulfillment of the requirements for the degree of Master of Science at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution August 1991

Description: This thesis describes the implementation and evaluation of an adaptive time-domain algorithm for signal enhancement from multiple-sensor observations. The algorithm is first derived as a noncausal time-domain algorithm, then converted into a causal, recursive form. A more computationally efficient gradient-based parameter estimation step is also presented. The results of several experiments using synthetic data are shown. These experiments first illustrate that the algorithm works on data meeting all the assumptions made by the algorithm, then provide a basis for comparing the performance of the algorithm against the performance of a noncausal frequency-domain algorithm solving the same problem. Finally, an evaluation is made of the performance of the simpler gradient-based parameter estimation step.