Theory and Implementation of Advanced Signal Processing for Active and Passive Sonar Systems

This is a collection of articles written by members of the Underwater Acoustic Signal Processing (UASP) Technical Committee. The first article, by D. W. Tufts, deals with the history of UASP prior to 1980. In this period, initial mathematical models were developed and the first experimental investigations of underwater acoustic propagation were performed. It was also recognized during this time that there are many similarities between radar and sonar signal processing. The article by J.P. Ianniello deals with research in passive and active sonar from 1980 to the present. Work in this period included experimental verification of algorithms that had been developed in the 1960s and 1970s (e.g. for adaptive beamforming), as well as the development of new approaches, which include acoustic propagation modeling in the design of signal processing algorithms. Such processing is referred to as matched field processing. A common task in passive sonar systems is to estimate the difference in times at which different sensors receive the same signal. Time-delay estimation is a first stage that feeds into subsequent processing blocks. I. Lourtie provides a concise review of work in this field. The article by J.C. Preisig deals with underwater acoustic communications. The underwater channel has several features that make reliable communication a challenging problem. Nevertheless, progress is being made by combining results from ocean acoustic modeling, communication theory, and signal processing. The final article, by J.M.F. Moura, deals with the future of signal processing in the ocean. In addition to considering advances in detection and localization, he deals with new applications such as acoustic tomography, physical oceanography, and synthetic aperture sonar.

It is well known that an optimum operating frequency exists for a particular sonar system with a specified range capability. In this paper, the extreme importance of the frequency curvature of the acoustic absorption-loss coefficient used in determining the optimum frequency is illustrated. A lower bound on optimum operating frequency, independent of the expression used for absorption loss, is derived. Results are presented for both active and passive sonar systems, and for four different, generally-accepted absorption-loss coefficient expressions.

Underwater multi-targets tracking has always been a difficult problem in active sonar tracking systems. In order to estimate the parameters of time-varying multi-targets moving in underwater environments, based on the Bayesian filtering framework, the Random Finite Set (RFS) is introduced to multi-targets tracking, which not only avoids the problem of data association in multi-targets tracking, but also realizes the estimation of the target number and their states simultaneously. Usually, the conventional Probability Hypothesis Density (PHD) and Cardinalized Probability Hypothesis Density (CPHD) algorithms assume that the detection probability is known as a priori, which is not suitable in many applications. Some methods have been proposed to estimate the detection probability, but most assume that it is constant both in time and surveillance region. In this paper, we model the detection probability through the active sonar equation. When fixed the false detection probability, we can get the analytic expression for the detection probability, which is related to target position. In addition, this novel detection probability is used in PHD and CPHD algorithms and applied to underwater active sonar tracking systems. Also, we use the adaptive ellipse gate strategy to reduce the computational load in PHD and CPHD algorithms. Under the linear Gaussian assumption, the proposed detection probability is illustrated in both Gaussian Mixture Probability Hypothesis Density (GM-PHD) and Gaussian Mixture Cardinalized Probability Hypothesis Density (GM-CPHD), respectively. Simulation results show that the proposed Pd-GM-PHD and Pd-GM-CPHD algorithms are more realistic and accuratein underwater active sonar tracking systems.

Active sonar systems are used in the detection and localization of underwater targets. While traditional approaches use pulsed active sonar (PAS) to transmit short bursts, high duty cycle (HDC) sonar systems have been the focus of recent research and can overcome the shortcomings of PAS. Since HDC sonar systems transmit a long pulse train waveform, we must address sub-pulse interference issues, which requires multiple orthogonal sub-pulses and generalized sinusoidal frequency modulated (GSFM) pulses are suitable for this purpose. Unfortunately, conventional GSFM pulse train design methods do not generate an adequate number of orthogonal sub-pulses. Therefore, we propose an improved GSFM pulse train waveform design approach for overcoming the limitations of these conventional methods. The proposed method is organized into two parts, the first focused on the assessment of the auto-correlation, and the second focused on the optimizing the orthogonality between sub-pulses. Cost function is also carefully designed for this purpose. From simulation experiments, we found that the proposed method was able to produce 30 orthogonal sub-pulses and an optimized waveform that demonstrated better detection performance than conventional waveforms under most conditions. We also verified the performance of the proposed method via actual sea experiments, with analysis showing superior performance matching that of simulations. An analysis on the bandwidth parameter K is also conducted and it is found K = 1.25 to be the best option.

Active sonar systems can provide good target detection potential but are limited in shallow water environments by the high level of reverberation produced by the interaction between the acoustic signal and the ocean bottom. The nature of the reverberation is highly variable and depends critically on the ocean and seabed properties, which are typically poorly known. This has motivated interest in techniques that are invariant to the environment. In passive sonar, a scalar parameter termed the waveguide invariant, has been introduced to describe the slope of striations observed in lofargrams. In this work, an invariant for active sonar is introduced. This active invariant is shown to be present in the time-frequency structure observed in sonar data from the Malta Plateau, and the structure agrees with results produced from normal mode simulations. The application of this feature in active tracking algorithms is discussed.

Active Doppler sonar systems are gaining popularity as a method to profile remotely the oceanic velocity field. Estimation of the variance with which an ideal single.beam sonar system measures relative Doppler motion is the subject of this paper. Theoretical formulations, simulated reverberation data, and sea data are considered to determine this variance. The sea data considered in this paper were selected carefully to avoid performance degrading physical/instrumental phenomena and bias corrected to remove further instrumental effects. Asymptotic performance, determined in this research, is realized when instrumental and physical phenomena do not measurably affect experimental performance. Oceanic and simulated data sets are processed with several velocity-estimation algorithms that are implemented in existing sonar systems. These algorithms are compared against each other and the velocity-estimation standard-deviation results tabulated. The algorithms are shown to possess similar performance characteristics when simulated, and sea data are processed in spite of the fact that some of these algorithms are suboptimal from a theoretical viewpoint. The results of this study show that the asymptotic performance of the Doppler system examined is approximately 25 percent above a theoretical lower bound.

This paper presents a scalable hardware and software architecture for a Passive Surveillance Sonar (PSS) system, enabling real-time processing of high-bandwidth data using multicore processors and high-speed telemetry interfaces. Larger arrays, required to detect quieter submarines in noisy environments, increase data loads; a novel scheme is proposed for efficient transfer between the digitizer and processor. The beamformer, the most computationally demanding component, is analyzed in its time-domain delay-and-sum, frequency-domain, and minimum variance distortionless response (MVDR) forms, including variants for different array structures. The architecture is evaluated with cylindrical arrays of up to 7,600 sensors generating 4 Gbps of data. The parallel and pipelined design achieves real-time performance and scalability. Experiments demonstrate the formation of 450 panoramic beams, detecting and tracking multiple targets with $$3^\circ$$ bearing accuracy.

Passive sonar systems are often used to detect marine mammal vocalizations in order to localize and track them. However, classifying the genus or species is often required and this typically requires expertise not readily available on the vessel. Perceptual signal features, similar to those employed in the human auditory system, have been used to reduce false alarms in active sonar by automatically discriminating between target and clutter echoes. Many of the features were inspired by research directed at discriminating the timbre of different musical instruments — a passive classification problem — which suggests it might be applied to classify marine mammal vocalizations. To test this hypothesis, the automatic aural classifier was trained and tested on a set of marine mammal vocalizations from four species: the northern right whale, sperm whale, humpback whale, and bowhead whale. This paper will provide an overview of the aural classifier’s architecture, describe the data set, and present some preliminary classification results.

The frequency domain beam forming techniques specially using 2D-FFT process are used mostly in passive sonar. It provides good results mostly in applications like towed arrays that are towed behind submarines and ships. The linear hydrophone array is sampled in time .The data is transformed from time domain into frequency domain by FFT process. Then another FFT is applied along the nth frequency bin in space. This provides good frequency contents of the towed arrays. The plot obtained from this 2D-FFT process is called k-omega plot. In which frequency is on x axis and wave number k on y axis. K-omega beam forming is widely used in passive sonar for detection, classification and localization purposes. The technique requires uniform linear array for beam forming. The linear hydrophone arrays are split into further sub arrays. These sub arrays are spaced spatially according to different cut frequencies. With the introduction of long passive arrays for naval and oil exploration purposes, the parallel K-omega processing problems becomes complex. Traditionally the implementation is done on cluster of IBM PowerPCs or more recently Intel Xeon clusters. The heavy computation load is often compromised with more optimized algorithms fit to specific architecture. This in practice requires major overhaul in the system if new feature is to be incorporated in the passive towed array sonar system. In this research paper the K-omega beam forming is implemented on IBM cell processor. The cell processor with its inherent parallelism in architecture provides an opportunity to execute processing of sub arrays concurrently. The K-omega beam forming is implemented on simulated 5 to 7 nested sub arrays and results are computed in simulator and tested against the Matlab outputs. The k omega beam forming used in this implementation depends on FFTW library along with IBM math, vector and matrix libraries. The implementation is benchmarked using IBM cell processor simulator. Another aspect of cell processor found in this research paper is its very fine memory handling capability. This is vital for Matrix calculations and iterative tasks. From the implementation it is evident that this can be easily ported to commercially available system i.e. Sony playstation 3. The Sony playstation 3 with Linux fedora core 6 is ideal for such an application and it can be used for training of passive sonar operator.

Passive sonar has been widely researched and used in underwater target tracking for its high covertness. In this paper, the detection and tracking of an unknown and time-varying number of targets for a towed sonar system is considered. To enhance the detection and tracking performance, a track-before-detect (TBD) based signal processing method is devised. Specifically, the raw bearing measurements generated after acoustic signal beamforming are directly used as the inputs of the proposed TBD processor. To handle the nonlinearity between target states and the raw bearing measurements, particle filter is employed to compute the joint multitarget probability density (JMPD) recursively through a Bayesian framework. Besides, the target initiation and termination steps are integrated into the particle filtering process as well to deal with the random births and deaths of multiple targets. The effectiveness of the proposed method are demonstrated by numerical experiment.

Marine mammals use sound for critical life functions through passive listening and, in some species, active sonar. Major recent advances have been made in understanding what sounds they make, how they hear, and how noise affects their hearing, behavior, and physiology. While data are lacking for whole groups, notably baleen whales, direct hearing measurements are available for most taxa. Some species, such as odontocete cetaceans, hear extremely broad frequency ranges, well into ultrasonic biosonar frequencies. Pinnipeds have more limited frequency ranges but are adapted for aerial and underwater hearing. Indirect data suggest whales use and are more sensitive to low frequencies. For most species, quite high received sound levels are required to temporarily or permanently affect hearing. Direct experimental measurements of behavioral responses have revealed strong species and context dependencies, with factors such as proximity, novelty, and spatial relationships influencing response probability as well as relative loudness. For active surveillance sonar systems operating in the 30-200 kHz range, whales and otariid pinnipeds are of much less concern in terms of audibility and potential behavioral disturbance than phocid pinniped and odontocete cetaceans, notably porpoises. Hearing loss or other injuries from such systems are quite unlikely in almost all practical scenarios.

A model is proposed for simulating the group dynamics and acoustic echoes of aggregate, discrete biologics in midfrequency (2–5 kHz) active undersea sonar. Using a Lagrangian description, group behavior is modeled by a set of Newtonian forces and stochastic perturbations on a collection of point masses. Interactions comprise a grouping force, which gives a characteristic separation distance, and an arrayal force, which tends to align the velocities of the group’s constituents. The group dynamics is characterized in terms of their statistical and asymptotic properties and studied as a function of the model parameters. Several models for the acoustic response are considered and studied for both individual and aggregate response. Finally, the resulting synthetic echoes are compared against those of actual biologics measured on a midfrequency active sonar system. The synthetic echoes are found to reproduce well both the qualitative characteristics of measured biologics and the resulting tracking and classification features found in a typical active sonar signal and information processing chain.

A modified adaptive noise cancelling (ANC) hybrid algorithm incorporating techniques of optimal wideband FIR filtering operation and the trimmed mean (TM)-levelization, a dynamic level-based thresholding process, is described in this paper for wideband active sonar echolocation system. This novel technique performs effective suppression of unwanted part of the training data thus localizing potentials of target returns. Furthermore, due to a massive parallelism in the system although the current electronic processing units such as GPU (graphic processing unit) and GPGPU (general-purpose GPU) work well, we propose the concept of smart-pixel (SP) as a possible solution towards hardware implementation of the above algorithm. In this architecture of parallelism, an array of optical devices is integrated onto one chip and then utilized as a computational cell to provide an alternative solution towards parallel implementation of the ANC hybrid algorithm.

Recent developments in side‐scan sonar technology have increased the potential for fundamental changes in our understanding of ocean basins. Developed in the late 1960s, “side looking” sonars have been widely used for the last two decades to obtain qualitative estimates of the acoustic properties of the materials of the seafloor. Modern developments in the ability to obtain spatially correct digital data from side‐scan sonar systems have resulted in images that can be subsequently processed, enhanced, and quantified. With appropriate processing, these acoustic images can be made to resemble easily recognizable optical photographs. Any geological interpretation of these images requires an understanding of the inherent limitations of the data acquisition system. When imagery is collected, these limitations are largely centered on the concept of resolution. In side‐scan sonar images, there are several different types of resolution, including along‐ and across‐track resolution, display resolution, and absolute instrumental resolution. All of these parameters play a critical role in our ability to calibrate and ultimately to interpret the new pictures of the ocean floor. Acoustic image processing is a new application of an old and well‐established technique. Digital optical images have benefited from several decades of development in processing techniques. These relatively sophisticated techniques have been applied to photographic images from satellites and spacecraft, images which are “noisy” and difficult to obtain but extremely valuable. Side‐scan sonar systems, on the other hand, have only recently been able to produce spatially correct, digital images of the seafloor. The application of digital signal‐processing techniques to side‐scan sonar data will now allow us to quantify what had been previously very subjective and qualitative interpretations of images of the seafloor. The goal of all this processing of acoustic images remains clear: the development of an interpretable map of the geology of the seafloor.

The passive and active sonar equations are reviewed with regard to the definitions of individual terms and their units. It is shown on the one hand that ambiguities exist in conventional definitions that can lead to misinterpretation, and on the other that under special conditions correction terms are needed that are not included in the standard equations. Revised definitions are proposed that remove the need for some of the correction terms and reduce the likelihood of misunderstanding. Particular emphasis is placed on the terms 'propagation loss' and 'source level' for both active and passive sonar. For example, the effect on changes in impedance between the source and receiver of the sound is considered in the case of passive sonar, and between the source, receiver and scatterer of sound in the case of active sonar.