submenu »

project home | overview | downloads | publications | related links

Parallel and Distributed Computing Architectures and Algorithms for Fault-Tolerant Sonar Arrays

Sponsor: Office of Naval Research -- US321/SS321

Quiet submarine threats and high clutter in the littoral undersea environment demand higher-gain acoustic sensors to be deployed for undersea surveillance.  This trend resulted in the use of high-element-count sonar arrays with increasing data rates and associated signal processing. These autonomous passive sonar array technologies are limited by poor fault-tolerance due to single points of failure and computational complexity that cannot be supported in real-time by conventional means. The limitations are especially evident for a large number of receiving nodes and with the continuing development of higher-fidelity algorithms such as adaptive and matched-field processing. The objectives of this research are to overcome such limitations with parallel and distributed computing (PDC) technology in the form of autonomous sonar arrays using in-array processing. This research leverages fault-tolerant distributed and parallel processing techniques to decrease cost and improve performance and reliability of large, autonomous, battery-powered, disposable sonar arrays.

Beamforming is a class of array processing that optimizes an array gain in a direction of interest. The determination of the direction of arrival relies on the detection of the time delay of the signal between sensors. Incoming signals are steered by complex-number vectors. If the beamformer is properly steered to an incoming signal, the multi-channel-input signals will be amplified coherently, maximizing power in the beamformed output; otherwise, the output of the beamformer is attenuated to some degree. Thus, peak points in the beamforming output indicate directions of arrival for sources. Based upon the processing approach, the beamforming algorithms are classified into several categories such as Conventional Beamforming (CBF), Adaptive Beamforming (ABF), and Matched Field Processing (MFP). In particular, the matched-field processing (MFP) focuses on acoustic propagation modeling of the ocean waveguide with signal processing algorithms. The MFP algorithm localizes acoustic sources in range and depth more precisely than plane-wave beamforming methods by using a full-wave acoustic propagation model instead of a simple plain-wave acoustic propagation model for the ocean.

With continuing development of higher-fidelity algorithms, conventional sonar array systems have the problems such as significant computational complexity not readily implemented in real-time, increasing memory capacity needs, and low fault tolerance capability. The use of distributed and parallel computing to perform computationally intensive beamforming algorithms provides an alternative with an increase in processor speed. Parallel processing algorithms coupled with advanced networking and distributed computing architectures can be used to turn telemetry nodes of arrays into processing nodes and thereby function as a distributed processing system for autonomous, in-situ beamforming. A parallel and distributed processing approach has the potential to eliminate the need for a centralized data collector and processor, and improve overall computational performance, dependability, and versatility. This architecture is composed of intelligent nodes connected by a network.  Each of the smart nodes, comprised of a hydrophone and a microprocessor, has its own processing power as well as requisite data collection and communication capability.  By using such distributed array architecture, the algorithmic workload is distributed and cost is reduced

In battery-powered and disposable sonar arrays system, faults are inevitable due to the harsh underwater conditions and limited power. Fault detection and self-healing algorithms are required to improve the reliability and increasing mission time of a sonar system. A key element of fault tolerance design studies is how best to improve fault coverage, reliability, and mission time while keeping low the power, weight, and cost factors. Success is measured in terms of system reliability, mission time, fault recovery, and the price paid for these in terms of power, weight, size, and cost as compared to the baseline sequential sonar array. 

Since 1996, this project has been targeted toward the development of new algorithms and supportive architectures for parallel in-situ processing of conventional beamforming, adaptive beamforming (ABF), matched-field tracking (MFT), and matched-field processing (MFP) applications on low-power, distributed, autonomous sonar arrays. Decompositions in various domains were developed to provide a basis for designing parallel beamforming algorithms, and experiments were conducted and results gathered and analyzed on a diverse set of testbed platforms. Building on our previous success with the parallelization of the split-aperture conventional beamforming (SA-CBF), narrowband and broadband processing for minimum variance distortionless response (MVDR), and subspace projection beamforming (SPB), a variety of decomposition, partitioning, and mapping strategies for baseline ABF, MFT, and MFP algorithms have been investigated, with network services and distributed system testbeds developed to support these investigations.

Recently, this research has been focused on satisfying ever-increasing computational trends and high-dependability requirement for adaptive MFP algorithms. One approach is to develop parallel algorithms for adaptive MFP algorithms to overcome intensive computational and memory demands in distributed array systems.  The other approaches are to present robust adaptive algorithms and fault-tolerant mechanisms to meet the highly reliable operational requirements for real-time sonar systems in harsh underwater environments.

OTHER HPC Group PROJECTS

GEMS Project

Reconfigurable Computing (RC) Empowered Grid Computing

Computational Framework for Simulating Joint Mechanics