Application Of Parallel Computation Technology In Matched-field Source Location

Matched-field processing (MFP) can break through the distance-limited limit of thetraditional passive localization way, and accurately estimate distance and depth of remote andextremely remote acoustic Source by employing a full-wave acoustic propagation modelinstead of a simple plane-wave acoustic propagation model for the ocean. So MFP isbecoming one of the most heated topics in the field of underwater acoustics research. Butbecause of very complicated environment condition for underwater sound, Match fieldpositioning was confronted with some questions of huge computation, Long time and hugestorage occupation. In order to providing supporting data for the fast simulation environment,this paper uses Parallel processing technique into Matched-field processing sourceslocalization. The processing tasks will be assigned reasonably to multiple processors of thecomputer system, so processing tasks may be finished cooperatively in shorter time, andaccelerate the computing speed.Through studying deeply and analyzing the inherent parallelism which existed inalgorithm of the match field sound source localization, parallel computing Platform has beenconstructed in the match field localization. Parallelization method was introduced in detail,which was applied to the match field localization algorithm. Aimed at the performance ofmatch field localization, this paper put forward the two computational methods in parallel,which includes the section decomposition (SD) method and the frequency decomposition (FD)method. The former is based on data parallelism in the processing of grid points, the latterexploits data parallelism between processing tasks of different frequency bins. And parallelalgorithm has been carried on the test and the corresponding analysis in the MPI parallelenvironment, in the way of interconnecting binuclear PC machines as the parallel computinghardware platform through fast internet, choosing Windows operate system and FORTRANcompiler.To further improve the efficiency of computation, parallel genetic algorithm wasintroduced as optimization tools of matched field processing. According to the actualcharacteristics of matched field processing, the operators and evaluation criteria of geneticalgorithm were designed reasonably in order to make its rapid convergence. In the parallelgenetic algorithm, the island model was adopted to complete inter-process communication.And an algorithm communication feature was analyzed and compared to learn affection oncomputation efficiency and accuracy of solve problems. The performances of parallel algorithms are simulated and tested in detail, in terms ofexecution times, communication times, speedup rations, parallel efficiencies of the sequentialand parallel algorithms. The simulation results showed: the method of parallel geneticalgorithm had more advantages than the traditional location method to solve the problem ofmatched field localization, such as fast calculating speed, pretty good adaptability, betterrobustness, and high location accuracy. Questions of the load big computation and longtime-consuming etc. could be settled effectively to exploiting parallelization in matched fieldprocessing methods, in order to provide settle path for fast sound source localization.