Research On Broadband Constant Beamwidth Beamforming Method For Small-scale Vector Array Based On GPU

Small underwater platforms such as unmanned underwater vehicle play a significant role in both military and civilian fields with a broad future development prospect.Due to the limitations of equipment cubage,the scale of sonar array carried by is also restricted.New problems such as the widening of beam main lobe width and the decreasing of target resolution are caused by the reduction of array aperture.Therefore,studying underwater acoustic signal processing methods applied to small-scale array comes to a vital research direction and focus.The vector hydrophone can make full use of sound pressure and vibration velocity information of sound field.Vector array signal processing has higher gain and better directivity because of the sound intensity obtained by joint processing and the directivity shaped like “8” carried by every vector hydrophone.In this paper,vector hydrophone is applied to small-scale array,and constant beamwidth beamforming methods of small-scale vector array are studied.Combined with GPU high-performance computing technology,a parallel acceleration algorithm is designed and implemented on hardware platform Jetson processor.Firstly,this paper presents basic principles of sonar array signal processing and vector array beamforming.Through analyzing the mathematical model of signal received by sound pressure array,the common narrowband beamforming methods of sound pressure array are reviewed and simulated.Based on above research,the mathematical models of signal received by single vector hydrophone and vector array are discussed.Narrowband beamforming methods for vector array are also simulated.By means of comparing the beamforming methods’ effect between sound pressure array and vector array,we can come to a conclusion that vector array is superior to sound pressure array.Secondly,the principles of three traditional constant beamwidth beamforming methods are studied and simulated,including linear combination of array method,Chebyshev weighted iteration method and spatial resampling method.Weighing up the processing effects of three methods above on small-scale arrays.In response to the disadvantage that the main lobe width of beam pattern is too broad,which is caused by using traditional constant beamwidth beamforming methods on small-scale array,a biomimetic auditory coupling algorithm that can effectively increase the effective aperture of array and a second-order cone programming constant beamwidth beamforming method that can control the main lobe width and side lobe level at the same time are studied.By combining two algorithms,a new method of constant beamwidth beamforming which is suitable for small-scale array is designed.The effectiveness of new method is verified through simulation experiments.Thirdly,the principles of GPU parallel acceleration algorithm and CUDA programming model are studied.Based on the characteristics of constant beamwidth beamforming algorithm,including of large data scale,complex and intensive computation,and long computation time,the serial algorithm is redesigned for heterogeneous parallelism by calling GPU multithreaded computation.A network transmission module has been designed to transmit data between PC and GPU platform Jetson processor.Finally,a lake trial is conducted.Using biomimetic auditory coupling and second-order cone programming constant beamwidth beamforming method on small-scale vector array to process the sea trial data and verifing the performance of algorithm.Using Jestson processor to test and validate the parallel constant beamwidth acceleration program.Experimental result shows that the constant beamwidth beamforming algorithm for small-scale vector array can achieve a narrow main lobe width,and the parallel algorithm has good acceleration effect,effectively reducing the computation time and achieving a real-time processing.