Design Of Real-time Simulation System Of Sonar Target Strength Based On Multi-GPU

With the rapid development of sonar detection technology and underwater weapon systems,the accuracy and speed requirements for forecasting the echo characteristics of underwater complex targets are becoming higher and higher.When forecasting the echo characteristics of a complex shape target with multiple reflections,the traditional Planar Elements Method(PEM)has low calculation accuracy,and it still takes a long time to determine the occlusion of a large number of the planar elements.Based on the Shooting and Bouncing Ray(SBR)method in the field of computational electromagnetics,this paper combines geometric optics(GO)and PEM scattering field integral algorithms to study sonar echo characteristics of underwater targets.Besides,we propose a high-performance parallel acceleration scheme based on a multi-GPU platform,and a real-time simulation system of sonar Target Strength(TS)is designed and implemented.The main work and contributions of this paper are as follows:Firstly,aiming at the problem that the ray-tracing procedure takes a long time in the SBR method,this paper applies two methods to improve the performance of the SBR method.The first is to construct the stackless KD-tree of the simulation target to improve the raytracing efficiency.And the second is to subdivide acoustic ray tube through a Multiresolution Grid Algorithm,which effectively reduce the memory resource and time consumption during calculation.Secondly,based on the multiresolution SBR method,this paper proposes a GPU parallel simulation scheme for sonar TS.Due to the sparse distribution of acoustic ray tube that need to be subdivided on the virtual aperture,and in order to take advantage of the coalesced access characteristics of global memory,this paper proposes a dense storage solution for acoustic ray tubes and rays on the GPU.For the time-consuming problems of main kernel functions of the algorithm,this paper performs in-depth performance optimization of the kernel functions from the aspects of CUDA thread calculation task allocation,GPU multilevel memory utilization,and occupancy analysis.The experiments show that the multiresolution SBR method proposed in this paper can achieve a performance improvement of 23%-80% compared with traditional SBR method based on GPU.Thirdly,this paper proposes a multi-GPU collaboration strategy based on Open MP and CUDA to achieve real-time simulation of sonar TS on a single-node multi-GPU platform.At the same time,in order to improve the load imbalance on the GPU devices,this paper proposes an L-shaped dynamic partitioning method of virtual aperture surface.Compared with one-dimensional uniform partitioning,the multiresolution SBR method's parallel performance improved by more than 20%.At this time,the calculation time of sonar TS based on multi-GPU is within 200 ms,which can meet the requirements of real-time simulation.Finally,in order to improve the integrity of the system simulation,a simulation demonstration system is designed and implemented.The system can use the graphical interface to complete the human-computer interaction functions such as simulation model selection,parameter configuration,model visualization,pseudo-color image and simulation curve drawing,which improves the practicability of this study.