Research On Radar Network Detection Power Computing Service Using GPU

Radar network detection power analysis is a widely used tool,which is often used in the military field to determine the radar cross section that the radar network can detect under the condition of no less than certain probability.It is a key step in radar military field,such as evaluation of radar network joint operation effectiveness,division of defense side responsibility area,optimization of radar network deployment location parameters,and planning of offensive penetration path.Image processor has powerful parallel floating-point computing capability.So now GPU is widely used in deep learning,radar data processing,image analysis and other fields.GPU has a large number of simple logic operations to achieve large-scale parallel computing,and radar network detection power calculation generally has the characteristics of many geographic space sampling points,large data throughput and large amount of computation,which causes the dilemma of calculation precision and calculation time.That is to say,the more sampling points,the greater the amount of calculation.However,the calculation time increases sharply while the calculation accuracy is obtained.In order to solve the dilemma between calculation speed and calculation accuracy,this paper proposes a GPU based algorithm for radar network detection power calculation and its optimization scheme.Based on the field model,the calculation process of the joint detection power is concretized into the construction of the field model and information processing.In addition,it is also a calculation method combining calculation and visualization,which directly generates the two-dimensional information of geotagged images required by terminal display.Finally,the algorithm is optimized by GPU programming optimization techniques such as compression solution space,optimized memory usage and optimized instruction usage,so as to make full use of GPU hardware.The GPU-based computing approach involves both hardware and development environment,that is,each different GPU needs to install the corresponding NVIDIA software driver.Moreover,the number of threads started by the algorithm itself in the GPU may cause errors due to different graphics cards,which leads to low portability of the algorithm.In view of the above difficulties,this paper encapsulates the GPU-based algorithm into a micro-service with ICE framework and provides an interface for calling.Finally,this paper designs three different experiments to verify the timeliness and scalability of the algorithm from different perspectives.The experimental analysis and comparison show that the new algorithm based on GPU can adapt to different grid resolution,radar detection probability and radar maximum detection height.The optimized GPU version is about 75 times more efficient than the serial version.