Fast Computing The Cross Ambiguity Function For Passive Radar Based On Black Box Optimization

Unlike active radar,which has a dedicated transmitter,passive radar utilizes illuminators of opportunity in the environment to detect targets and estimate target parameters.As it only receives electromagnetic waves,passive radar has many aspects of superiory over active radar,such as enhanced battlefield survivability and lower cost to implement.In passive radar system,the cross ambiguity function between the echo signal from targets and the reference is calculated online to detect targets and estimate targets’ distances and velocities.However,it is often the case that the energy of environmental noise is much higher than that of the echo signal.Thus,a long-time coherent integration is often required to increase the signal-to-noise ratio.Furthermore,in order to detect longdistance targets,the cross ambiguity function has to be calculated for a large time delay range.Besides these,new signals which have very high frequences,such as digital video broadcast-satellite signal,are being exploited as the illuminators.All these facts greatly increase the computational expense for the cross ambiguity function.Although many fast algorithms and hardware techniques have been introduced to accelerate the computation,computing the cross ambiguity function is still a bottleneck for real-time target detection and localization in passive radar system.This thesis tries to reduce the computational cost of cross ambiguity function utilizing a different idea.The major work consists of two parts.(1)The traditional way for radar target detection is calculating the value of cross ambiguity function at knots covering all possible time delay and Doppler frequency ranges,and then searching for the peak corresponding to the target.Obviously,this method is computationally demanding for a large target searching range.In this thesis,black box optimization is utilized to guide the calculation of cross ambiguity function,so that the calculation procedure moves in the direction of increasing the value of cross ambiguity function.In this way it avoids traversing all the time delay and Doppler frequency points.Following this idea,set membership global optimization(SMGO)algorithm is selected to accelerate the computational procedure for cross ambiguity function because it does not depend on gradients and has proved convergency property.Then,a SMGO-based fast algorithm for computing cross ambiguity function is proposed,which also integrates traditional fast algorithms like fast fourier transform(FFT)and Zoom FFT.The feasibility and efficiency of the proposed algorithm are verified through comparisons with traditional delay traversal method and method based on genetic algorithm.The numerical simulations in MATLAB demonstrate that SMGO-based algorithm has improved efficiency for searching the peak of cross ambiguity function.(2)Since hardware acceleration is widely used in practice for the computation of cross ambiguity function,a GPU-based version for the proposed algorithm is also designed to accelerate the computational procedure in further.The program is developed in C++ using the CUDA technology such that computational routines like loop cycles and FFT can be paralleled.Comparisons with the MATLAB version program and the C++ version program without utilizing GPU demonstrate that the parallel computation provided by GPU can accelerate the algorithm in further.Moreover,the accelerating capability provided by GPU is verified by numerical examples which have different delay searching ranges and different data lengths.The experimental results show that the accelerating performance is propotional to the amount of data processed,but restricted by the memory on GPU.In general,compared with traditional methods,the SMGO-based algorithm can significantly reduce the evaluation number of cross ambiguity function.Thus,it has the potential to accelerate the target localization procedure for passive radar in engineering practice.