Research Of Direction Finding Algorithms In Broadband Interferometer System And Its GPU Implementation

Recent years, wideband signals have been increasingly applied to the fields ofcommunication, radar and direction finding control. Therefore, how to accuratelymeasure and extract the direction of incoming waves becomes particularly important inthe wideband environment. Direction finding algorithms in the wideband directionfinding system are studied, where the traditional space angle algorithm is improved andimplemented on the GPU platform.In this thesis, three kinds of interferometer algorithms are studied. The first one isthe standard correlation interferometer algorithm which can eliminate the effects ofnon-ideal factors by comparing the measured phase differences and sample phasedifferences. In wideband environment, higher computation ability of the system isrequired to solve the concurrent data from multiple channels. when the number of thechannels increases large enough, it is hard for the standard correlation interferometer toensure the real-time performance. To solve this problem, another two algorithms basedon auxiliary baseline, the space angle correlation interferometer algorithm and theextend-baseline algorithm, both of which have relative smaller computation comparingto the standard correlation interferometer algorithm, are studied. The space anglealgorithm introduces the space angle with parallel auxiliary baselines, usingone-dimensional search of space angle to replace the two-dimensional search of azimuthand pitching angle. Therefore, the sample library size as well as the operation time arereduced. The extend-baseline algorithm uses the auxiliary baseline to solve the phaseambiguity problem. At last, above algorithms are compared in the same simulationenvironment to provide a reference for practical application.In actual direction finding system, it is found that the ideal algorithm of spaceangle is not suitable for the actual direction finding system. In order to solve theproblem, conventional algorithm is modified where it contains two parts, including therough-measurement part and the accurate-measurement part. In rough-measurement part,three kinds of improved methods are used to improve the anti-system ability, including increasing the baseline group, baseline combination and the result extension. Theycompute all possible angles by using the ideal sample library of space angle algorithmto reduce related computation times. The accurate-measurement part uses thepre-sampled library to avoid the effects of imperfect factors. Finally the simulationresults demonstrate that the direction finding precision can be improved with theimproved algorithm.In order to further improve the real-time performance, this thesis utilizes theNIVIDA’s CUDA programming platform to implement the GPU parallel acceleration ofimproved algorithm. By reconstructing the related processing function and optimizingthe parallelism of space angle, the best performance is achieved. At end of this thesis,the execution time of improved algorithms is tested to verify the higher efficiency of theGPU implementation.