Correlation Interferometer On Circular Array And GPU Realiazation

As the signal environment of the electronic warfare becomes more and morecomplicated, the frequency band of the signal is also wider than before gradually. In thiscase, it becomes much more important to find the direction of the arrival signal.Interferometer direction-finding system, with high direction finding precision, simplealgorithm principle, high real-time performance and other advantages, has become thekey direction finding technology in broadband direction-finding systems which arewidely applied nowadays.The key algorithm of broadband direction-finding system is studied in detail in thisdissertation, and the main contents of this paper are listed as follows:Firstly, the errors of direction-finding system are analyzed and then severalcorresponding technical indexes are given. A detailed analysis on main error sources ofinterferometer direction-finding system is made, and several methods to reduce errorsare proposed. On this basis, the definition and measurement methods of four technicalindexes are presented to provide the judging standard for the results of simulationexperiments in this paper.Secondly, a2-dimensional correlation interferometer algorithm base on dimensionsplit is proposed in this dissertation. On the basis of conventional correlationinterferometer direction finding algorithm, cosine function is chosen as the similarityfunction and the problems of phase ambiguity and boundary mutation are solvedeffectively. Meanwhile, the application of conic fitting improves the results of thealgorithm as well. According to the characteristics of correlation coefficient, animproved method based on dimension split which simplifies the original2-dimensionalangle searching by dividing it into two1-dimensional searching, and accelerates thealgorithm greatly and at the same time maintains high direction finding precision, is putforward. Then the conditions for dimension split are also discussed.Thirdly, correlation interferometer algorithm based on RBF neural network isstudied in this dissertation. Two commonly used learning algorithms are utilized fornetwork training, and their models for correlation interferometer direction finding are given separately. Once get the reasonable network parameters, RBF net which is trainedby OLS algorithm performs better than correlation interferometer algorithm.Finally, GPU platform is adopted to realize correlation interferometer algorithm.This paper utilizes the high parallelism of GPU platform to design CUDA program, andoptimizes it according to its characteristic for the sake of improving the executionefficiency if possible. The experimental results show that optimized program has abetter performance in GPU platform than CPU platform.