Research On Some Key Issues Of Super-resolution Direction Finding Based On Supervised Learning

The sensor array received the data which contained the direction of arrival(DOA)of the spatial signals.So we could extract the anglular parameter of the sources to determine the position of the signals.After years of research and development,so many super-resolution DOA algorithms had emerged.These methods had good performance to estimate DOA.The thesis mainly focused on the key issues of super-resolution DOA technology,including the optimization and improvement of estimating DOA for narrowband independent sources,the realization of coherent signals direction finding in colored noise environment,and wideband hybrid sources DOA estimation.The paper first studied the narrowband,coherent and wideband signal models,and then procided theoretical analysis of several classical DOA estimation algorithms.Then the narrowband DOA estimation method was on the basis of support vector regression(SVR).Extract the signal subspace as feature vector.Employ genetic method to optimize the parameters of the SVR model,the DOA estimation model was constructed by training samples,and finally estimate the DOA through the trained model.Compared other methods,the effectiveness of this algorithm could be certificated.In colored noise environment,the coherent DOA estimation algorithm put forward on the basis of back propagation neural network(BPNN).At the same time,particle swarm optimization(PSO)was able to improve the stability and generalization performance of BPNN.The parameter features which only contained message of coherent sources were extracted as the input of BPNN,and the angles of the signals were used as the output,thereby conforming a sample to train and test the BPNN.The simulation outcomes certified the performance of the method had been effectively improved.The DOA estimation algorithm for wideband mixed signals(independent signals and coherent sources coexist)based on random forest regression(RFR)was proposed.The input data of RFR was the triangular elements on the covariance matrix at different frequency points,and the output was the signal angle parameters.Theoretical analysis and simulations demonstrated that the proposed method had high directional accuracy and better directional performance.