DOA Estimation Algorithm Of Coherent Sources Based On Single Snapshot

Direction of Arrival estimation which has been widely applied for radar,mobile communication,sonar,and physical prospecting is a hot topic in array signal processing.The research on the DOA estimation algorithm of coherent sources based on single-snapshot is of great importance in practical application.Because the target-originated signal cannot be effectively estimated with the subspace-based high-resolution algorithms for angle when there are coherent sources.And the array receiving data is limited in length for which only the data of limited times or single snapshot can be used for emergencies of short time,fast-moving target,or poor receiving data.Although deep learning technique has been widely applied in signal processing fields based on its powerful self-adaptation and its solution for nonlinear problems,there are very few researches on DOA estimation.Therefore,this article conducts a research on the DOA estimation algorithm of coherent sources based on single snapshot from the two levels of traditional methods and deep learning method.The specific work can be summarized as follows: This article has done the following work on the basis of the original single snapshot algorithm:(1)The DOA estimation algorithm of single snapshot based on weighted spatial smoothing is suggested to sufficiently utilize the array receiving data of single snapshot.The core concept of this algorithm is to acquire the weighting matrix by the pretreatment for the pseudocovariance constructed by the data of single snapshot with the application of the embedded space smooth algorithm.We can obtain both the coherence and DOA estimation between sources through the weight sum with higher performance of sense finding compared with the DOA estimation algorithm of single snapshot available.(2)The ESPRIT algorithm of coherent signal based on single snapshot is suggested for the low accuracy of angle measurement by the traditional DOA estimation algorithm of single snapshot under the condition of low signal-noise ratio.The core concept of this algorithm is to achieve the angle estimation of incident signals by the direct utilization of snapshot data in one time for the construction of two Toeplitz matrix that meets the rotation invariance and the eigenvalue decomposition of one time based on the principle of matrix-penciled ESPRIT algorithm.There is no need of calculation for covariance matrix with this algorithm and it improves the accuracy of angle measurement compared with the ESPRIT algorithm of single snapshot available.(3)The DOA estimation algorithm based on deep learning is data-driven.It transforms the DOA estimation as pattern classification to establish the Deep Neural Networks model by the cascade connection of auto-coder and parallel classifier.It extracts the real part and imaginary part of upper triangular in covariance matrix of array as the input of feature vector for DNN network by the quantization of 1 bit for the array receiving single snapshot data based on the concept of 1 bit quantization considering the influence of high cost,high power consumption and high complexity of DNN model in the array receiving signal model.It trains the DNN model by the training set with DOA label which are constructed in advance.The auto-encoder in this DNN model may conduct the noise reduction processing for the input feature vector and output the final angle estimation result through the multilayer classifier.The simulation experiment indicates that the algorithm provides higher accuracy for angle measurement and stronger generalization capability.