A neural network approach to DOA estimation with array errors

Abstract

A linear array composed of N array elements with element position error is used as a receiving array to receive the original training data generated by the signal sources within the range of M groupsof training intervals. The training data of the input layer of the neural network are obtained by matrix operation and normalization from the original training data set. According to the dimension andprecision requirement of training data Rm and signal source arrival angle theta m, the number of hidden layers and the number of neurons in each hidden layer are set, and the dimension of weight matrix and threshold matrix of each layer are determined according to the number of neurons in input layer, hidden layer and output layer, so as to construct a neural network; M groups of training data are input to the input layer of the neural network, and the optimal weight matrix and the optimal threshold matrix reflecting the mapping relationship between the training data R and the angle of arrival of the signal source are obtained through the neural network training. The final output value of the neural network is the estimated value of the angle of arrival of the test signal by using the modified optimal weight matrix and the optimal threshold matrix to carry on the forward operation to the test data RC.

Description

technical field [0001] The invention belongs to the technical field of signal processing, and in particular relates to a neural network method for estimating the arrival angle of a signal source under the condition of an array error. Background technique [0002] The high-resolution algorithms for array signal angle of arrival estimation require the array to be in an ideal state, that is, there is no error in the array. The performance of subspace-like high-resolution algorithms is improved. Array error has become a bottleneck problem that restricts the performance of signal parameter estimation. For this reason, scholars have proposed a large number of error correction methods, including active error correction and passive error correction. These error correction methods require a known error model, error The modeling of the array is a very complicated process, especially the modeling of the coupling error and the comprehensive error under the existence of multiple errors ...

AI Technical Summary

Problems solved by technology

[0002] The high-resolution algorithms for array signal angle of arrival estimation require the array to be in an ideal state, that is, there is no error in the array. performance of subspace-like high-resolution algorithms

Array error has become a bottleneck problem that restricts the performance of signal parameter estimation. For this reason, scholars have proposed a large number of error correction methods, including active error correction and passive error correction. These error correction methods require a known error model, error The modeling of the array is a very complicated process, especially the modeling of the coupling error and the comprehensive error under the existence of multiple errors will be more difficult. When the error model does not match the real error of the array, it will inevitably lead to subsequent Performance degradation of error correction algorithms

The performance of the active error correction method is not only affected by the error model and the accuracy of the signal source parameters. When the signal source parameters are inaccurate, the error estimation performance will also deteriorate. Although the passive correction method does not require known signal source but the complexity of the algorithm is often high

Images