Research On The Application Of ANM Theory In DOA Estimation

With the increasingly complex electromagnetic environment,the performance requirements for wave direction estimation（Direction of Arrival,DOA）high resolution and high accuracy have gradually increased,so that the methods of DOA estimation have been continuously developed.The classical subspace algorithm can achieve super resolution of the signal.However,they do not directly deal with coherent sources,and the angle estimation accuracy is not excellent in low SNR and few snapshots.The compressed sensing estimation algorithms are more robust regarding the snapshots and SNR.However,the estimation performance of compressed sensing algorithms can be affected by the grid mismatch.The development of atomic norm minimization theory solves the grid mismatch problem in compressed sensing algorithms.So this paper mainly applies it to array amplitude phase error self correction,channel compression and deep learning based on ANM theory.The results achieved are as follows:Firstly,aiming at the gain phase error problem in the actual system,a gain phase error self correcting DOA estimation algorithm based on ANM is proposed.By analyzing the channel gain phase error and combining with the ANM model,a self correcting model conforming to the gain phase error is constructed,and the equivalent semi positive definite programming process of the model is given.Then,the Convex Optimization toolbox is used to obtain the optimal solution of the optimized variable.Finally,Vandermonde decomposition is used to obtain the estimated incident signal information,and the estimated angle is used to solve the error matrix.Simulation results show that compared with the classical MUSIC based iterative self correcting DOA estimation algorithm,the proposed algorithm can further improve the DOA accuracy under the same conditions.Secondly,for channel compression structure,a channel compression DOA estimation algorithm based on ANM was proposed.On the one hand,the use of channel compression structure can reduce the number of channels of the RF front end,thereby reducing the design complexity of the RF front end;On the other hand,the resolution of ANM algorithm can be improved by using this channel compression structure.The essence of this algorithm is to combine the compression matrix in the channel compression structure and the ANM model to construct a DOA estimation model suitable for channel compression.And then establish the Toeplitz matrix according to the optimal solution of the semi positive definite programming problem,and obtain the estimation results of the signal DOA parameters through its Vandermonde decomposition.Simulation experiments show that compared with the MUSIC and ₁ svd algorithm,the proposed algorithm improves the resolution and DOA estimation performance under the premise of the same number of channels.Finally,the ANM model was applied to deep learning and an unsupervised grid less DOA estimation algorithm was proposed.The use of deep learning has changed DOA estimation from model driven to data driven,reducing the possible model assumption error problems in model driven types.The essence of the implementation of the deep learning based DOA estimation algorithm is to equivalent the semi positive definite programming process equivalent to the loss function of the unmeshed unsupervised network,and get rid of the influence of grid mismatch and label collection through the custom loss function.Then,combined with the appropriate network model training,the optimal model output solution is obtained.Finally,the Vandermonde decomposition method of the Toeplitz matrix of the optimal solution is also used to obtain the DOA parameter information to be estimated.Simulation experiments show that compared with the unsupervised deep learning method and the DCN based deep learning DOA estimation algorithm,the proposed unsupervised grid less algorithm has certain advantages in improving the performance of DOA estimation.