Research On Deep Learning Delay Angle Estimation Method For Wireless Positionin

Wireless location is an important research content of the new generation communication technology.As more and more devices are connected to the communication system,the positioning scenarios become diversified and complex,which brings greater challenges to the positioning methods using delay and arrival angle estimation.However,the traditional time delay and arrival angle methods cannot meet the high precision positioning requirements of different complex scenes.Considering the powerful learning ability of deep learning,deep learning technology is being tried to be used in the localization of signal delay and arrival angle estimation of new methods.However,the deep learning method for signal estimation still needs to be further improved in three aspects: parameter coding,information bottleneck of multiparameter estimation,and migration between training and deployment scenarios.Therefore,this paper focuses on deep learning delay and angle estimation methods to solve the problems in the above three aspects and improve the estimation performance.(1)The existing deep learning algorithm is mainly based on the end-to-end network structure,so it has the problems of too large influence of local variables and insufficient generalization for time delay estimation.When using deep learning with unique thermal coding to estimate time delay,it still cannot meet the requirements in terms of generalization and convergence speed.In order to solve the above problems,this paper proposes a delay estimation algorithm using sinc check delay values for sparse coding.Based on the sparsity and correlation of delay values,the target of neural network estimation is transformed into a sparse sinc kernel representation of delay values.Experiments and simulations show that the proposed algorithm is better than traditional MUSIC,OMP,SPICE and end-to-end deep learning algorithms in accuracy and stability.(2)Based on the multi-task joint estimation algorithm of deep learning delay and arrival angle,due to the more complex information contained in the multi-parameter,the hidden variable of the middle layer of deep learning contains insufficient information.In order to make the hidden variable contain as much effective information as possible,the multi-task variational information bottleneck algorithm is proposed for joint estimation of delay and Angle.The discrete hidden variable is transformed into a continuously distributed hidden variable,and the information of time delay and arrival angle contained in the intermediate layer hidden variable is maximized by maximizing the mutual information between the input signal and the hidden variable,which significantly improves the joint estimation accuracy of time delay and arrival angle.(3)Aiming at the deep learning wireless positioning problem,in order to solve the problem of the deep learning positioning accuracy decline caused by scene changes,firstly,a migration algorithm for the joint estimation network of time delay and arrival angle required by positioning is proposed.Then,considering the problem of catastrophic forgetting in the continuous migration process of multi-scene positioning network,the positioning accuracy of training scene is decreased.In order to solve this problem,an adaptive method to adjust the learning weight of importance parameters is proposed to solve the problem of catastrophic forgetting in transfer learning.Simulation experiments show that the proposed algorithm improves the positioning accuracy of source domain scene by 0.8m when there is no accuracy loss in target domain scene.