Research On OFDM Signal Positioning Parameters Estimation Methods Based On Deep Learning

With the rapid development of the economy and continuous progress in technology,location information has become an indispensable part of people’s lives.It can be used not only for routine navigation and target tracking but also for emerging applications such as product recommendation,smart healthcare,and reverse car finding.All of these applications require high accuracy and real-time positioning.With the wide application of Orthogonal Frequency Division Multiplexing(OFDM)technology in various communication systems,how to quickly obtain highprecision positioning parameters(arrival angle,arrival time,etc.)from OFDM signals in complex and changing electromagnetic environments has a broad application prospect in the field of wireless positioning.Therefore,this paper studies the accurate estimation of the arrival angle and arrival time of OFDM signals and the related localization algorithms.The main work of this paper includes the following parts:1.Under low signal-to-noise ratio(SNR)conditions,traditional parameter estimation algorithms have large errors and are difficult to meet the requirements of high-precision positioning in 5G.This paper proposes a parameter estimation method based on a cascaded neural network.The method first uses a noising filter neural network to filter low SNR signals and then uses an estimation neural network to obtain high-precision parameter estimates.Simulation results show that the proposed method can effectively improve estimation accuracy for low SNR signals.2.In the case of dynamic channel conditions,the acquisition of signal features and the generalization ability of neural networks are the main factors limiting the accuracy of parameter estimation.This paper proposes a parameter estimation method based on fractional Fourier transform(FRFT)and deep residual network to address this problem.The method first uses FRFT to extend the channel frequency response(CFR)of OFDM signals and then designs a variant deep residual network based on the extended signal features for parameter estimation.The effectiveness and robustness of the proposed method are verified through simulation experiments under different channel conditions.3.Based on the proposed parameter estimation method,an adaptive Kalman filtering algorithm is proposed to solve the position coordinates of the mobile terminal.The method uses Bayesian variational inference to update the measurement noise covariance of the Kalman filtering equation in each filtering cycle and dynamically adjusts the state noise covariance by analyzing the working state of the neural network.The proposed method is validated through simulation experiments in different motion scenarios,and the 90th percentile of positioning error is within 0.6 meters,which can meet the requirements of sub-meter positioning in 5G.