Research On Signal Estimation Based On Deep Learning In OFDM System

This paper mainly studies how to improve the channel estimation and use the deep learning algorithm in Orthogonal Frequency Division Multiplexing(OFDM)systems.In this article,we mainly improve the pilot-based channel estimation method.The traditional pilot-based channel estimation method uses channel characteristics and interpolation of pilot positions to estimate all channel state information(CSI).In this paper,the deep network is used instead of the interpolation estimation method to more accurately simulate the derivative.The association between frequency location channel information and all channel information.To solve the channel fading,we used the channel simulator of the University of Vienna to generate data and trained a deconvolutional neural network offline.From the experimental results,our deep model performs better than the minimum mean square error(MMSE)estimation algorithm in solving channel distortion and detecting transmission signals.At the same time,in order to reduce the channel estimation error under low SNR,two solutions are proposed in this paper.One is to modify and perfect the model,and add a learning dimension reduction layer at the end of the base model to design a new deep convolution model.Compared with the traditional channel estimation algorithm and the general deconvolutional neural network model,the model is more robust to channel estimation under low SNR.The other is to add the GAN structure to the base model and introduce new high-quality data from the perspective of information theory.This model is trained better.