| Orthogonal Frequency Division Multiplexing(OFDM)can effectively overcome frequencyselective fading and improve spectral efficiency.Meanwhile,Deep Learning(DL)has recently shown promising potential for application in wireless communication systems due to its outstanding performance in various fields.Against this background,this paper explores channel estimation algorithms based on deep learning in OFDM systems.The main contributions of this paper are as follows:Firstly,the OFDM system model and classical channel estimation algorithms are introduced.Then,simulation experiments are conducted on the amplitude spectrum,phase spectrum,and power spectrum of the Jakes channel model.In addition,the theoretical knowledge of deep learning is briefly described.Current deep learning-based channel estimation algorithms are summarized,and the feasibility and effectiveness of the deep learning-based channel estimation algorithm are studied by computer simulation of representative neural networks in existing works.Considering that accurate channel estimation in traditional channel estimation algorithms relies on sufficient pilot information,compressive sensing theory can achieve good estimation performance with less pilot overhead.Based on the OFDM system,a sparsity-based channel estimation algorithm using deep learning compressive sensing(DLCS)is designed in this paper.This algorithm first estimates the sparsity of the channel and then uses a compressive sensing reconstruction algorithm with known sparsity to estimate the channel response,thereby improving the feasibility of compressive sensing algorithms in communication.Simulation results show that the proposed DLCS algorithm not only saves pilot overhead compared to the traditional least squares algorithm but also achieves better estimation performance.Since the proposed DLCS algorithm combines deep neural networks and compressive sensing modules to obtain sparse channel responses,to reduce the complexity of the algorithm,this paper considers directly using deep learning to estimate the channel response.Therefore,a channel estimation algorithm based on attention-aided fully convolutional network(AFCN)is designed in this paper.The algorithm models the channel response as an image super-resolution problem in computer vision,where the result of the least squares algorithm is treated as a low-resolution image,and the proposed AFCN structure is used to recover the high-resolution image.The proposed AFCN uses channel attention to assign different weights based on the importance of feature map channels,improving the performance of the algorithm.Simulation results show that the channel estimation algorithm based on AFCN performs better than the linear least mean square error algorithm,and the algorithm can maintain good performance with less pilot overhead. |
