Research On Channel Estimation Based On Deep Learning

In the communication physical layer centered on Orthogonal Frequency Division Multiplexing(OFDM),the receiver utilizes known pilot signals for channel estimation and obtains complete channel response by interpolation.This method is difficult to further improve estimation performance when dealing with increasingly complex dynamic channel scenarios.In recent years,due to the great success of deep learning in the field of artificial intelligence,many wireless communication researchers have tried to introduce deep learning into the physical layer technology of wireless communication.In this paper,the combination of OFDM channel estimation and deep learning is studied,aiming to improve the channel estimation performance and overcome the problems in traditional channel estimation algorithms through deep learning technology.The main work is as follows:First,this paper describes the similarity between the pilot-aided channel estimation algorithm and the image super-resolution reconstruction.By analyzing the characteristics of channel response image,this paper designs a new super-resolution neural network InceptionCENet for channel estimation.By introducing the Inception structure and global residual,the designed network has excellent feature extraction and noise reduction ability,and has less training parameters than other superresolution networks.The simulation results show that the performance of InceptionCENet-based channel estimation algorithm is far better than LS channel estimation algorithm.Compared to other current super-resolution networks,it has a performance gain of 3dB.It almost has the same channel estimation performance as MMSE and LMMSE.The algorithm does not depend on the prior knowledge of the channel,so it is a channel estimation algorithm with potential and practical significance.Secondly,this paper proposes a pilot pattern design scheme based on block principal component analysis(PCA)for InceptionCENet.In consider of the similarity between the selection of pilot position and the process of data dimensionality reduction,the pattern design scheme uses the block PCA algorithm to make the pilot pattern contain as many channel response features as possible,so as to obtain the optimal pilot pattern for InceptionCENet.The simulation results show that the block-based PCA pilot pattern design scheme can effectively improve the channel estimation performance of InceptionCENet compared with the equally-spaced pilot pattern.When the number of pilots is small,the performance improvement can reach 2dB.With the increasingly tight communication resources,the block based PCA pilot pattern design method can effectively reduce pilot overhead by about 30%,making it an effective pilot pattern design method for super-resolution channel estimation.