Channel Estimation And Optimization Of Cell-free Massive MIMO System

Cell-free massive MIMO system,as a new communication architecture that overturns the traditional cellular network architecture,has significantly improved performance compared to the traditional massive MIMO system.It is one of the key directions of the next generation of communication technology evolution.This system can provide communication services for a larger user community,and the increase in the number of users poses a challenge to the system’s channel estimation.Compared with traditional massive MIMO system,the resource consumption in the channel estimation phase of the system shows an explosive growth,which also becomes an important impediment to its widespread application.Therefore,it is of great practical significance to explore the channel estimation and optimization of this system.In this paper,the deep-learning idea is introduced,and the channel estimation problem is transformed into the picture processing problem by using the generation of antagonism network and convolution network and their extended models.A two-stage channel estimation scheme based on the generation of antagonism network and a two-stage channel estimation scheme based on convolution network are presented respectively.The specific research contributions are as follows:Firstly,an N2V-cGAN channel estimation scheme for twostage cell-free massive MIMO systems based on N2V denoising algorithm and cGAN is proposed to solve the problem of resource consumption in the channel estimation phase of cellfree massive MIMO systems,and the performance of the proposed scheme is verified by simulation.Generators of denoiser is based on U-Net structure,conditional generation is based on U-Net++structure,while discriminators of cGAN are based on CNN structure design.Compared with channel estimation schemes of the same type,N2V denoising algorithm makes the scheme independent of the data of noise image pairs or clean target images,achieves the recovery of single noise image,reduces the difficulty and cost of preparing training data,and is also suitable for special situations where it is difficult to obtain image pairs or clean target images.Generators and discriminators of cGAN can achieve better training effect through game,which makes channel estimation more accurate.The experimental results show that the proposed scheme performs better than the traditional channel estimation scheme and the original cGAN channel estimation scheme in the channel estimation of cell-free massive MIMO system.Within the range of SNR and pilot sequence length studied by us,the significant breakthrough of the proposed scheme is that the channel estimation performance of the proposed scheme is exponentially better than that of the traditional scheme in low SNR situations,which indicates that the proposed scheme can better adapt to harsh communication environment.At the same time,the scheme proposed with a shorter pilot sequence performs very well,which makes it possible to further save the pilot resources.Finally,the proposed scheme performs better when the antenna size is large,which makes it possible to expand the base station antenna size to achieve better communication performance.Then,in order to further improve the channel estimation effect,a FSRCNN-DnCNN channel estimation scheme for twostage cellular-free large-scale MIMO systems based on the structure of FSRCNN and DnCNN in convolution neural network is proposed,which can reduce the consumption of pilot resources and make the channel estimation more accurate,and the performance of the scheme is verified by simulation.The scheme consists of coarse channel estimation and denoising.The use of FRSCNN-based channel estimation model significantly improves the training speed and efficiency while guaranteeing the accuracy of channel estimation.The introduction of DnCNNbased denoising model makes channel estimation more accurate and channel performance better when the antenna size is large.The scheme uses a supervised learning based denoising algorithm with the real channel as the target,which significantly improves the quality of the estimated channel.The experimental results show that the proposed scheme performs better than traditional channel estimation scheme and N2V-cGAN scheme in channel estimation of cell-free massive MIMO system.At the same time,the training efficiency is similar to that of the previous section,which solves the problem that the denoising effect in the previous section decreases slightly because the channel information is not introduced into the denoising process,so the estimation effect is improved in all situations.