Research On Terahertz High Speed Transmission Technology Based On Deep Learning

With the gradual commercialization of 5G,millimeter wave spectrum resources will become increasingly scarce.At this time,terahertz,which is higher than millimeter wave band,will be the main trend of future communication development.In recent years,THz communication has been recognized as a promising technology that can provide enough spectrum resources and ultra-high data rate for the sixth generation wireless communi-cation system.Due to the serious path attenuation and molecular absorption of terahertz signal,long-distance communication will cause great damage to the signal strength,so the short-distance indoor scene is the most suitable scene for terahertz communication.However,terahertz wave is easily blocked by obstacles,resulting in communication in-terruption.For this reason,intelligent reflector(IRS)is considered as a promising technology.It can reduce the vulnerability of blocking and enhance the coverage of indoor scene by adjusting the discrete phase shift of IRS unit and interacting with the incident THz wave in a controllable way.Based on the hardware structure of graphene driven IRS,a THz MIMO communication system model assisted by IRS is established.Then,the channel estimation problem under this model is studied.Taking advantage of the strong sparsity of THz band channel,an algorithm based on compressed sensing sparse recovery is proposed to solve the channel estimation parameters.But this algorithm involves lattice,with the increase of the number of antennas,the complexity will increase exponentially.In order to further reduce the complexity,a two-stage neural network fitting sparse recovery process is built.The perfect CSI can be obtained by channel estimation algorithm.In order to max-imize the channel capacity,this paper transforms the nonconvex optimization problem of IRS phase matrix into a codebook search problem,and discusses the performance of several search algorithms.Finally,considering the non ideal situation,the hybrid beamforming matrix at both ends of BS and MS can not be obtained directly from the phase matrix of IRS,so this paper directly raises the problem to the codebook selection of BS hybrid beamforming matrix,IRS phase matrix and MS hybrid beamforming matrix at the same time,so that the channel capacity corresponding to the three optimal matrices obtained at the same time with the most optimal channel capacity.Obviously,this is a problem that can not be solved by traditional optimization algorithm.At the same time,the direct search algorithm is also very complex,because it involves the simultaneous optimization of three variables.In order to solve this problem,this paper proposes a multi task codebook selection network structure,which can make full use of the data characteristics of the input characteristic matrix through a two-stage network architecture,so as to simultaneously optimize the codebook The codebook selection of three matrices which maximize the channel capacity is obtained respectively...