Research On Terahertz Channel Estimation Method Based On Residual Conditional Generative Adversarial Networks

Wireless channel estimation efficiency is critical to the performance of wireless communication systems.Because water vapor attenuation,path loss and other problems will cause the signal to change before it reaches the receiving end,the task of channel estimation is to use the received signal to accurately estimate the model parameters of the channel model and improve the modulation efficiency of the output signal at the output end.And because the 6G network will use the terahertz frequency band as the transmission channel of wireless signals to solve the gradual shortage of 5G spectrum resources,it is of great significance to study the terahertz channel estimation.Terahertz channel estimation research is conducive to the construction of a stable and efficient 6G communication system.After integrating other innovative technologies,it can improve performance and user experience,such as human digital twins,high-speed Internet access in the air,and smart transportation.The application scenarios are very broad.This paper studies THz channel estimation in the V2I(Vehicle-to-Infrastructure)scenario based on the conditional generative adversarial network model.First of all,this paper constructs the application scenario of V2 I communication in outdoor sunny weather.The leaky wave antenna based on TE1(Lowest transverse-electric)mode is used as the antenna structure of this scenario to make the terahertz signal more efficient and undistorted during the propagation process.The terahertz signal at the receiving end and the channel model parameters in the transmission process are obtained by simulating the V2 I scenario,and the data is tested using the basic model.Second,improvements are made based on the base model test results.Use Res Block and Liner to replace and improve the network structure of the basic model,increase the network depth and prevent data distribution changes and network model degradation.Improve the generator loss function by adding a generative loss to improve the generation efficiency.Using the Focal loss function to replace the discriminator loss function improves the uneven discrimination.Then the structures of different Res Blocks are compared,the residual conditional generative adversarial network structure is determined,and the terahertz channel estimation experiment is carried out.Experiments show that the residual conditional generative adversarial network reduces the loss of terahertz channel estimation by 9.68% compared with the basic model,and the discriminator’s discriminant loss gap is reduced from about 10 to about0.5.Finally,the reasons for the insufficient stability of residual conditional generative adversarial networks are analyzed,and it is considered that there may be local mode collapse and generator gradient disappearance problems.Improvements are made by secondary splicing of condition information and improving data input methods.After comparing the terahertz channel estimation experiments with different transmission distances,different frequencies and different residual layers,it is proved that the improved model has better performance and stability,and the model is lighter and easier to deploy on the application side.The mean absolute percentage error is only 4.12%,which is 14.24% lower than the base model.And by comparing with other methods,the average absolute percentage error of the improved residual conditional generative adversarial network in this paper is at least 12.35% lower than other methods.The improved residual conditional generative adversarial network in this paper is used for terahertz channel estimation in V2 I communication application scenarios.The model has excellent performance and has certain application value.