Research On Channel Estimation Of Symbiotic Radio Network Based On Deep Learning

With the rapid development of wireless communication and the Internet of Things,the fifth-generation mobile communication technology(5G)is expected to achieve universal wireless connectivity of at least 100 billion devices,which has also increasingly highlighted the problems of lack of spectrum resources and excessive energy consumption,and has become an important factor limiting the continued development of 5G.In recent years,Symbiotic Radio(SR),reconfigurable intelligent surface(RIS),and deep learning technologies have provided new ideas to solve these problems.RIS is composed of a large number of passive reflection units,which can achieve low-cost and low-power symbiotic radio networks,improving the spectrum and energy efficiency of wireless networks.However,it is crucial to obtain accurate channel state information(CSI)in order to fully utilize RIS assisted communication,and the high-dimensional cascaded channels of symbiotic radio systems make obtaining CSI in SR systems a challenge.Based on the above reasons,this article focuses on combining deep learning technology with RIS based SR systems,studying the pilot design and channel estimation issues of this system,reducing pilot overhead and improving channel estimation performance through deep learning.The main work and achievements of this article are as follows:First of all,In multi user SR systems,in view of the limitations of existing pilot design methods that are difficult to solve inter user interference,this paper proposes a non orthogonal pilot designer based on Extreme Learning Machines(ELM),which can not only effectively reduce pilot overhead,but also reduce inter user interference and improve the accuracy of channel estimation.Further,in order to solve the problem of ELM hidden layer node redundancy,this paper uses Particle Swarm Optimization(PSO)to optimize ELM hidden layer node parameters.Simulation results show that the proposed non orthogonal pilot designer can achieve similar performance to orthogonal pilots with shorter pilot lengths,and effectively reduce inter user interference.Secondly,Current deep learning methods generally use deterministic loss functions,which greatly limits the performance of models in channel estimation tasks.Therefore,this paper proposes a channel estimator for SR systems based on Conditional Generated Adversary Nets(CGAN),which learns adaptive loss functions according to different tasks and data sets to correctly train the network.In order to further improve channel estimation performance,this paper aims to minimize the mean square error of channel estimation and jointly optimize the proposed pilot designer and channel estimator.Simulation results show that this joint optimization method can maintain good channel estimation performance in environments with smaller pilot size and lower signal-to-noise ratio,improving the accuracy and robustness of channel estimation.