Research On Advanced Transceivers For Intelligent Communication

With the development of human civilization and the expansion of the scope of production and living,human needs for a wider range of communications,more timely connections,and deeper perceptions have accelerated the research on sixth-generation mobile communications(6G).The debate on the key technological changes is one of the core issues of the 6G concept discussion.The key direction involved includes intelligent communication technology.Intelligent connectivity with AI as the core will be the key to enable the 6G vision.This paper is oriented to the research on advanced transceiver design for intelligent commu-nication.First,message passing algorithms and the applications of deep learning in the wireless physical layer are studied.This paper introduces the derivation of a general message-passing algorithm that is a variation of the belief-propagation algorithm,and gives the derivation process and iterative steps of a simplified approximate message passing algorithm(AMP).In order to overcome the stability defect of AMP,a modified orthogonal approximation message passing algorithm(OAMP)is proposed.The following part introduces the application of deep learning algorithms in wireless communication,mainly including the role of deep learning in each module of the wireless physical layer by giving typical cases of deep learn-ing based wireless physical layer application.Secondly,this paper provides a solution of the 1-bit precoding problem in a multi-user multi-input multi-output system and proposes an iterative discrete estimation algorithm based neural network optimized advanced iterative precoding network.The preocding network optimizes the damping factor using a deep learning method by taking the damping factor as a trainable variable,and overcomes the problem that the discrete function cannot propagate gradients backwards.Numerical simulation results show that the adopted iterative discrete estimation algorithm has great advantages in iterative extreme performance,and the neural network optimized precoding network designed in this paper can achieve 2dB gains for bit error rate targets for 10-4 under 20 iterations,thereby reducing the computational complexity.Finally,this paper studies the advanced receiver architecture based on intelligent com-munication in a GFDM multi-carrier system.This paper proposes a channel estimation method based on OAMP firstly,which greatly improves the estimation performance in GFDM systems.Based on the iterative channel estimation and signal detection algorithms,model-driven networks OAMP-Est-Net and OAMP-Det-Net are designed respectively.The networks are designed by iterative unfolding,and fewer key parameters are selected for training.Numerical results show that the proposed advanced receiver algorithm can achieve better performance than the classic solution,and the model-driven network further improves the performance based on the advanced algorithm proposed.