Neural Network Based Resource Management And Localization Techniques For Millimeter-wave Heterogeneous Networks

With the increasing demand for communication service,wireless communication technologies are continuing to evolve.As a representative of high spectrum,millimeter-wave communication technology can make great contributions in Enhanced Mobile Broadband(e MBB),Ultra-Reliable Low Latency Communications(URLLC),Massive Machine Type Communications(m MTC).However,there are still many key technical challenges in millimeter-wave communications,and this thesis focuses on the following aspects,mainly including(1)Resource management issues using millimeter-wave multi-connectivity technology and integrated access backhaul technology.(2)Energy efficiency optimization issues using millimeter-wave multiconnectivity technology.(3)Precise millimeter-wave positioning issues in outdoor scenario.This thesis conducts an in-depth study of the above problems and proposes the corresponding solutions based on neural network technology to achieve good performance with much reduced computational complexity.First,a Pointer Network based neural network algorithm is proposed for the joint user association and resource allocation problem in millimeter-wave wireless communication systems considering multi-connectivity and wireless backhaul technologies.Specifically,based on optimal decomposition,this thesis decomposes the original problem into two subproblems,the user association subproblem and the spectrum resource allocation subproblem.A Pointer Network-based neural network algorithm is proposed to obtain the suboptimal solution of the problem.Simulation results show that the proposed algorithm has a fast convergence speed and can achieve good performance.Secondly,a Deep Unfolding based neural network algorithm is proposed for the multiconnectivity energy efficiency problem in millimeter-wave wireless communication systems.Specifically,this thesis first designs a three-stage iterative algorithm to solve the upper bound problem.After that,this thesis designs the neural network components and add learnable parameters to unfold the three-stage iterative algorithm.The proposed neural network algorithm can not only solve the above upper bound quickly,but also obtain the feasible solution of the original problem quickly.Simulation results show that the proposed algorithm has a fast convergence speed and is capable of achieving good performance and with excellent generalization performance.Finally,a Vision Transformer based neural network algorithm is proposed for the millimeterwave fingerprint localization problem in the line-of-sight and non-line-of-sight(LOS/NLOS)channel coexistence scenario.Specifically,this thesis constructs a beamforming fingerprint database using millimeter-wave power delay profile data.Then,this thesis performs data mining to extract location information from the fingerprint database.In order to extract the accurate location information,a fingerprint segmentation algorithm is first proposed to extract the noise information.After that,a millimeter-wave fingerprint localization model based on Vision Transformer neural network is proposed.Simulation results show that the proposed fingerprint segmentation algorithm can effectively suppress the noise of fingerprint,and the proposed Vision Transformer-based neural network algorithm can achieve high accuracy localization.The proposed algorithm can achieve better localization performance with fewer number of parameters compared with existing literature.