| Terahertz(THz)communications have been envisioned as one of the promising alternative technologies to provide ultra-high data transmission and sufficient spectrum resources for the sixth generation(6G).However,except for the obvious advantages,THz communications still face some severe challenges in practical application scenarios.On the one hand,due to the high frequency of THz waves,the path loss and the reflection loss of THz signals are very serious,and conventional reflecting surfaces(e.g.,wall,floor,wardrobe)are unable to meet high-quality communication requirements.On the other hand,since THz beam is narrow and strongly directional,THz signals are easily blocked by obstacles,and the coverage capability of THz signals is seriously weakened,leading to a THz communication interruption.To tackle this problem,intelligent reflecting surface(IRS)can smartly control the propagation direction of THz waves by adjusting phase shifts of reflecting elements,and converts the complex and changeable wireless communication environment into the intelligent and controllable propagation environment.By deploying IRS,THz signals can bypass obstacles,which improves coverage capability and spectrum efficiency of THz communications.This dissertation mainly investigates the organic combination of THz communication and IRS in the future 6G,and the specific research contents and contributions are listed as follows:1.This dissertation proposes an efficient graphene-based IRS to solve the complex hardware structure design problem at THz frequency band.To this end,the electric properties of the graphene are explored,and the relationship between conductivity and applied voltage is analyzed theoretically.Then,based on the Fabry-Perot resonance model,a three-layer dielectric IRS element structure is proposed and Fabry-Perot resonant cavity.Simulation results show that the phase shift response is controlled up to 306.82 degrees and the reflecting amplitude efficiency is more than 50%.In addition,this dissertation establishes the channel model of IRS,and transforms the channel estimation problem into the compressed sensing problem.Subsequently,a low-complexity iterative atom pruning based subspace pursuit(IAP-SP)algorithm is developed to dynamically reduce the atoms in the sensing matrix that are uncorrelated with the signal residual during the iterative process.Simulation results show that the proposed IAP-SP algorithm can approach the sparse recovery performance of the conventional subspace pursuit algorithm,and meanwhile greatly decreases computational complexity of the channel estimation.2.In the IRS-assisted THz single-user multiple-input multiple-output(MIMO)system,this dissertation develops adaptive gradient descent algorithm and alternating optimization algorithm to solve the joint design problem of active and passive beamforming.Firstly,the optimal beamforming matrices of base station and users are obtained by executing singular value decomposition for THz cascaded channel,and the adaptive gradient descent algorithm is utilized to optimize the phase shift matrix of IRS.In contrast with the conventional gradient descent algorithm,the adaptive gradient descent algorithm is able to dynamically update the step size during the iterative process,which is determined by the second-order Taylor expansion formulation.To reduce the calculation load brought by the gradient descent process,a low-complexity alternating optimization algorithm is proposed by alternately designing active beamforming via the beam vector approximation algorithm and the passive beamforming of the RIS via the linear search algorithm.Compared with the gradient descent algorithm,simulation results show that the proposed alternating optimization algorithm can achieve a better compromise between communication rate and computational complexity.3.In the IRS-assisted THz multi-user MIMO system,this dissertation proposes a multi-task learning based analog beam selection(MTL-ABS)framework to handle the analog beam classification problems at base station,IRS and users.Firstly,an iterative alternating search algorithm is developed,which approaches the optimal beam selection performance of the exhaustive search algorithm and greatly decreases the complexity.Then,based on the dataset generated by the iterative alternating search algorithm,the MTL-ABS framework is developed to further cut down the computation overhead of the beam selection,which employs residual network module and self-attention mechanism to combat the network degradation and mine intrinsic THz channel features,respectively.In addition,a blockwise network analysis method is designed to prove the convergence of the MTL-ABS structure.Simulation results show that the MTL-ABS framework is able to substantially reduce the complexity of analog beam selection and achieves the nearoptimal sum-rate performance in contrast with conventional heuristic search algorithms.4.In the IRS-assisted fully-connected THz cell-free MIMO system,this dissertation proposes an efficient alternating optimization algorithm to tackle active and passive beamforming design problems.Firstly,the nonconvex sum-rate maximization problem is converted into a sequence of tractable subproblems,which are respectively settled by alternating direction method of multipliers,subgradient decent algorithm and manifold optimization algorithm.Then,a novel partially-connected THz cell-free MIMO architecture is developed to reduce the number of communication links between base stations and users and to diminish communication costs.Specifically,the base station selection problem is transformed into a binary integer quadratic programming(BIQP),and the relaxed linear approximation algorithm is put forward to further convert the BIQP problem into a binary integer linear programming problem.Simulation results demonstrate that,compared with fully-connected THz cell-free MIMO system,the partially-connected THz cell-free MIMO system greatly reduces the communication overhead in the case of negligible sum-rate performance penalty. |
PDF Full Text Request