Research On The Energy Efficient Techniques In The Cooperative Communication Networks

Global mobile data traffic has recently seen a significant increase,and the wireless cellular networks need to expand the capacity accordingly.Energy Ef-ficiency(EE)of the systems and the users becomes one of the key performance indicators gradually as the wireless capacity increases dramatically.Since the MIMO relay systems can attain high rate with limited energy,they have the potential to enhance the EE.On the other hand,the cooperative radio resource management techniques like mobility load balancing and mobility robustness optimization enable satisfied transmissions by consuming appropriate amount of energy to exchange overheads.Hence,this dissertation mainly focuses the energy efficient MIMO-relay transmissions and cooperative mobility robust-ness optimization.Specifically,the contributions are summarized as follows.Firstly,the user-centric energy efficient precoding design scheme is pro-posed in the MIMO two-way relay systems to satisfy the different requirements of the spectral efficiency(SE)and EE for the users.Specifically,the proposed precoding scheme aims to maximize SE and EE of the users simultaneously,and achieves the feasible SE-EE trade-offs.In order to handle the performance degradation caused by the energy and information orthogonal receptions in the MIMO two-way relay systems,where the sources are energy-constrained,the joint power control and relay precoding scheme is proposed.In order to maxi-mize the system rates under the power constraints of the sources and the relays,the proposed scheme decomposes the original problems into two coupled sub-problems,i.e.,the power control and the relay procoding designs.Moreover,the global solutions with the closed forms for the power control sub-problems and the approximating sub-optimal solutions for the relay procoding design sub-problems are obtained.In further,the jointly one-phase transmission scheme of the sources and the relays is proposed to improve the system SE in the co-frequency co-time full-duplex MISO one-way relay-aided wireless powered communication net-works.The proposed scheme utilizes the energy recycling and co-frequency co-time full duplex techniques elegantly and to improve the performance sig-nificantly by enabling the two-hop information and energy transmissions to be conducted in the same frequency and time,where the proposed algorithm achieves the global optimal with low complexity.The precoder design scheme in the full-duplex MIMO one-way relay-aided wireless powered communica-tion networks is proposed.The proposed scheme derives the optimal singular matrices of the precoding matrices of the sources and relays theoretically,and obtains the optimal singular values by solving the decomposed sub-problems.Lastly,in order to solve the frequent handover(HO)problems,which may cause the throughput and EE decline,and get rid of the assumption of the full knowledge of the network dynamics,the data-driven energy efficient handover optimization is studied.Specifically,the two-layer framework with the coop-erative deep reinforcement learning is proposed.The centralized controllers in the top layer partition users into clusters with unsupervised learning,where the users in the different areas can have different mobility patterns.Afterwards,within each cluster,the asynchronous deep reinforcement learning scheme is developed to control the HO processes.In this scheme,the deep neural net-work(DNN)approximates the HO controller learned by each UE.Due to the generalization ability of the DNN.such function approximators can represent the whole state space with limited weights,which can avoid the degraded per-formance for the newly-arriving users during the learning transitions.Then two methods are proposed to utilize the pre-trained networks:on-line vs.off-line.In the on-line method,the users fetch the weights from the parameter servers as the nre-trained networks periodically and keep learning.While in the off-line method,the users treat the pre-trained networks as static controllers.Final-ly,the supervised learning is utilized in initializing the DNN controller and to accelerate the reinforcement learning.