Research On Power Optimization Of Wireless Relay Communication System

At present,communication technologies have been applied to all aspects of social life,and wireless relay communication technology is the research focus in the field of communication.Compared with traditional wireless communication system,wireless relay communication system have more flexible application scenarios,broader range of signal coverage and better communication quality.On the other hand,the orthogonal frequency division multiplexing(OFDM)technology have the characteristics of high spectrum resource utilization and strong resistance to frequency selective fading.Therefore,the combination of relay and OFDM technology can significantly improve the communication performance of wireless communication systems.However,the introduction of new relay nodes will inevitably increase the power consumption of the wireless communication system,so the power allocation problem becomes particularly important.The thesis proposes a low-complexity power allocation algorithm to effectively reduce the total power consumption of the relay-assisted OFDM wireless system.The main research content of the thesis is as follows:(1)The thesis studies the power allocation problem of an OFDM communication system with three single-antenna nodes.Subject to user's quality of service requirements,we jointly optimize the transmission power of the source node and the relay node in different subcarriers,so that the system total power consumption can be minimized.This is a complicated non-convex problem with multiple tightly-coupled variables,so it is difficult to obtain the global optimal solution directly.Therefore,the thesis firstly apply the Lagrangian Relaxation algorithm to the power minimization problem to obtain multiple decouped nonconvex problems;then,the thesis introduces the block successive upper-bound minimization method to alternately optimize decoupled power variable,and finally we can obtain a high-quality sub-optimal solution of original problem.At last,the numerical simulations show the optimized performance of this algorithm.(2)The thesis considers using deep neural network(DNN)to learn the iterative optimization algorithm proposed in(1),the main idea is try to learn its input/output relation by neural network structure.First,we need to obtain the dataset required by the DNN.Secondly,the thesis determines the network structure,selects the activation function,learning rate and batch size.Next,In order to improve the learning speed and fitting accuracy,the thesis further optimizes the network from three aspects: improving the loss function,selecting the optimal number of hidden layers and its nodes.Finally,the thesis compares the data predicted by DNN and the data obtained by optimization algorithm proposed in(1),and compares the running time of the two methods to prove the efficiency and real-time nature of the neural network.