Joint Relay Selection and Power Allocation Based on Deep Neural Network in the Cooperative Relay-Eavesdropper Channel

The traditional cooperative communication algorithms show inherent limitations facing massive data processing and ultra-high-speed communication requirements in the future communication scenarios. A joint relay selection and power allocation optimization scheme based on deep neural network is introduced modeling the relay selection and power allocation problems as multi-classification and regression problems respectively. Taking all the channel state information (CSI) as the input feature, a multi-layer neural network is designed to complete relay selection and power allocation tasks jointly. The simulation results show that the joint relay selection and power allocation scheme based on deep neural network (DNN) not only achieves the secrecy rate close to the semi-definite relaxation (SDR) -based method, but also greatly reduces the complexity of signal processing to promisingly realize real-time secure communication. The SDR-based algorithm takes 123510s on 3000 test examples while DNN-based scheme only takes 3.6s.