Deep Learning Assisted Relay Matching in Multi-user Pair and Multi-relay Untrusted Networks

In this paper, we study the relay matching based on deep learning in multiple user-pair and multiple untrusted-relay networks. In our previous work, Kuhn-Munkras (KM) algorithm was utilized with the goal of maximizing the system secrecy rate based on the weight design. However, the computational complexity of KM algorithm is very high with the number of user pairs and relays increased. Due to the fact that deep neural network (DNN) is capable of dealing with the complex nonlinear relationship and reducing computational complexity, the deep learning assisted maximum weight matching between user pairs and untrusted relays is adopted to address this issue. And batch normalization is used to accelerate the network convergence. The simulation results show that the proposed approach almost obtain the same average user secrecy rate as the conventional KM algorithm, while the system complexity is reduced.