Abstract : In this paper, we study wireless energy transfer full-duplex (FD) Internet-of-Things (IoT) networks, where multiple FD IoT relays are deployed to assist short-packet communications between a source and a robot destination with multiple antennas in automation factories. Taking into account two residual interference (RSI) models for FD relays, we propose a full relay selection (FRS) scheme to maximize the end-to-end signal-to-noise ratio of packet transmissions aiming at improving the block error rate (BLER) and throughput of the considered system. Towards real-time configurations, we design a deep learning framework based on the FRS scheme to accurately predict the average BLER and system throughput via a short inference process. Simulation results reveal the significant effects of RSI models on the performance of FD IoT networks. Furthermore, the CNN design achieves the lowest root-mean-squared error among other schemes such as the conventional CNN and deep neural network. Furthermore, the DL framework can estimate similar BLER and throughput values as the FRS scheme, but with significantly reduced complexity and execution time, showing the potential of DL design in dealing with complex scenarios of heterogeneous IoT networks.
Index terms : Deep neural network , residual self-interference , short-packet communication , wireless power transfer