A Deep CNN Design for Secure Short-Packet Communications in WET IoT Networks with Multiple Eavesdroppers

This paper studies secure short-packet communications in wireless energy transfer Internet-of-Things (WET IoT) networks. The best node selection (BeNS) scheme is proposed to select the best IoT device for assisting the secure transmission between a source and a destination robot in the presence of multiple eavesdroppers. Due to the high complexity of analysis, a novel convolutional neural network (CNN) with multiple efficient processing blocks is designed to evaluate the secure performance of the system in a short execution time. Simulation results show that the proposed BeNS scheme outperforms the random node selection one in terms of secure block error rate (BLER) and secrecy throughput. Moreover, the proposed CNN framework achieves comparable secure BLER and secrecy throughput as the BeNS scheme but with a significant reduction in execution time. Furthermore, the proposed CNN model achieves the lowest root-mean-square error (RMSE) among existing models such as deep neural networks and conventional CNN in the considered dataset.