Deep Learning-Based Transmit Power Control for Wireless-Powered Secure Communications With Heterogeneous Channel Uncertainty

Abstract

In this paper, we investigate a method of transmit power control (TPC) for wireless-powered secure communications with untrusted energy harvesting receivers (EH-Rxs), which are allowed to harvest energy but not to decode information from the signals sent by transmitters. In practice, channel state information (CSI) is sometimes inaccurate due to the time-varying nature of wireless channels and quantization errors in the CSI feedback. Furthermore, the CSI on the EH link between transmitters and EH-Rxs is likely to be more inaccurate than the CSI on the signal link when the EH-Rxs act deceitfully, so the level of uncertainty in the CSI can be different for the signal link and the EH link. In the presence of this heterogeneity of channel uncertainty, a deep learning (DL)-based TPC strategy is proposed to maintain the confidentiality of information at the untrusted EH-Rxs whilst guaranteeing that the required energy can still be harvested. In particular, the modified CSI for each signal and EH link, which is generated from the estimated CSI through the addition of random noise, is taken into account in the training of a deep neural network (DNN) to compensate appropriately for the heterogeneous errors in the CSI. Simulation results confirm that the proposed scheme provides a good approximation to the optimal TPC strategy, even in the presence of severely heterogeneous channel errors, such that it outperforms conventional baseline schemes and shows a near-optimal secrecy performance whilst achieving a significantly lower computation time.