Deep Learning Aided Secure Transmission in Wirelessly Powered Untrusted Relaying in the Face of Hardware Impairments

Limited power and computational resources make the employment of complex classical encryption schemes unrealistic in resource-limited networks, e.g., the Internet of Things (IoT). To this end, physical layer security (PLS) has shown great potential in securing such resource-limited networks. To further combat the power scarcity in IoT nodes, radio frequency (RF) based energy harvesting (EH) is an attractive energy source while relaying can enhance the energy efficiency and extend the range of data transmission. Additionally, due to deploying low-cost hardware, imperfections in the RF chain of IoT transceivers are common. Against this background, in this paper, we investigate an untrusted EH relay-aided secure communication with RF impairments. Specifically, the relay simultaneously receives the desired signal from the source and the jamming from the destination in the first phase. Hence the relay is unable to extract the confidential desired signal. The resultant composite signal is then amplified by the relay in the second phase by using the energy harvested from the composite signal followed by its transmission to the destination. Since the destination is the original source of the jamming, its effect can be readily subtracted from the composite signal to recover the original desired signal of the source. Moreover, in the face of hardware impairments (HWIs) in all nodes, maintaining optimal power management both at the source and destination may impose excessive computations on an IoT node. We solve this problem by deep learning (DL) based optimal power management maximizing the secrecy rate based on the instantaneous channel coefficients. We show that our learning-based scheme can reach the accuracy of the exhaustive search method despite its considerably lower computational complexity. Moreover, we developed an optimization framework for judiciously sharing HWIs across the nodes, so that we attain the maximum secrecy rate. To derive an efficient solution, we utilize a majorization-minimization (MM) algorithm, which is a particular instance in the family of successive convex approximation (SCA) methods. The simulation results show that the proposed HWI aware design considerably improves the secrecy rate.