Securing Cognitive IoT Networks: Reinforcement Learning for Adaptive Physical Layer Defense

Single-input multiple-output (SIMO) configurations are commonly employed in applications where managing power consumption is critical. This is the case for battery-powered sensors and low-power Internet of Things (IoT) devices, which may also employ energy harvesting (EH) strategies to augment their battery longevity. These systems can further access the underutilized spectrum through cognitive radio networks (CRNs). This study focuses on mitigating eavesdropping concerns in such configurations through the application of physical layer security (PLS). More specifically, the paper proposes a PLS technique to evaluate and enhance the confidentiality of secondary users' (SUs) transmissions for SIMO underlay CRN, considering the presence of an eavesdropper. The secondary user receiver incorporates the power splitting-EH method to extract energy, subsequently utilizing it to generate jamming signals to perplex the eavesdropper. We compare the results obtained when the eavesdropper extracts energy from the SUs' transmissions and when it chooses to only decode the wiretapped messages. We implement a deep reinforcement learning (DRL) approach [1], specifically the deep Q-network, to optimize the transmission power and thereby maximize the secrecy rate of the SUs and demonstrate that the performance of the approach surpasses that of the benchmarks.