Deep-Reinforcement-Learning-Driven Secrecy Design for Intelligent-Reflecting-Surface-Based 6G-IoT Networks

The sixth-generation (6G) wireless communication has called for higher bandwidth and massive connectivity of Internet of Things (IoT) devices. The increased connectivity also demands advanced levels of network security, which are critical to maintain due to severe signal attenuation at higher frequencies. Intelligent reflecting surface (IRS) is an increasingly popular, efficient, solution to cater to higher data rates, better coverage range, and reduced signal blockages. In this article, an IRS-based model is proposed to address the issue of network security under trusted-untrusted device diversity, where the untrusted devices may potentially eavesdrop on the trusted devices. A mathematical design of the system model is presented, and an optimization problem is formulated. The secrecy rate of the trusted devices is maximized while guaranteeing Quality of Service (QoS) to all the legitimate, trusted, and untrusted devices. A deep deterministic policy gradient (DDPG) algorithm is devised to jointly optimize the active and passive beamforming matrices owing to the complex and continuous nature of action and state spaces. The results confirm a maximum gain of 2-2.5 times in the sum secrecy rate of trusted devices under the proposed model, as compared to the benchmark cases. The results also ensure the throughput performance of all trusted and untrusted devices. The performance of the proposed DDPG model is evaluated under meticulously selected hyper-parameters.

Automated summary

In order to meet the higher bandwidth and massive connectivity demands of the sixth-generation (6G) wireless communication, intelligent reflecting surface (IRS) has become an efficient solution to improve data rates, coverage range, and signal blockages. This article proposes an IRS-based model to address network security issues in the presence of trusted-untrusted device diversity, maximizing secrecy rates for trusted devices while maintaining Quality of Service (QoS) for all legitimate devices. A deep deterministic policy gradient (DDPG) algorithm is utilized for joint optimization of active and passive beamforming matrices. Experimental results demonstrate a significant improvement in secrecy rates and throughput performance compared to benchmark cases, validating the effectiveness of the proposed model.