[Paper Review] AI-enabled STAR-RIS aided MISO ISAC Secure Communications

Zhengyu Zhu, Mengfei Gong+3 more|arXiv (Cornell University)|Feb 26, 2024

Space Satellite Systems and Control4 citations

TL;DR

This paper proposes an AI-driven design for a STAR-RIS-aided MISO ISAC system to maximize the long-term average secrecy rate of legitimate users while ensuring sensing and communication quality. By jointly optimizing beamforming, receive filtering, and STAR-RIS coefficients using deep reinforcement learning (DDPG and SAC), the system achieves superior security and spectral efficiency compared to conventional RIS and double-spliced RIS benchmarks.

ABSTRACT

A simultaneous transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) aided integrated sensing and communication (ISAC) dual-secure communication system is studied in this paper. The sensed target and legitimate users (LUs) are situated on the opposite sides of the STAR-RIS, and the energy splitting and time switching protocols are applied in the STAR-RIS, respectively. The long-term average security rate for LUs is maximized by the joint design of the base station (BS) transmit beamforming and receive filter, along with the STAR-RIS transmitting and reflecting coefficients, under guarantying the echo signal-to-noise ratio thresholds and rate constraints for the LUs. Since the channel information changes over time, conventional convex optimization techniques cannot provide the optimal performance for the system, and result in excessively high computational complexity in the exploration of the long-term gains for the system. Taking continuity control decisions into account, the deep deterministic policy gradient and soft actor-critic algorithms based on off-policy are applied to address the complex non-convex problem. Simulation results comprehensively evaluate the performance of the proposed two reinforcement learning algorithms and demonstrate that STAR-RIS is remarkably better than the two benchmarks in the ISAC system.

Motivation & Objective

• Address the challenge of secure, high-rate communication in high-frequency ISAC systems with limited coverage and high pathloss.
• Overcome the limitations of conventional RIS, which only reflect signals and serve users on one side, by leveraging STAR-RIS for 360° coverage.
• Simultaneously satisfy sensing requirements (echo SNR threshold) and communication rate constraints in a dual-secure ISAC environment.
• Maximize the long-term average secrecy rate of legitimate users through joint optimization of BS beamforming, receive filtering, and STAR-RIS transmission/reflection coefficients.
• Tackle the non-convex, time-varying optimization problem caused by dynamic channel states using sample-efficient deep reinforcement learning algorithms.

Proposed method

• Employ a STAR-RIS that splits the incident signal into transmitted and reflected parts using either energy splitting (ES) or time switching (TS) protocols.
• Formulate a non-convex optimization problem to maximize the long-term average secrecy rate under minimum echo SNR and user rate constraints.
• Apply deep deterministic policy gradient (DDPG) and soft actor-critic (SAC) algorithms—off-policy deep reinforcement learning methods—to learn optimal control policies for beamforming, receive filtering, and STAR-RIS coefficients.
• Use neural networks to approximate the policy and value functions, enabling end-to-end joint optimization of system parameters in a dynamic environment.
• Train the agents using experience replay and target networks to improve stability and sample efficiency, especially under high-dimensional action spaces.
• Ensure fairness and convergence by normalizing state and action spaces and tuning hyperparameters such as learning rate and entropy coefficient.

Experimental results

Research questions

• RQ1 How does the integration of STAR-RIS enhance the secrecy rate and coverage in MISO ISAC systems compared to conventional RIS?
• RQ2 Can deep reinforcement learning effectively solve the non-convex, time-varying optimization problem in dynamic ISAC environments?
• RQ3 What is the performance gain of STAR-RIS over double-spliced RIS and conventional RIS in terms of average secrecy rate?
• RQ4 How do the TS and ES protocols for STAR-RIS impact system secrecy rate and resource allocation under varying SNR constraints?
• RQ5 Which deep reinforcement learning algorithm—DDPG or SAC—achieves better performance and convergence in this ISAC security optimization task?

Key findings

• The proposed STAR-RIS-aided ISAC system achieves significantly higher average secrecy rates than both conventional RIS and double-spliced RIS benchmarks, demonstrating the superiority of 360° coverage and full-signal manipulation.
• The SAC algorithm outperforms DDPG in terms of final reward and convergence stability, achieving higher average secrecy rates due to its maximum entropy regularization and dual Q-network design.
• DDPG converges faster than SAC, but SAC yields better final performance, indicating a trade-off between training speed and solution quality.
• Increasing the number of STAR-RIS elements (N) improves secrecy rate performance for both algorithms, with a more pronounced gain under SAC.
• The average running time per episode increases with N, and SAC incurs higher computational cost than DDPG due to its more complex network architecture.
• Higher BS transmit power increases the average secrecy rate, but the gain diminishes at high power levels, indicating diminishing returns.
• The ES protocol results in lower secrecy rates than TS, especially when echo SNR exceeds 16 dB, where communication security is compromised due to excessive resource allocation to sensing.