可解释机器学习驱动的智能超表面安全通信方法

Explainable Machine Learning Driven RIS Assisted Secure Communication

电磁环境的不确定性和不可操控性对无线通信安全带来复杂挑战,智能超表面因其能够灵活调控无线传播环境已被广泛应用于安全通信领域。尽管通过高效可靠的波束成形算法设计,智能超表面能够大幅提升无线通信系统的安全性能。但是,传统的高复杂度安全波束成形算法无法满足实时调控的现实需求;与此同时,基于人工智能算法的安全波束成形算法缺乏必要的可解释性。针对上述挑战,提出一种可解释深度强化学习方法以增强智能超表面安全通信决策的实时性和可解释性。具体地,首先给出了一种鲁棒近端策略优化算法,以解决在非法节点信息未知条件下的最优安全策略生成问题;然后,通过引入级联决策树作为策略函数逼近器,进一步增强策略可解释性。仿真结果表明,在非法节点存在显著不确定性的无线环境下,所提算法能够实时生成可解释的安全策略并提升频谱效率。

RIS has been widely used in secure communication due to their ability to flexibly regulate wireless propagation environments.However,traditional secure beamforming algorithms cannot meet the practical needs of real-time regulation,and artificial intelligence based secure beamforming algorithms lack necessary interpretability.In response to the above challenges,this article proposes for the first time an interpretable deep reinforcement learning method to enhance the real-time and interpretability of secure communication decisions.Specifically,a robust proximal policy optimization algorithm was first proposed to solve the problem of optimal security policy generation under unknown illegal node information.Then,a cascaded decision tree was introduced as a policy function approximator to further enhance the interpretability of the policy.The simulation results show that in a time-varying environment with significant uncertainty in illegal nodes,the proposed algorithm can generate interpretable security policies in real-time to improve spectral efficiency.