Due to the inherent openness of wireless channels, data transmission in user-centric ultra-dense networks (UUDNs) is vulnerable to eavesdropping and malicious jamming attacks (EJA). This paper addresses the user-centric clustering, beamforming, and artificial noise design to combat with the EJA, and an accelerated deep reinforcement learning (DRL) based approach has been proposed for the downlink of UUDNs from a security, reliability and energy efficiency (SREE) perspective. In contrast to the existing approaches, the proposed approach divides the UE's communication behavior into association phase and transmission phase to perform a fast and adaptive response to the EJA. For the association phase, the proposed approach aims at achieving a fast response to EJA by optimization algorithms from a short-term perspective. For the transmission phase, we propose a DRL based algorithm to achieve optimal adaptive response by dealing with the dynamic and persistent effects of EJA from a long-term perspective. To accelerate the training process, fast response achieved in the association phase will be regarded as the initial experience to update the parameters of the deep neural networks adopted by the DRL based algorithm for transmission phase. Comparing with the existing methods, simulation results show that our proposed accelerated DRL approach achieves better performance in terms of SREE and convergence speed.

中文翻译：

---

用于节能的、以用户为中心的超密集网络的安全可靠的下行链路传输：加速 DRL 方法


由于无线信道固有的开放性，以用户为中心的超密集网络（UUDN）中的数据传输很容易受到窃听和恶意干扰攻击（EJA）。本文提出了以用户为中心的聚类、波束成形和人工噪声设计来对抗 EJA，并从安全性、可靠性和能源效率（SREE）的角度提出了一种基于加速深度强化学习（DRL）的 UUDN 下行链路方法。 ） 看法。与现有方法相比，所提出的方法将UE的通信行为划分为关联阶段和传输阶段，以对EJA执行快速且自适应的响应。对于关联阶段，所提出的方法旨在从短期角度通过优化算法实现对 EJA 的快速响应。对于传输阶段，我们提出了一种基于 DRL 的算法，通过从长期角度处理 EJA 的动态和持久影响来实现最佳自适应响应。为了加速训练过程，在关联阶段实现的快速响应将被视为更新基于 DRL 的传输阶段算法所采用的深度神经网络参数的初始经验。与现有方法相比，仿真结果表明，我们提出的加速 DRL 方法在 SREE 和收敛速度方面取得了更好的性能。