基于深度强化学习的认知物联网资源分配的策略研究

Cognitive Radio IoT Resource Allocation Strategy Based on Deep Reinforcement Learning

能量采集(Energy Harvesting,EH)和认知无线电(Cognitive Radio,CR)技术的组合可为物联网设备提供持续的能量,并有效地提高物联网系统的频谱效率。然而,在衬底模式下的认知物联网(Cognitive Radio IoT,CIoT)系统中,物联网设备之间的无线通信常常遭受窃听攻击。针对存在多窃听者条件下的CIoT系统无线通信场景,以保密速率作为系统保密性能指标。为解决所提的资源分配问题,将长短期记忆网络(Long-Term Memory Network,LSTM)、生成对抗网络(Generative Adversarial Networks,GAN)和深度强化学习(Deep Reinforcement Learning,DRL)算法相结合,设计一种联合能量采集时间和传输功率分配方案。数值仿真表明,与其他基准算法相比,所提方法能够有效地提高系统保密性能。

The combination of EH(Energy Harvesting)and CR(Cognitive Radio)technology can provide continuous energy for IoT devices and effectively improve the spectrum efficiency of IoT system.However,in the CIoT(Cognitive Radio IoT)system under underlay mode,wireless communications between IoT devices are often subject to eavesdropping attacks.For the wireless communication scenario of CIoT systems with multiple eavesdroppers,this paper takes secrecy rate as the secrecy performance index.To address the proposed resource allocation problem,this paper resorts to a combination of LSTM(Long Short-Term Memory Network),GAN(Generative Adversarial Network)and DRL(Deep Reinforcement Learning Algorithm),and then develops a joint EH time and transmission power allocation scheme.Numerical simulation results indicate that the proposed method can effectively improve the system secrecy performance compared with other benchmark algorithms.