基于深度学习的有源智能超表面通信系统

Active Reconfigurable Intelligent Surface-aided Deep Learning Communication Systems

智能超表面(Reconfigurable Intelligent Surface,RIS)作为未来无线通信系统中最受关注的物理层技术之一,开创了由适应环境到重构电磁传播环境的全新通信范式。然而由于“乘性衰落”效应,RIS在典型的通信场景中只能实现微不足道的容量增益,而这在许多现有工作中被广泛忽视。针对上述现象,有源RIS可以通过主动放大反射信号,有效克服“乘性衰落”的高路径损失。为此提出了一种基于端到端(End-to-End,E2E)学习策略的有源RIS辅助的通信系统。通过深度学习网络,可以联合优化基站(Base Station,BS)以及RIS处的预编码与功率分配,以及用户(User Equipment,UE)的合并矩阵设计,避免了传统方案交替优化带来的高复杂度。具体来说,利用三个深度神经网络(Deep Neural Network,DNN)分别实现BS的预编码矩阵,BS与RIS处功率分配以及UE端的合并矩阵设计,并利用一个可学习参数向量表征RIS中的相位设置。仿真结果表明,所提出的基于深度学习的有源RIS传输方案相对于传统的无源RIS通信方案与无RIS方案,实现了更优的误比特率(Bit Error Rate,BER)性能。

Reconfigurable Intelligent Surfaces(RIS)represent one of the most promising physical layer technologies for future wireless communication systems,creating a novel communications paradigm that evolves from adapting to environmental conditions to reconstructing electromagnetic propagation environment.However,due to the“multiplicative fading”effect,RIS can only achieve negligible capacity gains in typical communication scenarios,a fact widely overlooked in many existing studies.To address this,active RIS can effectively counteract the significant path loss of“multiplicative fading”by actively amplifying the reflected signals.In this paper,we introduce a communication system aided by an active RIS that employs an End-to-End(E2E)learning strategy.By using a deep learning network,we can jointly optimize the precoding and power allocation ratio at the Base Station(BS)and RIS,as well as the combiner matrix design at the User Equipment(UE),thus avoiding the high complexity resulting from the alternating optimization inherent in traditional schemes.Specifically,we utilize three Deep Neural Networks(DNN)to implement the precoding matrix and power allocation at BS,and the combiner matrix design on UE,and use a learnable parameter vector to characterize the phase shifts in RIS.Simulation results demonstrate that the proposed deep learning-based active RIS transmission scheme outperforms conventional passive RIS and no-RIS schemes in terms of Bit Error Rate(BER).