IRS-aided Communications Without Channel State Information Relying on Deep Reinforcement Learning
IRS-aided Communications Without Channel State Information Relying on Deep Reinforcement Learning
复制标题
DOI：
10.1109/globecom48099.2022.10001413
复制
发表时间：
2022-12
期刊：
影响因子：
--
通讯作者：
Hiroaki Hashida;Y. Kawamoto;Nei Kato;M. Iwabuchi;T. Murakami
中科院分区：
文献类型：
--
作者：
Hiroaki Hashida;Y. Kawamoto;Nei Kato;M. Iwabuchi;T. Murakami
由众多无源元件组成的智能反射面（IRS）被认为是一种很有前途的智能无线通信技术。然而，由于IRS的被动特性，对信道状态信息进行显式估计以适当调整其反射系数具有挑战性。本研究提出了一种基于深度强化学习的算法，该算法从无线环境中学习基站（BS）的预编码向量和IRS相移来解决这一问题。我们为该算法开发了一个基于波束模式的学习框架，该框架间接地将无线环境映射到相移，以管理由BS和IRS的多个元素引起的大型状态-动作空间。仿真结果表明，该算法能够从环境中学习，建立传输策略，提高用户的传输速率。此外，结果验证了基于波束模式学习框架的算法比直接建立相移映射的方法更有效，并且可扩展到IRS元素的数量。
An intelligent reflecting surface (IRS), which comprises numerous passive elements, is considered a promising technology for smart wireless communication. However, the passive characteristics of an IRS render the explicit estimation of the channel state information to appropriately adjust its reflection coefficient challenging. This study proposes a deep reinforcement learning-based algorithm that learns the precoding vector of the base station (BS) and the IRS phase shift from the wireless environment to address this problem. We develop a beam-pattern-based learning framework for this algorithm that indirectly maps the wireless environment to the phase shift to manage the large state-action space caused by the multiple elements of the BS and IRS. Based on the simulation results, the proposed algorithm can learn from the environment and establish a transmission strategy that improves the user's transmission rate. Furthermore, the results validate that the proposed algorithm based on the beam pattern learning framework is more efficient and scalable to the number of IRS elements compared to the method that directly builds a mapping to the phase shift.