基于改进DRQN的分布式无线网络动态频谱接入策略研究
Research on Dynamic Spectrum Access Strategy in Distributed Wireless Networks Based on Improved DRQN
王启凤 1徐伟强1
作者信息
- 1. 浙江理工大学信息科学与工程学院,浙江 杭州 310018
- 折叠
摘要
频谱资源作为通信发展的核心资源,面临着资源匮乏的问题,提高频谱利用率是解决这一问题的重要途径之一.提出了一种基于改进深度循环Q网络(Deep Recurrent Q-Network,DRQN)算法的分布式动态频谱接入策略,并研究了认知无线网络中具有带宽约束的多个非授权用户(Secondary Users,SU)在存在频谱感知缺陷和用户间干扰的情况下的动态频谱接入策略,认知用户需要在不干扰授权用户使用的前提下接入合适的频段.整体目标是减少用户间干扰,尽可能提高频谱利用率.所提算法采用双向长短期记忆(Bidirectional Long Short-Term Memory,Bi-LSTM)网络和注意力机制,以提高DRQN算法的性能.此外,针对强化学习存在的过估计问题,引入了交叉熵机制.仿真结果表明,所提算法具备优异的性能.
Abstract
As a core resource for the development of communications,spectrum resources are facing the problem of resource scarcity,and improving spectrum utilization is one of the important ways to address this problem.This paper proposes a distributed dynamic spectrum access strategy based on an improved DRQN(Deep Recurrent Q-Network)algorithm,investigates dynamic spectrum access strategies for multiple SUs(Secondary Users)with bandwidth constraints in cognitive wireless networks in the presence of spectrum perception defects and inter-user interference,where cognitive users need to access the appropriate frequency bands without interfering with the use of authorized users.The overall goal is to reduce inter-user interference and maximize spectrum utilization.The proposed algorithm employs Bi-LSTM(Bi-directional Long Short-Term Memory)network and attention mechanism to enhance the performance of DRQN algorithm.In addition,a cross-entropy mechanism is introduced to address the overestimation problem of reinforcement learning.Simulation results demonstrate that the proposed algorithm has fairly good performance.
关键词
动态频谱接入/深度强化学习/双向LSTM网络/注意力机制/交叉熵损失Key words
dynamic spectrum access/deep reinforcement learning/bi-directional LSTM/attention mechanism/cross-entropy loss引用本文复制引用
出版年
2025
NETL
NSTL
万方数据