隐空间采样与隐蔽特征提取的CR-GAN复杂无线信道建模

CR-GAN Complex Wireless Channel Modeling with Hidden Space Sampling and Hidden Feature Extraction

为了更准确地建模随机无线信道,提出一种自适应增强条件生成对抗网络信道建模方法.其采用扩展的生成对抗网络(Generative Adversarial Network,GAN)开展训练,以近似估计无线信道响应,模拟真实无线环境信道.为了改善GAN训练稳定性和学习能力,引入条件信息和梯度惩罚项,并提出一种增强条件生成对抗网络框架,用于提取信道隐蔽特征.此外,还提出隐空间采样策略,以增加随机变量与生成数据的互信息量,提高所提框架的信道建模性能.仿真表明:所提框架能很好地模拟复杂无线信道分布.在信噪比为10 dB时,与现有GAN训练方法相比,其归一化均方误差性能改善约24%.

To accurately model random wireless channels,an adaptive channel modeling framework based on a strengthened conditional generative adversarial network(GAN)is proposed.It utilizes the extended GAN for training to approximately estimate the response of wireless channels and thus stimulate the actual wireless channels.To improve both the GAN training stability and learning capability,conditional information and gradient penalty terms are introduced.Besides,a strengthened conditional GAN frame,named condition reinforcement GAN(CR-GAN),is proposed to extract the essential hidden characteristics of wireless channels.In addition,a hidden space sampling strategy is utilized to increase the mutual information between the potential variables and generative data for the improved channel modeling performance of the proposed framework.Simulation results demonstrate that,at a signal-to-noise ratio of 10 dB,the proposed CR-GAN framework outperforms current GAN-based models by reducing 24% of the normalized mean squared error.