基于GAN的广播通讯过程多链路信息融合方法

Multilink information fusion method for broadcast communication process based on Generative Adversarial Networks

为了降低广播媒体噪声干扰、提升融合效果,提出了基于生成对抗网络的广播通讯过程多链路信息融合方法。通过小波去噪的信道估计算法清除广播通讯过程中的多链路信息噪声,将各链路信息划分为高频信息和低频信息;构建级联生成对抗网络,第一级网络依据汇总后低频信息构建多链路信息的基本结构,输出初级融合信息,第二级网络以低频到高频的转换信息为辅助任务,使第一级网络初级融合信息的细节信息更饱满。实验结果表明,该方法的信道去噪估计均方误差低;融合效果受噪声干扰较小,且融合超调量始终低于规定阈值。

In order to reduce the noise interference of broadcast media and improve the fusion effect,a multi link information fusion method for broadcast communication process based on Generative Adversarial Networks(GAN)is proposed.Through the channel estimation algorithm of wavelet denoising,the multilink information noise in the process of broadcast communication is removed,and each link information is divided into high-frequency information and low-frequency information.Then,a cascade GAN is constructed.The first level network constructs the basic structure of multilink information according to the summarized low-frequency information and outputs the primary fusion information.The second level network takes the conversion information from low-frequency to high-frequency as the auxiliary task to enrich the detailed information of the primary fusion information of the first level network.The experiment results show that the mean square error of channel denoising estimation is low,the fusion effect is less disturbed by noise,and the fusion overshoot is always lower than the specified threshold.