基于双注意力机制门控循环单元的光通信采样定时偏差估计方案

Sampling time offset estimation scheme based on dual attention gate recurrent unit for optical communication

在光通信中,由接收器的模数转换器产生的采样定时偏差(sampling time offset,STO)是接入网中需要估计的一个重要参数。随着现代信息传输量的增加,要求的传输速度也越来越快。现有的传统估计方法存在与升高的调制格式不兼容、估计误差大、估计时间长等问题。此方案提出了一种用于误差估计的双重注意机制门循环单元神经网络(dual attention gated recurrent unit,DAGRU)。DAGRU可以根据信号序列,从全局级别和群级别两个层面分别使用注意力机制,选择性地提取有用信息作为学习的特征。实验结果证明,DAGRU可用于多种调制格式的长距离传输,适用于1.25倍的采样率,抗噪声能力强,鲁棒性好。

In optical communication,the sampling time offset(STO)which is generated by the receiver’s analog-to-digital converter,is an important parameter to be estimated in the access network.As the volume of modern information transmission increases,the required transmission speed is also getting faster.The existing conventional estimation methods suffer from incompatibility with elevated modulation formats,large estimation errors,and long estimation times.This scheme proposes a dual attention gated recurrent unit(DAGRU)for offset estimation.DAGRU can selectively extract useful information as learned features based on signal sequences,using attention mechanisms at two levels,global-level and group-level,respectively.Experimental results demonstrate that DAGRU can be used for long-distance transmission in a variety of modulation formats,is suitable for sampling rates of 1.25 times,is highly resistant to noise,and has good robustness.