MZI wavelength interleaving filter based on double microring structure

To improve the output characteristics of all-fiber Mach-Zehnder interferometer(MZI)-interleaver,a dual microring-assisted MZI interleaver is designed.According to its structure,the output expression of the device is derived from the signal flow diagram.After simulation analysis,the optimal structural parameters during the transmission process are obtained.In addition,the coupling coefficient and transmission loss of the coupler are analyzed.The results show that the improved interleaver output line wave is closer to the square wave,and its 25 dB cutoff band bandwidth and 0.5 dB passband bandwidth are significantly improved,with the values of 41.2 GHz and 18.9 GHz,respectively.The device has a certain resistantability to deviation,and the transmission loss has less influence on the extinction characteristics of the filter.

一种无需非线性结构的光学卷积神经网络

卷积神经网络在视觉处理方面具有独特的优势。最近的一些研究使用光学的卷积神经网络来实现更为快速和低功耗的图像处理系统。我们的工作提出了一种无需额外非线性结构的光学卷积神经网络,使用绝对值函数作为激活函数,设置了特定的三层卷积神经网络,该网络对MNIST和Fashion-MNIST数据集的识别准确率与目前普遍使用的激活函数相差不大。通过对软件仿真与硬件仿真的结果进行对比,发现图像经过第一层卷积和非线性后的结果误差不超过2%,输出结果验证了我们的光学卷积神经网络对图像处理的有效性。这为实现高效可编程的光学卷积神经网络提供了可行方案。