Investigation of impairments separability in direct detection optical performance monitoring based on UMAP technique

Zhao Shen,Xiangye Zeng,Jingyi Wang,Jianfei Liu,Jia Lu,Jie Ma,Yilin Zhang,Baoshuo Fan

DOI: https://doi.org/10.1007/s10043-024-00878-4

IF: 1.2

2024-04-07

Optical Review

Abstract:This paper focuses on the channel impairments separability of two histogram-based features, asynchronous amplitude histograms (AAH) and asynchronous delay-tap plot (ADTP), commonly used in direct-detection optical performance monitoring (OPM) techniques. This paper presents an in-depth study of the conditions under which these two histogram features are applicable in OPM. These high-dimensional features, AAH and ADTP, are dimensionally reduced using a state-of-the-art data visualization algorithm called Uniform Manifold Approximation and Projection (UMAP) algorithm. After data visualization, it can be found these two histogram-based features have some limitations in distinguishing between different levels of impairments in some specific cases. These features cannot achieve high accuracy in monitoring optical performance in these given situations, no matter how complex the classifier is designed. Extensive simulation experiments were performed to study the classification performance of the two histogram features in the single and multiple impairments cases. The results show that both AAH and ADTP can be used to monitor cumulative dispersion (CD) and optical signal to noise ratio (OSNR) in the case of the single impairment. In addition, the monitoring performance of both features is better for dispersion in the case of multiple impairments coexistence, while both have limitations for OSNR monitoring. However, the anti-dispersion interference ability of ADTP is better than that of AAH. The plausibility of the study results is verified by estimating the channel impairments under different conditions using a deep neural network-based (DNN) identifier. The impairments separation visualization results of UMAP are highly consistent with the estimation results of the DNN-based classifier, achieving the interconnection of usefulness and practicality.

optics