Research Article
A Lightweight 1D-CNN with Adaptive Learning for Multi-Class QoT Impairment Diagnosis
@INPROCEEDINGS{10.1007/978-3-032-34048-1_2, author={Mo-Han Liu and Chao Wang and Xian-Ming Chen}, title={A Lightweight 1D-CNN with Adaptive Learning for Multi-Class QoT Impairment Diagnosis}, proceedings={Smart Grid and Internet of Things. 9th EAI International Conference, SGIoT 2025, Taichung, Taiwan, December 6--7, 2025, Proceedings}, proceedings_a={SGIOT}, year={2026}, month={10}, keywords={Optical Communication Qot Prediction 1d-Cnn Dynamic Learning Rate Fault Diagnosis}, doi={10.1007/978-3-032-34048-1_2} }- Mo-Han Liu
Chao Wang
Xian-Ming Chen
Year: 2026
A Lightweight 1D-CNN with Adaptive Learning for Multi-Class QoT Impairment Diagnosis
SGIOT
Springer
DOI: 10.1007/978-3-032-34048-1_2
Abstract
This paper presents a lightweight one-dimensional convolutional neural network (1D-CNN) model designed for multi-class quality of transmission (QoT) fault diagnosis in optical communication systems. The pro-posed model incorporates a dynamic learning rate scheduler and a probabilistic data augmentation technique, which collectively enhance diagnostic robustness and convergence stability. Extensive experimental results demonstrate an overall accuracy exceeding 97%, with the precision for each individual class surpassing 0.97. The superior generalization capability and fault discriminability of the model are further corroborated by normalized confusion matrices and PR/ROC curves. Compared to conventional CNN baseline models, the proposed approach achieves more efficient feature extraction while significantly mitigating overfitting. Future work will focus on model pruning and deployment on embedded systems to facilitate real-time QoT monitoring.