Intelligent Machine Learning–Based Routing with Feature Extraction for Optical Benes Networks

Optical Benes networks are effective switching architectures for high-capacity communication systems. However, conventional routing algorithms primarily emphasize connectivity while often overlooking path quality, which often results in severe transmission loss along worst-case paths. To address this limitation, we propose an intelligent routing framework that integrates a feature extraction module with the K-Nearest Neighbors (KNN) algorithm. The proposed method guides path selection more effectively and avoids worst-case routing scenarios through effective preprocessing and feature extraction from routing tables. A 30 Gbps PAM4 transmission system is simulated to evaluate the proposed approach. For performance comparison, conventional routing methods, as well as Support Vector Machine (SVM), and Convolutional Neural Network (CNN) routing methods are considered. The results reflect significant improvements in routing accuracy (from 55% to 72.85%) with KNN, which significantly outperforming the CNN (52.23%) and SVM (51.06%) approaches while achieving the lowest computational cost of all tested methods (0.1–1 ms per iteration). Furthermore, the proposed approach reduces the power penalties, enhances the Extinction Ratio (EXT), and lowers the Symbol Error Rate (SER). Analysis using eye diagrams confirms superior signal integrity at lower received power levels. These findings demonstrate that the feature-enhanced KNN routing algorithm is an efficient and intelligent solution that not only ensures connectivity but also optimizes path quality, paving the way for scalable, high-speed optical Benes networks.