Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces the link-adaptive digital twin (LA-DT) for hybrid-amplified ultra-wideband optical networks, addressing the challenges of accurate physical-layer modeling and robust generalized signal-to-noise ratio (GSNR) estimation. By decomposing the GSNR modeling task into three key power predictions and employing a novel neural network architecture with linear modulation layers, the LA-DT significantly enhances generalization across diverse scenarios. Experimental results show that the LA-DT achieves substantial improvements in prediction accuracy, with RMSE reductions of up to 58.4% compared to baseline methods, while also demonstrating strong adaptability to previously unseen conditions with limited data.
Achieving a 55.8% improvement in GSNR estimation accuracy, the LA-DT adapts rapidly to new scenarios with minimal data, revolutionizing physical-layer modeling in ultra-wideband networks.
Ultra-wideband optical networks represent a practical solution for expanding communication capacity and supporting emerging applications such as artificial intelligence data center interconnection and 6G-oriented intelligent networks. Accurate physical-layer modeling has become increasingly essential to ensure reliable ultra-wideband network operation and capacity optimization, particularly under the intensified inter-channel stimulated Raman scattering (ISRS) effect. This paper proposes the link-adaptive digital twin (LA-DT) of hybrid-amplified ultra-wideband links to overcome the generalization and speed limitations of existing modeling methods, achieving accurate physical-layer modeling as well as robust generalized signal-to-noise ratio (GSNR) estimation across diverse links. First, to address the heterogeneity of EDFAs, the GSNR modeling task is decomposed into three key power predictions, including amplified spontaneous emission (ASE), nonlinear interference (NLI), and signal powers before entering the EDFA. Second, to enhance generalization in cross-scenario power prediction, three dedicated DT models are developed based on a novel, to our knowledge, neural network architecture that introduces linear modulation layers (LMLs). Third, to enable rapid adaptation to previously unseen scenarios with only limited data, three domain discriminators are designed to guide the few-shot fine-tuning of the LMLs. Finally, the proposed LA-DT explicitly accounts for insertion loss induced by Raman amplifiers (RAs), thereby improving modeling reliability under practical deployment conditions. Results demonstrate that the LA-DT achieves accurate physical-layer modeling across 35 scenarios with diverse fiber lengths, launch powers, Raman pump powers, and insertion losses. Compared with the baseline method, the LA-DT reduces the average root-mean-square error (RMSE) of NLI, ASE, and signal power predictions to 0.151, 0.111, and 0.113 dBm, respectively, corresponding to improvements of 56.0%, 58.4%, and 52.7%, and achieves an average GSNR estimation RMSE as low as 0.114 dBm, representing a 55.8% improvement. Moreover, for 12 previously unseen scenarios, the LA-DT maintains high accuracy through domain-discriminator-guided few-shot fine-tuning with only 20 samples per scenario, achieving an average GSNR estimation RMSE of 0.159 dB and demonstrating strong adaptability and robustness.