Learning to Tune Optical WANs: A Field Deployment of Noise Models in Optical Networks
summary
Builds a Gaussian Noise model of a New York ISP optical backbone including amplifiers, fiber spans, ROADMs and transceivers and achieves under 4% modeling error. Treats the model as an opaque box and runs hyperparameter search over amplifier gain targets (0.5 dB steps, safety limits, Raman forced max) to find peak end-to-end SNR. Applies gains in a phased field deployment and records a stable 4.2 dB SNR improvement across three wavelengths and roughly 60 Gbps (10.4%) capacity gain.
AI-written summary · about
chapters (12)
- 0:00Optical WANs and research goals
- 0:59Key components of optical networks
- 2:46Signal power noise and SNR trade-offs
- 4:32Tuning multiple amplifiers
- 5:56Operator tuning methods and risks
- 6:40Physical view of the network
- 8:54Modeling with Gaussian noise abstraction
- 10:20EDFA and Raman amplifiers
- 12:07Modeling hurdles and accuracy results
- 12:51Optimizing amplifier configurations
- 13:54Deploying optimized configuration
- 14:55Results and capacity increase
related talks
17mDOTE: Rethinking (Predictive) WAN Traffic Engineering
10m20,000 Leagues Inside the Optical Fiber
09m20,000 Leagues Inside the Optical Fiber
13mEfficient Multi-WAN Transport for 5G with OTTER
14mOrbWeaver: Using IDLE Cycles in Programmable Networks for Opportunistic Coordination
24mWiring the Planet: Scaling Meta's Global Optical Network
16mDetecting Ephemeral Optical Events with OpTel
17mMeta’s Network Journey to Enable AI