Nested Optical System Models Combining Machine Learning and Behavioral Approaches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for modeling optical systems, such as fully numeric and semi-analytic approaches, face challenges in accuracy and computational efficiency, especially when applied to complex and interconnected optical networks.
Innovation Solution
The use of nested models that combine machine learning and behavioral approaches, decomposing a large system model into smaller sub-models, allows for improved isolation of cause and effect, enhanced accuracy, and reduced computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fully numeric approaches are used to solve electromagnetic wave propagation equations, then accuracy in accounting for non-linear effects is improved, but computational resources and time requirements increase rapidly
Solution Approach 1:
The patent segments the optical network into multiple spans, each with its own neural network model. This divides the large-scale system modeling problem into smaller, manageable sub-problems that can be solved independently and then combined, reducing overall computational complexity while maintaining accuracy.
Solution Approach 2:
The patent implements nested models where span-level neural network models are contained within a network-level model. Each span model captures local non-linear effects, and these nested span models collectively represent the entire network, allowing accurate modeling of complex interactions without requiring a single monolithic model.
2Productivity
If semi-analytic approaches are used to divide signal propagation into separate components, then computational efficiency is improved, but modeling accuracy is sacrificed
Solution Approach 1:
The patent merges machine learning capabilities with semi-analytic modeling approaches. Neural network models are trained to learn the complex non-linear relationships that traditional semi-analytic methods approximate separately, combining the computational efficiency of semi-analytic methods with the accuracy of data-driven approaches.
Solution Approach 2:
The patent transforms the modeling approach by changing from fixed analytical parameters to adaptive neural network parameters. The neural networks learn optimal parameter representations from training data, allowing the model to adapt to different network conditions and maintain accuracy across varying operating points while preserving computational efficiency.
3Adaptability or versatility
If a single entity machine learning model is used to model the complete optical system, then system-level prediction capability is improved, but complexity, data collection requirements, and computation increase
Solution Approach 1:
The patent segments the monolithic system-level model into distributed span-level models. Each span model operates independently with local data, eliminating the need for centralized collection of vast amounts of system-level data while maintaining the ability to predict overall network behavior through composition of span predictions.
Solution Approach 2:
The patent creates universal span models that can be applied to any span in the network regardless of specific local conditions. These multi-functional models handle various propagation effects and network configurations through a unified neural network architecture, reducing overall model complexity while maintaining versatility.
Data Source
AI summary
A method of modeling an optical system includes obtaining input data for a channel in an optical system that includes a transmitter, one or more spans, and a receiver; processing the input data with a transmitter sub-model, output data from the transmitter sub-model with one or more span sub-models, and output data from one or more span sub-models with a receiver sub-model; and providing output data for the channel based on the processing. The method can also include, prior to the receiving, training an optical system model for the optical system; and decomposing the optical system model into the transmitter sub-model, the one or more span sub-models, and the receiver sub-model.