Deep-learning surrogate model for rapid quantum-noise computation in dual-pump fiber-optic phase-sensitive amplifiers
Abstract
Fiber-optic parametric amplifiers in a phase-sensitive amplifier (PSA) configuration are well known for their low noise properties. Quantum mechanics dictates that these amplifiers can have a 0 dB noise figure (NF) (ratio of signal-to-noise ratio at the input and output of the amplifier) under certain conditions. However, for a dual-pump configuration, when the system is highly nonlinear, fully quantum calculations of NF become non-trivial. In such cases, semi-quantum models of NF evaluation exists, however, they are often complicated and resource intensive. To simplify that, in this work, first we use a semi-quantum NF calculation framework for the generation of numerically simulated input-output dataset of NF over a large parameter spaces. Then we use the data to train an optimized deep neural network (DNN) for rapid NF prediction. Utilizing the DNN, we demonstrate about 400 times reduction in the computation time compared to the standard semi-quantum NF evaluation approach with a mean absolute error (MAE) of 0.0015 dB. Using clustering we demonstrate a rapid exploration of the parameter space with respect to high gain (20 dB) and low NF (0.3 dB) targets and retrieve associated design parameters. These results illustrate the efficacy of machine learning-based approaches for rapid search of high-dimensional parameter spaces especially in compute intensive photonic systems.
Related
- Machine Learning Surrogate Model for Accelerating InGaAs/GaAs Quantum Well Laser Design
- Towards Surrogate Based Dequantization of Quantum Reinforcement Learning
- Operator learning for efficient quantum computation
- When Similarity Is Interaction-Driven: Quantum Kernels for Regime-Sensitive Learning
- Fault-tolerant quantum computation with static atomic buses
BibTeX
@misc{bouasria2026deeplearning,
title = {{Deep-learning surrogate model for rapid quantum-noise computation in dual-pump fiber-optic phase-sensitive amplifiers}},
author = {Yousra Bouasria and Debanuj Chatterjee and Abdessamad Belfakir and Debabrota Basu},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.08437},
eprint = {2611.08437},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.08437},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Deep-learning surrogate model for rapid quantum-noise computation in dual-pump fiber-optic phase-sensitive amplifiers
AU - Yousra Bouasria
AU - Debanuj Chatterjee
AU - Abdessamad Belfakir
AU - Debabrota Basu
PY - 2026
DA - 2026/11/05
PB - arXiv
DO - 10.48550/arxiv.2611.08437
UR - https://arxiv.org/abs/2611.08437
AB - Fiber-optic parametric amplifiers in a phase-sensitive amplifier (PSA) configuration are well known for their low noise properties. Quantum mechanics dictates that these amplifiers can have a 0 dB noise figure (NF) (ratio of signal-to-noise ratio at the input and output of the amplifier) under certain conditions. However, for a dual-pump configuration, when the system is highly nonlinear, fully quantum calculations of NF become non-trivial. In such cases, semi-quantum models of NF evaluation exists, however, they are often complicated and resource intensive. To simplify that, in this work, first we use a semi-quantum NF calculation framework for the generation of numerically simulated input-output dataset of NF over a large parameter spaces. Then we use the data to train an optimized deep neural network (DNN) for rapid NF prediction. Utilizing the DNN, we demonstrate about 400 times reduction in the computation time compared to the standard semi-quantum NF evaluation approach with a mean absolute error (MAE) of 0.0015 dB. Using clustering we demonstrate a rapid exploration of the parameter space with respect to high gain (20 dB) and low NF (0.3 dB) targets and retrieve associated design parameters. These results illustrate the efficacy of machine learning-based approaches for rapid search of high-dimensional parameter spaces especially in compute intensive photonic systems.
ER -