FPGA-accelerated mode decomposition for multimode fiber-based communication

Mode division multiplexing (MDM) using multimode fibers (MMFs) is key to meeting the demand for higher data rates and advancing internet technologies. However, optical transmission within MMFs presents challenges, particularly due to mode crosstalk, which complicates the use of MMFs to increase syst...

Full description

Saved in:
Bibliographic Details
Main Authors: Qian Zhang, Yuedi Zhang, Juergen Czarske
Format: Article
Language:English
Published: Light Publishing Group 2025-08-01
Series:Light: Advanced Manufacturing
Subjects:
Online Access:https://www.light-am.com/article/doi/10.37188/lam.2025.031
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849240161993359360
author Qian Zhang
Yuedi Zhang
Juergen Czarske
author_facet Qian Zhang
Yuedi Zhang
Juergen Czarske
author_sort Qian Zhang
collection DOAJ
description Mode division multiplexing (MDM) using multimode fibers (MMFs) is key to meeting the demand for higher data rates and advancing internet technologies. However, optical transmission within MMFs presents challenges, particularly due to mode crosstalk, which complicates the use of MMFs to increase system capacity. Quantitatively analyzing the output of MMFs is essential not only for telecommunications but also for applications like fiber sensors, fiber lasers, and endoscopy. With the success of deep neural networks (DNNs), AI-driven mode decomposition (MD) has emerged as a leading solution for MMFs. However, almost all implementations rely on Graphics Processing Units (GPUs), which have high computational and system integration demands. Additionally, achieving the critical latency for real-time data transfer in closed-loop systems remains a challenge. In this work, we propose using field-programmable gate arrays (FPGAs) to perform neural network inference for MD, marking the first use of FPGAs for this application, which is important, since the latency of closed-loop control could be significantly lower than at GPUs. A convolutional neural network (CNN) is trained on synthetic data to predict mode weights (amplitude and phase) from intensity images. After quantizing the model’s parameters, the CNN is executed on an FPGA using fixed-point arithmetic. The results demonstrate that the FPGA-based neural network can accurately decompose up to six modes. The FPGA’s customization and high efficiency provide substantial advantages, with low power consumption (2.4 Watts) and rapid inference (over 100 Hz), offering practical solutions for real-time applications. The proposed FPGA-based MD solution, coupled with closed-loop control, shows promise for applications in fiber characterization, communications, and beyond.
format Article
id doaj-art-7f6b47f169d3494dac7d06d3fd973f71
institution Kabale University
issn 2689-9620
language English
publishDate 2025-08-01
publisher Light Publishing Group
record_format Article
series Light: Advanced Manufacturing
spelling doaj-art-7f6b47f169d3494dac7d06d3fd973f712025-08-20T04:00:43ZengLight Publishing GroupLight: Advanced Manufacturing2689-96202025-08-016227228310.37188/lam.2025.031FPGA-accelerated mode decomposition for multimode fiber-based communicationQian ZhangYuedi ZhangJuergen CzarskeMode division multiplexing (MDM) using multimode fibers (MMFs) is key to meeting the demand for higher data rates and advancing internet technologies. However, optical transmission within MMFs presents challenges, particularly due to mode crosstalk, which complicates the use of MMFs to increase system capacity. Quantitatively analyzing the output of MMFs is essential not only for telecommunications but also for applications like fiber sensors, fiber lasers, and endoscopy. With the success of deep neural networks (DNNs), AI-driven mode decomposition (MD) has emerged as a leading solution for MMFs. However, almost all implementations rely on Graphics Processing Units (GPUs), which have high computational and system integration demands. Additionally, achieving the critical latency for real-time data transfer in closed-loop systems remains a challenge. In this work, we propose using field-programmable gate arrays (FPGAs) to perform neural network inference for MD, marking the first use of FPGAs for this application, which is important, since the latency of closed-loop control could be significantly lower than at GPUs. A convolutional neural network (CNN) is trained on synthetic data to predict mode weights (amplitude and phase) from intensity images. After quantizing the model’s parameters, the CNN is executed on an FPGA using fixed-point arithmetic. The results demonstrate that the FPGA-based neural network can accurately decompose up to six modes. The FPGA’s customization and high efficiency provide substantial advantages, with low power consumption (2.4 Watts) and rapid inference (over 100 Hz), offering practical solutions for real-time applications. The proposed FPGA-based MD solution, coupled with closed-loop control, shows promise for applications in fiber characterization, communications, and beyond.https://www.light-am.com/article/doi/10.37188/lam.2025.031multimode fiberfpgadeep learningmode demultiplexingclosed-loop communication.
spellingShingle Qian Zhang
Yuedi Zhang
Juergen Czarske
FPGA-accelerated mode decomposition for multimode fiber-based communication
Light: Advanced Manufacturing
multimode fiber
fpga
deep learning
mode demultiplexing
closed-loop communication.
title FPGA-accelerated mode decomposition for multimode fiber-based communication
title_full FPGA-accelerated mode decomposition for multimode fiber-based communication
title_fullStr FPGA-accelerated mode decomposition for multimode fiber-based communication
title_full_unstemmed FPGA-accelerated mode decomposition for multimode fiber-based communication
title_short FPGA-accelerated mode decomposition for multimode fiber-based communication
title_sort fpga accelerated mode decomposition for multimode fiber based communication
topic multimode fiber
fpga
deep learning
mode demultiplexing
closed-loop communication.
url https://www.light-am.com/article/doi/10.37188/lam.2025.031
work_keys_str_mv AT qianzhang fpgaacceleratedmodedecompositionformultimodefiberbasedcommunication
AT yuedizhang fpgaacceleratedmodedecompositionformultimodefiberbasedcommunication
AT juergenczarske fpgaacceleratedmodedecompositionformultimodefiberbasedcommunication