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...
Saved in:
| Main Authors: | , , |
|---|---|
| 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!
|
| Summary: | 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. |
|---|---|
| ISSN: | 2689-9620 |