Vision transformers motivating superior OAM mode recognition in optical communications

Area of Science:

• Optical Communications
• Artificial Intelligence
• Signal Processing

Background:

• Atmospheric turbulence (AT) severely degrades signal strength and causes intermodal crosstalk in free-space optical (FSO) communications.
• This degradation reduces the accuracy of orbital angular momentum (OAM) mode detection, hindering FSO system performance.
• Accurate OAM mode recognition is crucial for enhancing data transmission capacity in FSO systems.

Purpose of the Study:

• To introduce a novel deep learning approach for accurate OAM mode recognition in FSO communications.
• To address the challenge of reduced OAM mode detection accuracy caused by atmospheric turbulence.
• To demonstrate the effectiveness of the proposed method in mitigating turbulence-induced signal impairments.

Main Methods:

• A deep learning model based on vision transformers (ViT) was developed for OAM mode recognition.
• The ViT model was pretrained on the large-scale IMAGENET dataset.
• The model was fine-tuned using a specific dataset of OAM beams subjected to varying strengths of atmospheric turbulence.

Main Results:

• The ViT-based method achieved nearly 100% accuracy in recognizing multiple OAM modes under weak-to-moderate turbulence.
• High accuracy (approximately 98%) was maintained even under strong turbulence conditions over long transmission distances (Cn2=1×10^-14).
• The advanced self-attention mechanism in ViT effectively captures global information, enhancing robustness.

Conclusions:

• The proposed ViT-based deep learning method offers a robust solution for detecting complex OAM beams in FSO systems.
• This approach effectively mitigates the adverse effects of atmospheric turbulence on OAM mode recognition.
• The findings pave the way for more reliable and high-performance OAM-based free-space optical communication systems.