Mode detection of misaligned orbital angular momentum beams based on convolutional neural network

Area of Science:

• Optical Communications
• Machine Learning
• Signal Processing

Background:

• Free-space optical (FSO) communication utilizes orbital angular momentum (OAM) beams to enhance channel capacity.
• Beam misalignment is a significant challenge that degrades performance in OAM-based FSO systems.
• Deep learning offers robust pattern recognition capabilities for mitigating such errors.

Purpose of the Study:

• To propose and improve a deep learning-based neural network structure for mitigating misalignment errors in OAM-based FSO communication.
• To enhance the accuracy and reliability of OAM mode detection under various turbulence conditions.

Main Methods:

• A convolutional neural network (CNN) architecture is employed for OAM mode detection.
• Data augmentation techniques are utilized during the training phase to improve model robustness.
• A novel view-pooling layer is introduced to compress feature maps from multiple receiving angles, effectively handling misalignment.

Main Results:

• The proposed method demonstrates high accuracy in OAM mode detection, even with beam misalignment.
• Accuracy remains above 99% for tilt angles less than 35° under weak turbulence (Cn2=1×10^-15 m^-2/3).
• The system achieves 93% and 88% accuracy under medium (Cn2=1×10^-14 m^-2/3) and strong (Cn2=1×10^-13 m^-2/3) turbulence, respectively.

Conclusions:

• The improved deep learning model effectively mitigates misalignment errors in OAM-based FSO communication.
• The view-pooling layer enhances the network's ability to handle spatial variations caused by misalignment.
• The proposed method significantly improves the robustness and reliability of FSO communication systems operating under turbulent conditions.