Computationally efficient self-attention assisted signal detection method for SIMO FSO communications with...

Area of Science:

• Optical communication systems
• Signal processing
• Machine learning applications

Background:

• Atmospheric turbulence severely degrades free-space optical (FSO) communication performance.
• Spatial diversity techniques mitigate turbulence, but combining and detection methods are critical.
• Deep learning (DL) offers solutions but often faces high computational complexity.

Purpose of the Study:

• To propose a novel, low-complexity self-attention neural network (SANN) for signal combining and detection in SIMO FSO systems.
• To reduce the computational overhead associated with DL-based methods in FSO communication.
• To enhance the practical implementation of advanced signal processing techniques in turbulent FSO channels.

Main Methods:

• Development of a self-attention neural network (SANN) utilizing lightweight self-attention mechanisms.
• Implementation of adaptive focusing on crucial signal components for efficient processing.
• Experimental validation using measured intensity fluctuations for realistic simulation of atmospheric turbulence.

Main Results:

• The proposed SANN achieves significant computational overhead reduction compared to CNN-based DL and MMSE methods.
• Maintained comparable or superior bit error rate (BER) performance.
• Demonstrated over 70% reduction in computational complexity compared to existing CNN-based detectors.

Conclusions:

• The SANN provides a computationally efficient framework for signal detection in SIMO FSO systems.
• This approach offers a practical solution for FSO systems where processing efficiency is crucial.
• The study highlights the potential of lightweight attention mechanisms in overcoming DL complexity barriers in FSO communication.