Performance analysis of a vision transformer based on a hierarchical spatial enhancement mechanism in the coherent...

Atmospheric turbulence significantly degrades the transmission performance of Coherent Free-Space Optical Communication (CFSOC) systems. Wavefront Sensor-less Adaptive Optics (WFS-less AO) systems offer an effective solution for compensating for turbulence-induced wavefront aberrations in real-time. The correction capability of WFS-less AO systems is determined by the control algorithm employed. In this work, we propose a Vision Transformer (VIT) based on a Hierarchical Spatial Enhancement Mechanism (HS-VIT) to serve as the control algorithm. We integrate a hierarchical spatial enhancement mechanism into the VIT architecture and employ a self-modulated feature aggregation method to enhance further HS-VIT's capability to correct wavefront aberrations under complex turbulence conditions. Theoretical analysis and numerical simulations demonstrate that the HS-VIT model yields performance improvements ranging from 20.9% to 65.4% over established deep learning models under strong turbulence. Experimental results indicate that the HS-VIT model effectively corrects wavefront distortion with a computation time of only 0.32 ms, offering an order-of-magnitude advantage in real-time performance over iterative control algorithms. Implementing the HS-VIT model in WFS-less AO systems significantly improves the communication performance of CFSOC systems.