Accurate prediction approach for the center position of a future light spot under atmospheric turbulence

Atmospheric turbulence significantly impacts satellite-to-ground laser communication (SGLC). To enhance the accuracy and stability of the pointing, acquisition, and tracking (PAT) system, we propose a hybrid method for predicting the center position of light spots, named feature matching recurrent prediction (FMRP). The foundational architecture of FMRP is based on a recurrent neural network (RNN) structure, with feature matching utilized to optimize the acquisition of input features and labels. Unlike most previous deep learning-based methods, a key advantage of FMRP is that it can achieve accurate position prediction only by processing real-time received spot images, eliminating the need for offline training, and it does not require the collection of a large dataset for model training. Experimental validation demonstrates that this method is effective under both weak and strong atmospheric turbulence, exhibiting high prediction accuracy and stability.