Towards an Artificial-Intelligence-Based Optical Scintillometer: Scaling Issue

TL;DR

This paper proposes a method for atmospheric turbulence sensing using deep neural networks that can be scaled across different propagation distances without retraining, utilizing either theoretical or simulation-based scaling factors.

Contribution

It introduces a novel scaling approach for DNN-based scintillometer measurements, eliminating the need for retraining when propagation distances change.

Findings

Scaling factor can be derived analytically from turbulence theory

Wave-optics simulations can also determine the scaling factor

The method enables distance-independent turbulence measurement

Abstract

Atmospheric turbulence strength (Cn2 parameter) sensing based on processing of intensity scintillation patterns with deep neural network (DNN) is considered. It is shown that DNN re-training with propagation distance change can be avoided by scaling of Cn2 values obtained using a DNN trained for a nominal distance L0 . The required Cn2 scaling factor can be obtained using either an analytical expression derived from the Kolmogorov turbulence theory (theory-based scaling), or through wave-optics numerical modeling and simulations (M&S-based scaling).