AI-empowered Real-Time Line-of-Sight Identification via Network Digital Twins

TL;DR

This paper presents an AI-based method for real-time Line-of-Sight identification using Network Digital Twins and Ray-Tracing, achieving higher accuracy and significantly reduced computational cost compared to existing models.

Contribution

It introduces training strategies for a general-purpose Deep Learning model trained on NDT-generated data, improving accuracy and efficiency for LoS classification.

Findings

Outperforms state-of-the-art models by 5-10% in accuracy across SNR conditions.

Reduces input size while maintaining performance.

Decreases inference FLOPs by 98.55%, enabling real-time application.

Abstract

The identification of Line-of-Sight (LoS) conditions is critical for ensuring reliable high-frequency communication links, which are particularly vulnerable to blockages and rapid channel variations. Network Digital Twins (NDTs) and Ray-Tracing (RT) techniques can significantly automate the large-scale collection and labeling of channel data, tailored to specific wireless environments. This paper examines the quality of Artificial Intelligence (AI) models trained on data generated by Network Digital Twins. We propose and evaluate training strategies for a general-purpose Deep Learning model, demonstrating superior performance compared to the current state-of-the-art. In terms of classification accuracy, our approach outperforms the state-of-the-art Deep Learning model by 5% in very low SNR conditions and by approximately 10% in medium-to-high SNR scenarios. Additionally, the proposed strategies effectively reduce the input size to the Deep Learning model while preserving its performance. The computational cost, measured in floating-point operations per second (FLOPs) during inference, is reduced by 98.55% relative to state-of-the-art solutions, making it ideal for real-time applications.