Machine-Learning-Based LOS Detection for 5G Signals with Applications in Airport Environments.

Area of Science:

• Wireless Communications
• Signal Processing
• Air Traffic Management

Background:

• Advanced Air Traffic Management (ATM) systems face high operational costs, especially for smaller airports.
• 5G signals offer potential for low-cost wireless positioning and sensing in ATM using Time-of-Arrival (ToA) and Angle-of-Arrival (AoA).
• Existing ToA/AoA methods are sensitive to multipath and Non-Line-of-Sight (NLOS) conditions, with limited research on 5G LOS detection.

Purpose of the Study:

• To investigate Line-of-Sight (LOS) / Non-Line-of-Sight (NLOS) detection methods for 5G signals within an Air Traffic Management (ATM) context.
• To evaluate both statistical/model-driven and data-driven/machine learning (ML) approaches for LOS/NLOS detection.
• To assess performance across three 5G channel models: Tapped Delay Line (TDL), Clustered Delay Line (CDL), and Winner II.

Main Methods:

• Utilized statistical/model-driven and machine learning (ML) approaches for LOS/NLOS detection.
• Employed Tapped Delay Line (TDL), Clustered Delay Line (CDL), and Winner II channel models for simulations.
• Validated findings through in-lab measurements using 5G signals and Yagi/3D-vector antennas.

Main Results:

• Machine learning-based detection achieved 80%-98% accuracy across TDL, CDL, and Winner II channel models using simulated data.
• The Tapped Delay Line (TDL) channel model presented the most significant challenge for LOS detection due to its limited features.
• In-lab measurements demonstrated 99%-100% detection probabilities with sufficient training data using XGBoost or Random Forest classifiers.

Conclusions:

• Machine learning offers a viable and accurate solution for 5G LOS/NLOS detection in ATM.
• The developed methods can enhance the feasibility of low-cost ATM solutions by leveraging existing 5G infrastructure.
• Further research and validation are supported by the high accuracy achieved in both simulated and real-world measurement scenarios.