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

Abstract

The operational costs of the advanced Air Traffic Management (ATM) solutions are often prohibitive in low- and medium-sized airports. Therefore, new and complementary solutions are currently under research in order to take advantage of existing infrastructure and offer low-cost alternatives. The 5G signals are particularly attractive in an ATM context due to their promising potential in wireless positioning and sensing via Time-of-Arrival (ToA) and Angle-of-Arrival (AoA) algorithms. However, ToA and AoA methods are known to be highly sensitive to the presence of multipath and Non-Line-of-Sight (NLOS) scenarios. Yet, LOS detection in the context of 5G signals has been poorly addressed in the literature so far, to the best of the Authors’ knowledge. This paper focuses on LOS/NLOS detection methods for 5G signals by using both statistical/model-driven and data-driven/machine learning (ML) approaches and three challenging channel model classes widely used in 5G: namely Tapped Delay Line (TDL), Clustered Delay Line (CDL) and Winner II channel models. We show that, with simulated data, the ML-based detection can reach between 80% and 98% detection accuracy for TDL, CDL and Winner II channel models and that TDL is the most challenging in terms of LOS detection capabilities, as its richness of features is the lowest compared to CDL and Winner II channels. We also validate the findings through in-lab measurements with 5G signals and Yagi and 3D-vector antenna and show that measurement-based detection probabilities can reach 99–100% with a sufficient amount of training data and XGBoost or Random Forest classifiers.

Machine Learning-Based Beamforming Algorithm for Massive MIMO Systems in 5G Networks

“…Shabih et al [24] propose machine learning based hybrid precoding for High Altitude Platform (HAP) Massive MIMO systems with reduced RF chains This study addresses the practical limitations of hardware in massive MIMO systems that have to be considered on a real-life level when implementing beamforming solutions. Lastly, the paper of Palihawadana et al [25] discusses LOS detection for 5G signals using machine learning in an airport-related application This work sheds light on the localization aspect of communication systems, a component closely tied to beamforming optimization in cases with a changing LOS condition. Overall, the related work highlights that AI-assisted beamforming and channel modeling applications utilize machine learning techniques for an array of communication environments.…”

5G Positioning: Advantageous Aspects, Taxonomy, State-of-the-art Review and Future Trends

“…Another approach is to apply the statistical as well as machine learning methods to identify mainly the LOS 5G signals, which would enhance the accuracy and robustness of such positioning in NLOS-rich areas [106]. Moreover, reinforcement learning (RL) based CNN model is able to mitigate the effect of multipath in the 5G downlink signal.…”

Analysis of 5G Technology in Airports: Requirements, Challenges, and Use Cases Evaluation

“…Studies have also investigated how 5G can enhance user equipment positioning accuracy without dedicated hardware, as is detailed in [23] or through the application of Machine Learning (ML) techniques [25], [26]. Different positioning solutions that 5G can offer include Downlink Time Difference of Arrival (DL-TDOA), Uplink Time Difference of Arrival (UL-TDOA), Multi-cell Round Trip Time (Multi-RTT), Downlink Angle of Departure (DL-AoD), Uplink Angle of Arrival (UL-AoA), and Enhanced Cell ID (E-DIC) along with a plethora of ML/Artificial Intelligence (AI) algorithms.…”