Machine Learning-Based Line-Of-Sight Prediction in Urban Manhattan-Like Environments - Conference paper summary

Aug 5, 2026

This paper addresses predicting whether a transmitter-receiver link is in Line-of-Sight (LOS) when only a limited set of high-level descriptors about the propagation environment and radio link is available. LOS prediction is formulated as a binary classification machine learning task, using a baseline Gradient Boosting Decision Trees (GBDT) classifier trained and tested on a synthetic raytracing dataset with Manhattan-like topologies. Generalization to unseen locations and environments is evaluated, showing good classification performance and accurate LOS probability modeling. Feature importance indicates learned decision rules consistent with common sense.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

Line-of-sight and its impact on wireless channels

Existing LOS probability models and limitations

Related Work

Machine learning in radio wave propagation

Method and Problem Setup

Dataset generation from ray tracing

LOS classification with ML

Numerical Results and Discussion

Classification performance and probability modeling

Conclusion

Key takeaways and future work

Machine Learning-Based Line-Of-Sight Prediction in Urban Manhattan-Like Environments Nicola Di Cicco∗ , Simone Del Prete†, Silvi Kodra†, Marina Barbiroli† Franco Fuschini†, Enrico M. Vitucci†, Vittorio Degli Esposti†, Massimo Tornatore∗ , ∗ Department of Electronics, Information and Bioengineering (DEIB), Politecnico di Milano, Italy †Department of Electrical, Electronic and Information Engineering (DEI), CNIT, University of Bologna, Italy Abstract—This paper considers the problem of predicting whether or not a transmitter and a receiver are in Line-of-Sight (LOS) condition. While this problem can be easily solved using a digital urban database and applying ray tracing, we consider the scenario in which only few high-level features descriptive of the propagation environment and of the radio link are available. LOS prediction is modelled as a binary classification Machine Learning problem, and a baseline classifier based on Gradient Boosting Decision Trees (GBDT) is proposed. A synthetic raytracing dataset of Manhattan-like topologies is generated for training and testing a GBDT classifier, and its generalization capabilities to both locations and environments unseen at training time are assessed. Results show that the GBDT model achieves good classification performance and provides accurate LOS probability modelling. By estimating feature importance, it can be concluded that the model learned simple decision rules that align with common sense. Index Terms—propagation modelling, ray tracing, line-of-sight probability, machine learning, datasets. I. INTRODUCTION The presence of Line-of-Sight (LOS) condition between two radio link ends represents one of the basic properties of a propagation environment. LOS determines the fundamental characteristics of the radio channel, [e.g. link](e.g. link) budget, fading statistics, time and angle spreading, and heavily impacts on the choice of the optimal transmission and coding technique. With the use of the mm-wave spectrum for 5G and beyond systems [1], required to cope with the ever-increasing demand for higher bitrates, the LOS condition becomes even more important due to the higher blocking effect of obstacles. As such, LOS probability has gained importance as a key property in wireless channel prediction and simulation. Several statistical propagation models, such as the ‘WINNER’ model [2] and the ITU-R Recommendation P.1411 [3] are based on the definition of different path loss formulation as a function of the LOS or Non-LOS (NLOS) condition. LOS probability is likewise important in spectrum-sharing studies, such as the ones conducted within the CEPT and ITU-R. In these studies, adjacent bands are allocated to services operating in the same geographical area, leading to design systems where minimum interference must be provided to the incumbent or protected users while maximizing the number of users simultaneously accessing the same spectrum [4] . Overall, the development of suitable models to determine LOS condition in urban environments based on general characteristics such as building density, street width, and link distance, is very valuable for all those cases where accurate information about the environment layout is unavailable, or would be too difficult or time-consuming to determine. Often, LOS probability is estimated through an empirical model fitted from some measurement data, and the output is typically a decreasing exponential function with the distance. The LOS probability is provided for typical environments namely Indoor Hotspot, Urban Macro, Urban Micro, and Rural Macro, as described in [5] . However, these models do not consider the actual geometry of the environment, like the building height or position in an urban scenario, or the antenna height. 3GPP reported a study on channels from 0.5GHz to 100GHz [6] . The report describes different environments (e.g., indoor office, street canyon, etc.), but all the functions are simply a negative expo

FAQs

How is the LOS prediction problem formulated in this work?

LOS prediction is cast as a binary classification machine learning problem that outputs whether a link is in Line-of-Sight condition.

What data and model are used to train and test the classifier?

A synthetic raytracing dataset of Manhattan-like topologies is generated for training and testing, and a baseline Gradient Boosting Decision Trees (GBDT) classifier is used.

How is the model’s generalization evaluated?

The paper assesses generalization capabilities to both locations and environments not seen during training, and reports good classification performance with accurate LOS probability modeling.

Aug 5, 2026

Upload by Lucas Martin