Improving QoS Prediction in Urban V2X Networks by Leveraging Data from Leading Vehicles and Historical Trends

Sanket Partani 1 , Michael Zentarra 2 , Anthony Kiggundu 2 and Hans D. Schotten 1 2 · 2025

Read Improving QoS Prediction in Urban V2X Networks by Leveraging Data from Leading Vehicles and Historical Trends by Sanket Partani 1 , Michael Zentarra 2 , Anthony Kiggundu 2 and Hans D. Schotten 1 2 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-16848.txt.

Introduction

Cars that talk to the network need warning of weak connections ahead, and a 2025 paper by Sanket Partani and co-authors improves those forecasts for city driving. Their method predicts download speed for a car using data from the car ahead of it, plus patterns from the past.

The team worked with a public Berlin data set for vehicle-to-network links over 5G. They picked input traits from both the target car and the lead car after a close look at the data, then trained several machine learning models. Adding the lead car's data improved the forecasts. The gain held across model types, which suggests the lead car acts like a scout that has just driven the same stretch of road.

Catalog record

CREATORSanket Partani 1 , Michael Zentarra 2 , Anthony Kiggundu 2 and Hans D. Schotten 1 2
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-16848

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Improving QoS Prediction in Urban V2X Networks by Leveraging Data from Leading Vehicles and Historical Trends Sanket Partani 1 , Michael Zentarra 2 , Anthony Kiggundu 2 and Hans D. Schotten 1 2 Abstract With the evolution of Vehicle-to-Everything (V2X) technology and increased deployment of 5G networks and edge computing, Predictive Quality of Service (PQoS) is seen as an enabler for resilient and adaptive V2X communication systems. PQoS incorporates data-driven techniques, such as Machine Learning (ML) , to forecast/predict Key Performing Indicators (KPIs) such as throughput, latency, etc. In this paper, we aim to predict downlink throughput in an urban environment using the Berlin V2X cellular dataset. We select features from the ego and lead vehicles to train different ML models to help improve the predicted throughput for the ego vehicle. We identify these features based on an in-depth exploratory data analysis. Results show an improvement in model performance when adding features from the lead vehicle. Moreover, we show that the improvement in model performance is model-agnostic. I Introduction The evolution of V2X technology and increasing integration with autonomous vehicles, as part of Connected, Cooperative and Automated Mobility (CCAM) , holds the potential to not only optimize traffic flows and improve road users’ safety, but also reduce emissions and help meet the climate action targets. The autonomous car of today is equipped with a suite of sensors, including but not limited to lidar, camera, GPS, and V2X , which is seen as an extended sensor. V2X encompasses a vehicle communicating not only with other vehicles ( V2V ), but also pedestrians ( V2P ), networks ( V2N ), roadside infrastructure ( V2I ), amongst others. V2X can provide timely and crucial data about traffic information, road conditions and pedestrian movement to the vehicle, thus, enabling informed decisions by either the driver or the self-driving systems. Several V2X applications like tele-operated driving, high-density platooning, as well as infotainment applications such as video streaming, have diverse and (often) stringent network requirements with respect to throughput, latency, etc. that demands a certain Quality of Service (QoS) level and availability from the communications network. By predicting QoS and providing in-advance notifications, PQoS emphasizes safety of road users, reliability in information exchange, end-user experience and satisfaction. Furthermore, PQoS can provide context awareness to applications depending on the driving conditions and users’ needs. This can regulate application-specific survival time and ensures that the service is either continued or gracefully degraded [ 1 ] . For example, in high latency scenarios, autonomous vehicles can be proactive by either decelerating, changing lanes or handover control to the driver to avoid an accident. Predicting QoS is indeed a complex task, further exacerbated by the high relative vehicle speeds, vehicle density (based on time of the day) and channel conditions (based on region - rural, urban and highway) prevalent in V2X scenarios. As part of Network Data Analytics Function (NWDAF) , 3GPP Release 16 [ 2 ] introduced an in-advance, subscribe/request-based notification system to notify V2X applications of estimated or expected change in QoS , with the possibility to switch to an Alternative QoS Profile (AQP) when QoS degrades. However, this does not ensure service continuity, as the application is notified once the change occurs. QoS can be predicted either on the User Equipment (UE) -side [ 3 , 4 , 5 , 6 , 7 ] , or the network-side [ 8 ] , or a combined approach [ 9 ] . Majority of the research for PQoS includes prediction by training models on historical data of the UE , i.e., ego vehicle itself (henceforth, vehicle and device will be used interchangeably). However, authors in [ 10 ] proposed a device-to-device (D2D) coverage prediction framework to predict QoS by using the received signal strength measurements from the lead device. Initial findings in [ 11 ] indicate that the data from the lead vehicle holds the potential to improve the PQoS of the ego vehicle. Authors in [ 12 ] highlight a new set of features from the lead vehicle and network operator to improve the performance of a Random Forest (RF) model to predict both uplink and downlink throughput of the ego vehicle and over different prediction horizons. A similar study [ 13 ] shows improvement throughput prediction of the XGBoost regressor by selecting a different combination of features from the lead vehicle and the network operator and over different time horizons. In this paper, we intend to build upon the insights of [ 12 ] and [ 13 ] , both of which have conducted their studies using a dataset collected in a highway environment. However, unlike [ 12 ] [ 13 ] , we aim to focus on improving QoS by means of instantaneous downlink throughput prediction in urban environments, i.e., focusing on the ego vehicle’s current state or in the very near future, by using both current (temporally-aligned) and historical (spatially-aligned) data from the lead device. Furthermore, we train 3 different types of ML models. In this regard, we pose the following research questions, the answers to which form the basis of our contributions: • Can the data from the lead vehicle improve the PQoS of the ego vehicle in an urban setting? • Can the historical data, i.e., temporal difference between the lead and ego vehicle at the same geo-coordinate be utilized to improve the PQoS of the ego vehicle? • How does feature engineering the datasets, i.e., introducing relative features impact the model performance? • If the selection of features from the lead vehicles shows an improvement in model performance, is it model-agnostic? We are using the Berlin V2X dataset [ 14 ] , primarily, as the dataset includes vehicles driving in close proximity in an urban setting. Moreover, it is publicly available ensuring reproducibility of results. At the time of writing this paper, this dataset has been previously used to predict downlink throughput by training a Light Gradient Boosting Machine (LightGBM) model on one operator’s data and evaluating on another operator’s data [ 15 ] ; identifying nodal points in V2X networks by predicting data rate [ 16 ] ; and handling concept drifts when predicting QoS in resource-constrained V2X environments [ 17 ] . The rest of the paper is organized as follows. Section II provides an in-depth correlation analysis of the different features of both the ego and lead device. The correlation analysis helps select features for training the ML models described in Section III . Section IV delivers answers to the aforementioned research questions and lastly, Section V concludes the paper and provides insights into future research directions. II Exploratory Data Analysis TABLE I : Berlin V2X Cellular Dataset - Overview Direction Downlink Uplink Operator IDs 1 2 1 2 Device IDs pc1, pc4 pc2, pc3 pc1, pc4 pc2, pc3 Target Datarate [kBit/s] 400 350,000 400 350,000 400 75,000 400 75,000 Measurement ID 3, 4 0, 1, 2, 14, 15, 16 1, 2, 14 0, 3, 4, 15, 16 8, 9, 10 5, 6, 7, 11, 12, 13 8, 9, 10 5, 6, 7, 11, 12, 13 The Berlin V2X dataset captures cellular data transmission in both directions - uplink and downlink. A brief overview is provided in Table I

Read the complete text