🤖 AI

MINT-V2X: A Mobility-Integrated Network Trajectory Dataset for Predictive Resource Management

This paper introduces MINT-V2X, a comprehensive and validated dataset that synchronizes vehicle mobility trajectories with realistic wireless network parameters to bridge the infrastructure gap in V2X research and enable superior predictive resource management.

Original authors: Abdullah Anjum, Abdolazim Rezaei, Mehdi Sookhak

Published 2026-07-28

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Abdullah Anjum, Abdolazim Rezaei, Mehdi Sookhak

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of self-driving cars not as a collection of metal boxes on wheels, but as a massive, high-speed game of tag played on a digital chessboard. In this game, every car is a player that needs to whisper secrets to its neighbors and shout warnings to the traffic lights, all in the blink of an eye. This is the world of Vehicle-to-Everything (V2X) communication. For these cars to stay safe and avoid crashing, they need two things to happen perfectly at the same time: they need to know exactly where they are going (their "trajectory"), and they need to know if the airwaves they are shouting through are clear or clogged with noise (the "network").

The problem is that for a long time, scientists have been studying these two things in separate rooms. One group has maps of where cars go, but no idea about the radio signals. Another group has charts of radio signals, but no idea which specific car is sending them. It's like trying to learn how to play a duet by listening to the violinist in one room and the drummer in another, never hearing how they sound together. Without a single, unified record that shows both the car's movement and the signal quality at the exact same moment, it's very hard to teach computers to predict traffic jams or signal failures before they happen.

This is where the new paper, titled "MINT-V2X," steps in to build a bridge between these two rooms. The researchers created a massive, digital "simulated city" to generate a brand-new dataset that finally combines these missing pieces. Instead of just guessing or looking at separate logs, they built a virtual world using three different computer programs working together like a dream team. First, they used a traffic simulator to create realistic cars driving through the streets of Corpus Christi, Texas. Then, they used a network simulator to act out exactly how radio waves bounce off buildings and travel between those cars and roadside towers. Finally, they glued these two simulations together so that every time a car moved a tiny bit, the radio signal updated instantly.

The result is a treasure trove of data called MINT-V2X. It contains nearly 10 million synchronized snapshots (9.87 million, to be exact) taken over three hours of simulated driving. In this digital world, 1,386 virtual cars zoom around, talking to 15 roadside towers. The data is so detailed that it records the car's speed, position, and acceleration, while simultaneously measuring the strength of the radio signal, how many messages get through, and how "busy" the airwaves are. The researchers didn't just throw this data together; they ran it through a strict 14-point checklist to make sure it followed the real-world rules of physics and communication standards, proving that their virtual world behaves very much like the real one.

The most exciting part of their work is a "test drive" to see if this new data is actually useful. They asked a simple question: "Can we predict how busy a roadside tower will be in the next few seconds?" They tried two approaches. The first approach was like looking at a weather report based only on what happened yesterday (using only past network data). The second approach was like looking at the weather report while also watching the cars on the road (using the new combined data). The result? The approach that used the car movement data was much better at predicting the future. It showed that if you know where the cars are heading, you can guess where the network traffic will clog up before it even happens.

The paper is careful to note that this is a simulation, not a recording of real cars on real streets. The virtual cars followed standard driving rules and didn't do anything crazy like aggressive racing or driving in tight "platoons." Also, the radio signals in this simulation assumed a mostly clear view between the car and the tower, without the heavy interference of tall buildings blocking the signal. However, the authors argue that this dataset is a crucial first step. It provides a solid, reproducible foundation that other scientists can use to build smarter, safer systems for our future roads, proving that when you finally let the traffic map and the radio map talk to each other, you get a much clearer picture of the future.

Technical Summary: MINT-V2X

Problem Statement

Connected and Autonomous Vehicles (CAVs) rely on low-latency Vehicle-to-Everything (V2X) communications for safety-critical applications such as cooperative driving and collision avoidance. Effective resource management in these dynamic environments requires predictive models capable of forecasting vehicle trajectories and corresponding network loads over 5–10 second horizons. However, a critical infrastructure gap exists: publicly available datasets typically offer either high-fidelity mobility data (e.g., nuScenes, Argoverse) or network metrics (e.g., CRAWDAD), but rarely both in a single, time-synchronized view.

Existing V2X datasets suffer from three primary limitations:

Integration Gap: No existing dataset provides per-vehicle kinematics, physical-layer metrics (SINR, PDR, CQI), and RSU associations in a single time-synchronized record at sub-second granularity.
Scale Gap: Current datasets (e.g., Berlin V2X) often contain only hundreds of vehicles, insufficient for realistic large-scale deployment modeling.
Validation Gap: Published datasets frequently lack rigorous validation against established physical-layer standards (3GPP, ETSI) and information-theoretic models.

Methodology

To address these gaps, the authors propose MINT-V2X, a dataset generated through a coupled simulation architecture. The methodology employs a three-layer framework connected via the Traffic Control Interface (TraCI):

Traffic Simulation (Layer 1): SUMO (Simulation of Urban Mobility) simulates urban traffic dynamics in Corpus Christi, Texas, using OpenStreetMap data. The simulation covers 61.19 km², runs for 3 hours (10,800 seconds), and generates 1,386 vehicles using a Poisson process and the Krauss car-following model.
Middleware (Layer 2): Veins provides bidirectional coupling between SUMO and the network simulator, instantiating an OMNeT++ module for each vehicle. Vehicle positions are synchronized at 10 Hz (0.1-second intervals).
Network Simulation (Layer 3): OMNeT++ with the Simu5G framework simulates the C-V2X physical layer (3GPP Release 14). The setup includes 15 Roadside Units (RSUs) arranged in a 5×3 grid. Vehicles transmit ETSI Cooperative Awareness Messages (CAMs) at 10 Hz. The simulation utilizes a log-distance path-loss model with Nakagami fading at the 5.9 GHz ITS band.

Data Collection and Preprocessing:
The pipeline records a 29-feature vector every 100 ms per vehicle, encompassing trajectory kinematics, network state (RSU association, distance, neighbor count), physical-layer metrics (SINR, CQI, PDR, MCS), and end-to-end performance (throughput, latency). The raw data undergoes standardized preprocessing, including temporal alignment, linear interpolation for short missing segments (<500 ms), outlier detection (speed changes >5 m/s), and Z-score normalization. The dataset is split temporally into training (60%), validation (15%), and testing (25%) sets to prevent information leakage.

Key Contributions

The paper presents four primary contributions:

Integrated Dataset Generation: The coupling of SUMO, Veins, and OMNeT++/Simu5G to generate time-aligned trajectory and network-state samples at 10 Hz.
Validation Framework: The implementation of a 14-point validation framework referencing 3GPP C-V2X, ETSI standards, and Shannon capacity theory.
MINT-V2X Release: A comprehensive dataset containing 9.87 million synchronized records from 1,386 vehicles over three hours of urban simulation, accompanied by complete reproducibility documentation and configuration files.
Fidelity Evidence: Quantitative proof of dataset consistency through network-metric correlations and trajectory-validation metrics that meet or exceed published benchmarks.

Results and Validation

The dataset was validated against 14 standardized tests, with all parameters passing against expected ranges defined by 3GPP, ETSI, and theoretical models. Key results include:

Scale: 9.87 million records from 1,386 vehicles across 15 RSUs.
Metric Correlations: Strong algorithmic consistency was demonstrated, specifically a CQI-SINR correlation of 0.993 and an SINR-PDR correlation of 0.946.
Physical Layer Fidelity: The path-loss exponent was empirically calculated at $n = 1.991$ , closely matching the theoretical free-space value of 2.0. The Channel Busy Ratio (CBR) distribution (73.2% in relaxed state) aligns with ETSI Decentralized Congestion Control expectations.
Connectivity: The simulation achieved a 99.78% connectivity ratio within RSU coverage zones.

Case Study: RSU Load Prediction
To demonstrate utility, the authors conducted a case study on predicting RSU load. They compared three models:

Load-only: Based on historical load.
Load + Communication: Added network metrics (SINR, CQI, PDR).
Load + Communication + Trajectory: Incorporated predicted vehicle positions and kinematics.

The model utilizing trajectory data achieved the highest coefficient of determination ( $R^2$ ), explaining over 90% of RSU load fluctuations. This result indicates that integrating trajectory data yields superior predictive performance compared to baselines relying solely on network history.

Significance

The paper claims that MINT-V2X is unique in four respects:

Granularity: It provides synchronized per-vehicle kinematics and physical-layer metrics at 100 ms intervals.
Ground Truth: It offers ground-truth RSU associations, enabling direct trajectory-load correlation.
Scale: It operates at an order of magnitude larger scale than comparable existing datasets.
Reproducibility: It provides full access to configuration files, source code, and the validation methodology.

By bridging the gap between trajectory and network data, the study enables the development of predictive models for proactive RSU resource planning, handover optimization, and mobility-aware network management. The authors position the 14-point validation framework as a reusable template for evaluating future simulation-based V2X datasets.