Performance Evaluation of V2X Communication Using Large-Scale Traffic Data

Vehicular communication (V2X) technologies are widely regarded as a cornerstone for cooperative and automated driving, yet their large-scale real-world deployment remains limited. As a result, understanding V2X performance under realistic, full-scale traffic conditions continues to be relevant. Most existing performance evaluations rely on synthetic traffic scenarios generated by simulators, which, while useful, may not fully capture the features of real-world traffic. In this paper, we present a large-scale, data-driven evaluation of V2X communication performance using real-world traffic datasets. Vehicle trajectories derived from the Highway Drone (HighD) and Intersection Drone (InD) datasets are converted into simulation-ready formats and coupled with a standardized V2X networking stack to enable message-level performance analysis for entire traffic populations comprising over hundred thousands vehicles across multiple locations. We evaluate key V2X performance indicators, including inter-generation gap, inter-packet gap, packet delivery ratio, and channel busy ratio, across both highway and urban intersection environments. Our results show that cooperative awareness services remain feasible at scale under realistic traffic conditions. In addition, the findings highlight how traffic density, mobility patterns, and communication range influence V2X performance and how synthetic traffic assumptions may overestimate channel congestion.

💡 Research Summary

This paper presents a large‑scale, data‑driven evaluation of vehicular‑to‑everything (V2X) communication performance using real‑world traffic recordings. The authors select two publicly available drone‑based datasets—HighD (highway) and InD (urban intersections)—because they satisfy three essential criteria: (1) they contain detailed vehicle dynamics (position, speed, heading, acceleration) captured from real traffic, (2) they cover medium‑ to large‑scale scenarios with tens of thousands of vehicles across multiple locations and road types, and (3) they record all vehicles traversing a fixed roadway segment, enabling multi‑vehicle message exchange analysis beyond ego‑centric perspectives.

To make the datasets usable for V2X simulation, the authors convert the original OpenDRIVE lanelet definitions into SUMO network files (.net.xml) using the SUMO netconvert tool. Vehicle trajectories stored in CSV files are transformed into SUMO route files (.rou.xml). A coordinate transformation aligns the original dataset origin with the SUMO origin, and each vehicle position is matched to the nearest network node or junction within a 4 m threshold. If a frame fails the threshold, it is skipped and later interpolated, preserving trajectory continuity. The conversion pipeline is validated by (i) confirming that no vehicles are lost and (ii) measuring the average running‑time difference (Δt) between original and SUMO‑generated trajectories, which stays below 0.02 s for all locations, indicating high fidelity.

The communication stack is built on the ETSI ITS‑G5 standard using the Artery framework (OMNeT++ + Vanetza) coupled with SUMO via TraCI. Simulation parameters reflect realistic DSRC settings: 5.9 GHz carrier, 10 MHz bandwidth, 200 mW transmit power, receiver sensitivity –82 dBm, and CAM sizes between 85 and 285 bytes. CAM generation intervals are configurable between 100 ms and 1000 ms, and Decentralized Congestion Control (DCC) is enabled to adapt transmission rates according to channel load.

Four performance metrics are evaluated: Inter‑Generation Gap (IGG), Inter‑Packet Gap (IPG), Packet Delivery Ratio (PDR), and Channel Busy Ratio (CBR). In the highway scenario, average vehicle density reaches 120 vehicles/km². When CAM intervals are ≤200 ms, PDR stays above 95 % and CBR remains around 55 %. At densities above 200 vehicles/km², CBR exceeds 70 %, DCC reduces the transmission rate, IPG spikes, and PDR drops below 85 %, demonstrating that high density leads to severe channel contention.

In the urban intersection scenario, frequent stop‑and‑go maneuvers cause rapid distance fluctuations. Average CBR is about 55 % but peaks at 80 % during dense bursts. DCC again throttles CAM intervals to ≥500 ms, yet PDR stays near 90 %, indicating that adaptive congestion control can mitigate the harsher dynamics of intersections.

Key insights include: (1) Real traffic density and mobility patterns affect channel load differently from synthetic traffic models, especially at intersections where abrupt speed changes cause transient congestion. (2) DCC (or similar adaptive mechanisms) is essential for maintaining service reliability in high‑density environments; without it, channel collisions would dramatically increase. (3) Synthetic traffic simulations tend to overestimate channel occupancy, suggesting that V2X system design must incorporate a safety margin based on empirical data. (4) The data‑driven methodology bridges the gap between autonomous‑driving research (which often uses naturalistic driving datasets) and V2X communication studies, providing a more realistic benchmark for standards development and policy making.

The authors conclude that coupling real‑world traffic datasets with a standardized V2X stack enables scalable, population‑level performance assessment that confirms the feasibility of cooperative awareness services at scale. Future work is outlined to include roadside unit placement optimization, multi‑band operation, and extending the analysis to vulnerable road users (pedestrians, cyclists) to fully capture the heterogeneous V2X ecosystem.