PaperlayerView exact revision on arXiv

Source-linked AI summary

Big Data Driven Vehicular Networks

Nan Cheng, Feng Lyu, Jiayin Chen, Wenchao Xu, Haibo Zhou, Shan Zhang, Xuemin, Shen

arXiv:1804.04203v1eess.SP

TL;DR

VANETs generate rapidly escalating heterogeneous data that must be transmitted efficiently and reliably, while related data can support VANET analysis and improvement. The article reviews 5G and alternative data-pipe approaches, discusses big-data-based characterization and protocol design, and applies machine learning to detect NLoS conditions. NLoS detection reaches about 92.5%–97.4% with NB and 93.7%–98.3% with SVM across highway, suburban, and urban scenarios, with SVM slightly outperforming NB.

  • Problem

    VANETs face escalating big-data demands, while their data also presents an opportunity to characterize networks and improve performance.

  • Method

    The article reviews 5G and opportunistic data pipes, discusses big-data-based VANET analysis, and applies NB and SVM to historical PDR features for NLoS detection.

  • Results

    NLoS detection accuracy reaches about 92.5%, 96.9%, and 97.4% with NB and 93.7%, 98.3%, and 98.3% with SVM in highway, suburban, and urban scenarios, respectively.

  • Takeaways & Limitations

    Big data can support VANET characterization and intelligent protocol design, while machine learning can detect NLoS conditions from VANET measurements.

Abstract

from arXiv · show

Vehicular communications networks (VANETs) enable information exchange among vehicles, other end devices and public networks, which plays a key role in road safety/infotainment, intelligent transportation system, and self-driving system. As the vehicular connectivity soars, and new on-road mobile applications and technologies emerge, VANETs are generating an ever-increasing amount of data, requiring fast and reliable transmissions through VANETs. On the other hand, a variety of VANETs related data can be analyzed and utilized to improve the performance of VANETs. In this article, we first review the VANETs technologies to efficiently and reliably transmit the big data. Then, the methods employing big data for studying VANETs characteristics and improving VANETs performance are discussed. Furthermore, we present a case study where machine learning schemes are applied to analyze the VANETs measurement data for efficiently detecting negative communication conditions.

I. INTRODUCTION

VANETs connect vehicles and networks to support transportation, safety, infotainment, and self-driving applications, while generating rapidly escalating big data. The article reviews ways to transmit and exploit this data, including machine-learning-based detection of negative communication conditions.

  • VANETs enable efficient and reliable information exchange through vehicle-to-vehicle and vehicle-to-infrastructure communications.
  • The data volume required, generated, collected, and transmitted by VANETs has escalated exponentially with mobile services and self-driving technologies.
  • VANETs big data exhibits the five characteristics known as volume, variety, velocity, value, and veracity.
  • Big data enables applications including smart cities and intelligent transportation systems, using large-scale traffic information for transportation services.
  • Supporting VANETs big data requires high data rates, large capacity, heterogeneous network integration, differentiated QoS, and capabilities for collection, storage, and computation.
  • The article examines big-data transmission, VANET characterization and improvement, and machine-learning detection of negative communication conditions.

II. BIG DATA IN VANETS

VANETs big data arises from heterogeneous vehicle, location, sensing, and service sources. These sources support applications ranging from diagnosis and traffic management to autonomous perception and infotainment.

  • VANETs big data comes from multiple heterogeneous sources with diverse volume, structure, value, and processing-delay requirements.
  • Vehicle sensing data supports vehicle diagnosis, road safety improvement, smart charging, and accident detection.
  • GPS data provides structured vehicle location and speed information for navigation, traffic management, routing optimization, and trajectory analysis.
  • Self-driving technologies use cameras and LiDAR to improve environmental perception, producing very large data volumes continuously.
  • Vehicular mobile services such as streaming, gaming, social networks, and user-generated content require or generate substantial data.

III. SUPPORTING BIG DATA IN VEHICULAR NETWORKS

Supporting VANETs big data requires coordinated aggregation, storage, transmission, and computation beyond traditional decentralized, bandwidth-limited technologies. The article highlights 5G as a high-capacity, low-latency platform with differentiated use cases for vehicular data.

  • Big-data operation in VANETs requires data aggregation, storage, transmission, and computation.
  • Raw data are processed to extract valuable information and transmitted to cloud or edge storage systems for further analysis.
  • Traditional IEEE 802.11p-based VANETs struggle with big-data demands because decentralized protocols and bandwidth limitations restrict resources, flexibility, and QoS support.
  • A. 5G Technologies: 5G is designed to support VANET big data with a 10 Gb/s data rate, less than 1 ms end-to-end latency, and machine-type communications.
  • A. 5G Technologies: 5G use cases—eMBB, URLLC, and mMTC—provide differentiated performance categories for VANET big-data gathering and transmission.
  • A. 5G Technologies: eMBB targets high-capacity vehicular data services, with a peak data rate of 10 Gb/s and mobile data volume of 10 Tb/s/km2.
  • A. 5G Technologies: URLLC addresses mission-critical vehicular services requiring less than 5 ms latency and higher than 99.999% reliability.
  • A. 5G Technologies: mMTC supports massive concurrent connectivity from densely deployed vehicular devices with low energy consumption and low latency.

B. Opportunistic Data Pipes

Because cellular capacity, deployment timing, and transmission cost constrain big-data support, alternative data pipes can offload traffic cost-effectively.

  • 5G capacity may still face congestion as VANET big data grows, while near-term 4G LTE capacity remains relatively limited.
  • WLANs, CRNs, and D2D communications can offload VANET big data from cellular networks in a cost-effective way.

1) WiFi Offloading:

Vehicular WiFi offloading exploits intermittent roadside coverage and vehicle mobility to decide when data should use WiFi or cellular transmission.

  • WiFi offloading uses small, intermittent roadside coverage areas, making access spatially and temporally opportunistic for vehicles.
  • Mobility prediction and prior knowledge of WiFi deployment can estimate future access opportunities and corresponding throughput.
  • Offloading decisions can account for application delay tolerance by waiting for WiFi or transmitting directly through cellular networks.

2) Cognitive Radio Technology:

The paper discusses cognitive radio and D2D communications as spectrum and connectivity technologies relevant to big-data transmission and VANET applications.

  • 2) Cognitive Radio Technology:: Cognitive radio lets unlicensed users opportunistically exploit vacant licensed spectrum bands allocated to licensed systems.
  • 2) Cognitive Radio Technology:: Cognitive radio can use underutilized spectrum resources for big-data transmission, but vehicle mobility may require frequent spectrum sensing.
  • 2) Cognitive Radio Technology:: D2D communications use proximity to connect mobile users directly without traversing base stations or backhaul networks.
  • 2) Cognitive Radio Technology:: The reviewed mobility and measurement datasets support channel and mobility modeling, vehicle-movement prediction, and intelligent protocol design.

A. Vehicle Mobility Trace Data

Vehicle mobility traces provide data for modeling movement and network connectivity, while communication measurements characterize vehicular channels across diverse environments.

  • A. Vehicle Mobility Trace Data: Vehicle mobility datasets can reveal practical mobility models, network connectivity, and spatial and temporal density distributions.
  • A. Vehicle Mobility Trace Data: Trace gaps caused by reporting intervals require preprocessing, including route prediction from maps, traffic signs, and past vehicle data.
  • A. Vehicle Mobility Trace Data: Mobility and network measurements inform position-based routing and MAC protocols adapted to changing topology and high vehicle mobility.
  • A. Vehicle Mobility Trace Data: Mobility traces support connectivity metrics such as link duration, average hops, connected vehicle pairs, and interconnect-time distributions.
  • A. Vehicle Mobility Trace Data: IEEE 802.11p devices on vehicles and roadside units collect realistic communication measurements across urban, suburban, rural, open-field, and freeway environments.
  • A. Vehicle Mobility Trace Data: WiFi offloading measurements reveal entry, production, and exit phases, with weaker connection quality and lower data rates during entry and exit.
  • A. Vehicle Mobility Trace Data: Vehicular channels require practical models because shadowing, Doppler shifts, and non-stationarity complicate wireless behavior.

V. CASE STUDY

The case study applies machine learning to VANET measurement data for online detection of negative communication conditions, particularly NLoS links.

  • V. CASE STUDY: The case study uses big data and machine learning schemes to support efficient protocol design in VANET communications.
  • V. CASE STUDY: Because NLoS conditions degrade V2V link performance, the proposed scheme learns from V2V measurement data to detect NLoS conditions online.
  • V. CASE STUDY: The approach is motivated by avoiding blind retransmissions in harsh NLoS conditions that may waste resources and increase interference.
  • V. CASE STUDY: Detected channel conditions can support protocols that allocate resources to LoS vehicles or use helper vehicles to relay NLoS packets.

B. Collecting V2V Communication Measurement Data Sets

The campaign collects synchronized V2V communication, GPS, and video data across highway, suburban, and urban roads to relate communication outcomes to driving environments.

  • Experimental setup: The campaign uses two experimental vehicles equipped with roof-mounted DSRC modules, with the transmitter sending a 300-byte packet every 100 ms.Both vehicles log transmitted and received packets, while cameras mounted on the front and rear glass record the process.
  • Data collected: Each data set combines communication traces, GPS traces, and recorded videos in a time-synchronized campaign.Communication traces identify received or dropped packets and support packet delivery ratio calculation; GPS provides speed, altitude, and distance information, while video records the communicating environment.
  • Campaign scope: Data collection covers highway, suburban, and urban roads in Shanghai, represented by data sets H, S, and U.The campaign lasts over two months, spans more than 1,500 kilometers, and reaches a total data size of up to 110GB.
  • Environmental context: Recorded videos identify road type, traffic conditions, and surrounding obstacles in the communicating environment.These observations support interpretation of communication traces alongside vehicle position and packet outcomes.

C. Supervised Machine Learning

The study uses supervised learning to detect non-line-of-sight conditions from historical packet delivery behavior, evaluates multiple metrics, and compares Naive Bayes with SVM across road scenarios.

  • Learning approach: Naive Bayes and Support Vector Machines are trained to detect NLoS conditions using labeled vehicular measurement data.Camera recordings label situations where communicating vehicles cannot visually see each other, providing approximations of radio NLoS conditions.
  • Metrics: Evaluation reports accuracy, precision, recall, and false positive rate from true-positive, true-negative, false-positive, and false-negative outcomes.These metrics quantify correct identification, NLoS identification quality, NLoS coverage, and mistaken NLoS alarms for LoS conditions.
  • Evaluation design: Ten-fold cross-validation evaluates models trained and validated across highway, suburban, and urban data sets.The data sets contain 16,425 highway samples, 16,033 suburban samples, and 27,439 urban samples.
  • Training-data robustness: Naive Bayes requires diverse training data to reach very high accuracy, then plateaus near 96.5% as training size increases.In the highway scenario, accuracy rises from 84.3% to 90.9% and then 96.4% at training proportions 0.1, 0.2, and 0.3, respectively.

VI. CONCLUSIONS

The article addresses both efficient big-data support in VANETs and the use of big data to characterize and improve them, illustrating the latter with machine-learning-based NLoS detection.

  • Big-data support: The paper introduces a framework combining 5G cellular networks with alternative opportunistic data pipes for efficient, reliable, and flexible VANET big-data support.This addresses transmission of VANET big data.
  • Big-data utilization: It discusses mechanisms that analyze and learn from big data to characterize VANETs and design intelligent protocols.The conclusion frames big-data use as a means of improving VANET operation.
  • Case study: The case study uses urban VANET measurement data and machine-learning schemes to detect NLoS conditions.The example demonstrates the paper’s approach to employing measurements for VANET improvement.
Loading 1804.04203v1…