Big Data Driven Vehicular Networks

Abstract

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 selfdriving system. As the vehicular connectivity soars, and new onroad 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.

Lessons Learned From Accident of Autonomous Vehicle Testing: An Edge Learning-Aided Offloading Framework

“…Liu et al [6] investigated the feature of a Rayleigh fading channel and proposed to train a long short-term memory (LSTM) model to predict the future channel parameters. Additionally, Cheng et al [7] studied a case where two classical supervise machine learning methods were used to detect the Non-Line-of-Sight (NLoS) conditions by learning the V2V measurement data. Nevertheless, none of the above solutions consider the impact of collected data quality to the inference accuracy of the trained DL model, which, however, is the critical reason for the Uber accident.…”

“…Therefore, to identify the pedestrian with maximum inference accuracy within a certain delay, the vehicle needs to perform pre-braking with the probability η m and offload their DLTs with the optimal offloading probability ̺ * m derived in (11). Calculate optimal offloading probability (̺ * m ); 5 Calculate the optimal inference error rate threshold (ǫ th * m ); 6 Evaluate the inference error rate, i.e., ǫ L m = g(Q, D V m ); 7 if ǫ L m ≥ ǫ th * m then 8 Offload the DLT to the MES; Observation 4. There exists a trade-off between the inference error rate and inference delay, as indicated by (6) and (7).…”

“…Calculate optimal offloading probability (̺ * m ); 5 Calculate the optimal inference error rate threshold (ǫ th * m ); 6 Evaluate the inference error rate, i.e., ǫ L m = g(Q, D V m ); 7 if ǫ L m ≥ ǫ th * m then 8 Offload the DLT to the MES; Observation 4. There exists a trade-off between the inference error rate and inference delay, as indicated by (6) and (7). In other words, the achievement of a low inference error rate is at the expense of inference delay.…”

Intelligent Network Slicing for V2X Services Toward 5G

“…Thus, collected mobile data, generated from heterogenous devices, needs to be pre-processed and converted into network-wide observation of vehicular networks, which can be used by SDCon to capture a global view on vehicular networks. important similarity with videos, which can be interpreted as a series of "images" [14]. According to this view, we can mark the collected mobile data on the map at its corresponding coordinate, which make a "snapshot", analogous to "image", of vehicular networks.…”

“…The first part is reward function in terms of latency and reliability requirements for V2X services. The second part is cost function for usage of communication, storage and computation resources 14. However, in our RL model, the reward function (i.e., network operating revenue function) is related to large set of variables, which will bring the curse of dimensionality to traditional RL algorithms.…”