4660 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO.

3, JUNE 2019

Blockchain for Secure and Efficient Data Sharing in

Vehicular Edge Computing and Networks
Jiawen Kang , Rong Yu , Member, IEEE, Xumin Huang, Maoqiang Wu, Sabita Maharjan , Member, IEEE,
Shengli Xie , Senior Member, IEEE, and Yan Zhang , Senior Member, IEEE

Abstract—The drastically increasing volume and the growing information mainly includes traffic-related data, such as road
trend on the types of data have brought in the possibility of real- and weather conditions, and parking lot occupancy. The sub-
izing advanced applications such as enhanced driving safety, and jective information includes things such as rating of a hotel
have enriched existing vehicular services through data sharing
among vehicles and data analysis. Due to limited resources with and quality of vehicular services [4]. Sharing of data has made
vehicles, vehicular edge computing and networks (VECONs) i.e., it possible to realize goals such as improved driving safety, and
the integration of mobile edge computing and vehicular networks, to obtain higher service quality during travelling.
can provide powerful computing and massive storage resources. Due to resource constraints, vehicles cannot support massive
However, road side units that primarily presume the role of vehic- data storage and large-scale data sharing. Vehicle-generated
ular edge computing servers cannot be fully trusted, which may
lead to serious security and privacy challenges for such inte- data becomes increasingly fine-grained and complex, which
grated platforms despite their promising potential and benefits. increases the burden on data transmission. Meanwhile, the
We exploit consortium blockchain and smart contract technolo- data more locally relevant for vehicles has spatial scope and
gies to achieve secure data storage and sharing in vehicular explicit lifetime of utility, such as current traffic information
edge networks. These technologies efficiently prevent data sharing at an intersection, which requires low latency and location
without authorization. In addition, we propose a reputation-based
data sharing scheme to ensure high-quality data sharing among awareness for vehicular data sharing [2]. To address these
vehicles. A three-weight subjective logic model is utilized for challenges, mobile edge computing is a promising paradigm
precisely managing reputation of the vehicles. Numerical results that can be embedded at the network edge infrastructures,
based on a real dataset show that our schemes achieve reasonable e.g., roadside units (RSUs), to support massive data stor-
efficiency and high-level of security for data sharing in VECONs. age, computing and sharing close to the vehicles [2], [5].
Index Terms—Blockchain, reputation management, security Vehicular networks integrated with mobile edge computing
and privacy, smart contracts, vehicular edge computing. are evolving toward vehicular edge computing and networks
(VECONs) [6].
Security and privacy issues are critical challenges for
I. I NTRODUCTION VECONs. RSUs in VECONs play an important role to tem-
ITH rapid development of vehicular telematics and
W applications, vehicles generate a huge amount and sev-
eral different types of data. For example, a self-driving vehicle
porally store and manage vehicular data. But the RSUs are
semi-trusted as they are usually distributed along the road
without any strong security measures, thus making them vul-
can create 1 GB data per second from cameras, radar, GPS, nerable to being compromised by attackers [2], [7], [8].
etc. [1]. Moreover, vehicles can cooperatively collect and share Vehicles therefore may not be willing to upload their data
data of common interest [2], [3]. Data collected by the vehicles to the RSUs because of privacy concerns. Likewise, peer to
consists of objective and subjective information. The objective peer (P2P) data sharing among vehicles raises the issues such
as data access without authorization and the need of ensuring
Manuscript received April 17, 2018; revised September 7, 2018; accepted security in a decentralized manner. These challenges influence
October 2, 2018. Date of publication October 11, 2018; date of current ver-
sion June 19, 2019. This work was supported in part by the NSFC under the sharing of vehicular data, and thus hinder the pace for
Grant 61379115, Grant 61422201, Grant 61501127, Grant 61370159, Grant development of VECONs [9].
61503083, Grant U1301255, and Grant U1501251, in part by the Science and Recently, blockchain technology has attracted growing
Technology Program of Guangdong Province under Grant 2015B010129001,
Grant 2015B010106010, Grant 2016A030313705, Grant 2014B090907010, attention and research work in the context of vehicular
and Grant 2015B010131014, and in part by the Projects funded by the networks because of its characteristics of decentralization,
Research Council of Norway under Grant 240079/F20. (Corresponding anonymity and trust. Blockchain can facilitate establish-
author: Yan Zhang.)
J. Kang, R. Yu, X. Huang, M. Wu, and S. Xie are with the School of ing a secure, trusted and decentralized intelligent trans-
Automation, Guangdong University of Technology, Guangzhou 510006, China port ecosystem, to address data sharing problems thus
(e-mail: kjwx886@163.com; yurong@ieee.org; huangxu_min@163.com; contributing in creating better usage of the transport
maoqiang.wu@vip.163.com; shlxie@gdut.edu.cn).
S. Maharjan is with the Simula Metropolitan Center for Digital infrastructures and resources [9]–[11]. Singh and Kim [12]
Engineering, Norway and University of Oslo, 0316 Oslo, Norway (e-mail: presented an intelligent vehicle-trust point mechanism using
sabita@simula.no). blockchain to support secure communications among vehi-
Y. Zhang is with Department of Informatics, University of Oslo, Norway
(e-mail: yanzhang@ieee.org). cles. However, due to high cost to establish a public
Digital Object Identifier 10.1109/JIOT.2018.2875542 blockchain in resource-limited vehicles, the existing methods
2327-4662 c 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
KANG et al.: BLOCKCHAIN FOR SECURE AND EFFICIENT DATA SHARING IN VECONs 4661

do not work well in P2P data sharing among vehicles in

VECONs.
Motivated by these developments, we exploit the consortium
blockchain technology to develop a secure P2P data sharing
system for vehicular data named vehicular blockchain in this
paper. Consortium blockchain is a specific blockchain with
multiple preselected nodes to establish the distributed shared
database with moderate cost [13], [14]. Here, the preselected
nodes are RSUs. Vehicular blockchain is established on RSUs
to publicly audit and store shared data and records of data
sharing. Yang et al. [8] proposed a public blockchain-based
trust management system, wherein vehicles validate received
messages from neighboring vehicles using Bayesian inference
model. Unlike that in [8], we also utilize smart contracts to
design vehicular data storage and sharing schemes, which are
self-executing scripts residing on blockchains and allow for
distributed automation of multistep processes. These smart
contract-based schemes enable data management automation
with high efficiency, and defend against second-hand data shar-
ing without authorization as well. Moreover, data quality is a
core element of the development of vehicular data sharing [9].
Vehicles as data sources may provide irrelevant or incorrect Fig. 1. Secure P2P data sharing system using consortium blockchains.
information to other vehicles due to defective sensors, compro-
mised firmware, or even selfish purpose [4], [8], [15]. Previous
researches have indicated that the quality of data depends on computations, collect local data from sensing devices, and
the vehicles’ reputation [16]. It is essential to design a mecha- upload data to the edge layer [2]. In the edge layer, several
nism to quantify vehicles’ reputation based on the interactions nearby RSUs (edge nodes) deployed along the road can be
among vehicles [4], [8]. Vehicles choose the best data provider combined to form a vehicular edge cluster. Each vehicle com-
according to reputation. municates with the nearest RSU to access a local vehicular
The main contributions of this paper are summarized as edge cluster. Vehicular edge clusters temporarily store the data
follows. from vehicles and deliver the data to a central cloud through
1) We propose to utilize consortium blockchain to establish wired connections if necessary. The central cloud in the cloud
a secure and distributed vehicular blockchain for data layer manages all vehicular edge clusters. This central cloud
management in VECONs. can be a data center of the ITS, which can store massive
2) We deploy smart contracts on the vehicular blockchain data permanently and carry out complicated and delay-tolerant
to achieve secure and efficient data storage on RSUs, computing tasks for vehicles. While the frequently used data
and data sharing among vehicles. and time-sensitive tasks from vehicles can be stored and exe-
3) We develop a reputation-based data sharing scheme with cuted at the edge of vehicular networks, e.g., local vehicular
three-weight subjective logic (TWSL) model to choose edge clusters.
more reliable data source to improve data credibility.
The rest of this paper is organized as follows. We introduce
the core system components of secure P2P data sharing sys- B. Vehicular Blockchain and Smart Contracts
tem using blockchain in Section II. We illustrate secure and To decrease cost of establishing a blockchain, unlike tradi-
efficient data storage and sharing schemes running on vehic- tional public blockchains [11], [17], [18], we utilize consor-
ular blockchain in Section III. We propose a reputation-based tium blockchain technology to form a vehicular blockchain,
data sharing scheme with subjective logic model for high- which performs distributed data storage and secure data shar-
quality sharing in Section IV. We provide security analysis ing. A consortium blockchain is a special blockchain in which
and numerical results in Section V. Section VI concludes this the consensus process is executed on preselected edge nodes,
paper. e.g., RSUs. Here, the consensus process is an important data
audit stage before adding the data into vehicular blockchain.
Some RSUs are chosen and authorized to carry out the consen-
II. C ORE S YSTEM C OMPONENTS FOR D ISTRIBUTED DATA
sus process in the vehicular blockchain. For distributed data
S TORAGE AND D ECENTRALIZED DATA S HARING
storage and secure P2P sharing among vehicles, we also design
A. Vehicular Edge Computing and Networks a model of vehicular blockchain using smart contract technolo-
VECONs are composed of a user layer, an edge layer, and gies in VECONs. This model includes data requestors, data
a cloud layer as shown in Fig. 1. In the user layer, vehicles providers, edge nodes, a data storage smart contract (DSSC)
equipped with onboard units can access services by com- and an information sharing smart contract (ISSC) running on
municating with RSUs. The onboard units perform simple the vehicular blockchain.
4662 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 3, JUNE 2019

vehicular blockchain, a consensus should be reached

among the edge nodes through a mechanism named
proof-of-storage in DSSC. A local controller gener-
ates a storage address list of raw data for each data
provider. The data provider searches and reads its raw
data according to the corresponding storage address list.
3) Proof-of-Storage About Storage Resource Contributions:
The edge node with the most contribution on stor-
age space in every vehicular edge cluster is rewarded
by vehicle coins over a period of time, which is an
incentive to encourage edge nodes to provide enough
storage space for local storage. Here, the vehicle coin
Fig. 2. Secure data storage and sharing using blockchain in vehicular edge is a specific crypto-currency for VECONs. The total
networks.
amount of contributed storage is recorded by a local
controller, which is the proof-of-storage for edge nodes
about storage resource contributions.
1) Data Requestors and Data Providers: Vehicles play dif- 4) ISSC for Decentralized Data Sharing: There are three
ferent roles in P2P vehicular data storage and sharing: data components in ISSC as follows.
requestors and data providers. The data requestors apply for 1) Metadata: A data provider first generates an index of
shared data from the data providers. The data providers collect its raw data as a metadata before uploading to vehicular
traffic-related information, and share their data stored in edge blockchain using DSSC. The metadata generally con-
nodes for getting rewards based on their contributions [16]. tains a pseudonym of the data provider, storage address
Each vehicle chooses its own role according to data demands of raw data in vehicular edge clusters, data description
and driving plan. (e.g., type, accuracy, size, sampling frequency, collec-
2) Edge Nodes: We consider RSUs at the edge of VECONs tion time, sharing permission, etc.), reputation opinions
are the edge devices (nodes). The RSUs are upgraded to have of vehicles, and digital signature for verification. A data
computational capabilities and storage space for computing sharing record includes entity information of data shar-
and storage services. A certain number of RSUs in the same ing, sharing range and so on. The metadata and sharing
area form a vehicular edge cluster. There are a local controller records are integrated into a block and uploaded to an
and a storage pool in each vehicular edge cluster as shown in edge node for verification among edge nodes. More
Fig. 2. The local controller works as a data broker to manage details are provided in Sections III-B and IV.
data requests from local data requestors. A storage pool stores 2) Proof-of-Work About Data Audit for Edge Nodes: Each
local data uploaded by vehicles. Each data requestor sends a edge node collects and verifies local metadata in its cov-
request about data demand to the nearest RSU after finding erage. All edge nodes broadcast their local metadata to
the best local data provider by ISSC. The data providers will other edge nodes in the vehicular blockchain. Every edge
make decisions about data sharing authorization. The RSUs node periodically structures newly received metadata
act as not only data aggregators in DSSC, but also miners in into a local data block, and competes to find an available
ISSC. hash value based on parameters of the local data block.
3) DSSC for Distributed Data Storage: DSSC for dis- Similar to traditional proof-of-work in Bitcoin [18], this
tributed data storage mainly consists of the following com- hash value should meet a preset difficulty controlled by
ponents. the whole blockchain system to adjust generation speed
1) Raw Data: Due to limited storage resource of data of new data blocks. The fastest edge node adds its local
providers, a variety of raw data, such as information data block to the vehicular blockchain using DSSC after
about ice on road, traffic conditions, parking lot occu- verification by other edge nodes, and thus it gets a cer-
pancy, and rating of a restaurant, are stored in edge tain amount of vehicle coins as rewards. Edge nodes
nodes. These data can be used for various types of can use received vehicle coins to further upgrade their
researches, e.g., data analysis. For security and pri- storage and computation resources.
vacy protection, the raw data should be anonymous,
and should be encrypted and attached with digital signa-
III. S ECURE AND E FFICIENT DATA S TORAGE AND
tures of data providers. The data providers use different
S HARING S CHEMES IN THE V EHICULAR B LOCKCHAIN
pseudonyms to encrypt different raw data for decreas-
ing the relevance of raw data generated by the same data A. Overview of Our Proposed Schemes
provider [19], [20]. In this paper, smart contracts are exploited to achieve secure,
2) Data Blocks: As shown in Fig. 2, edge nodes (i.e., reliable, and efficient data sharing. A smart contract is a script
RSUs) working as data aggregators will periodically resided on blockchains to enable automation of multistep pro-
integrate received raw data into a data block, and broad- cesses, which cannot be modified or interrupted because of
cast the data block to other edge nodes for verification. the distributed nature. For this reason, the usage of smart con-
Before a new data block is inserted into immutable tracts could improve the reliability, efficiency, and security of
KANG et al.: BLOCKCHAIN FOR SECURE AND EFFICIENT DATA SHARING IN VECONs 4663

TABLE I
M AIN S YMBOLS U SED IN T HIS PAPER

Fig. 3. Consensus process for ISSC.

pseudonym being used (PIDki ), and the corresponding

signature SigPIDk and certificate CertPIDk to ensure relia-
i i
the vehicular blockchain. Two smart contracts, i.e., DSSC and bility and truthfulness of the request. After receiving the
ISSC, are deployed on vehicular blockchain to enable secure request, DAGj verifies the request and sends a response
and decentralized data sharing. back to vi . If vi is allowed to upload data, vi will send
Fig. 2 shows that vehicles driving along the road gener- its shared data (Data) as a record after encryption with
ate and upload raw data and their corresponding metadata to the public key PKPIDk of PIDki to DAGj , namely
i
nearby RSUs by DSSC. The raw data will be securely stored
in the vehicular blockchain using proof-of-storage. Meanwhile, vi → DAGj : Record
for efficiently decentralized data sharing, data requestors first = EPKDAGj (Data_1||CertPIDk ||SigPIDk ||timestamp)
i i
search data through ISSC, then find out related information where Data_1 = EPK (Data||timestamp)
PIDki
of data of interest. The data requestors communicate with the
data providers to apply for access authorization. After that, SigPIDk = SignSK (Data_1).
i PIDki
the data requestors pay the data providers using vehicle coins.
3) Step 3 (Generating Data Blocks): Each DAG in the same
With the help of proof-of-work about data audit, raw data,
vehicular edge cluster periodically gathers uploaded
and sharing records are audited and verified by RSUs, then
Records from local vehicles. A DAG j generates a new
added into vehicular blockchain. More details about the pro-
data block with a timestamp and broadcasts them to
posed schemes are given as follows. The main symbols used
other DAGs on vehicular blockchain for audit and ver-
in this paper are listed in Table I.
ification. During a period of time, the DAG with the
most contribution of storage resource in a vehicular
B. Secure Data Storage Scheme Using DSSC edge cluster, that is recorded by the local controller in
1) Raw Data Storage in the Vehicular Blockchain: The the vehicular edge cluster, can work as the leader of
raw data from data providers can be stored in the vehicular block generation in this round. This leader collects all
blockchain through the following steps. received Records and generates a Merkle hash value of
1) Step 1 (System Initialization and Key Generation): In the Records linked to the prior block in the vehicular
the vehicular blockchain, elliptic curve digital signature blockchain [18]. After that, the new block is broad-
algorithm and asymmetric cryptography are used for sys- casted to all DAGs, and then added into the vehicular
tem initialization. Every vehicle becomes a legitimate blockchain.
vehicle after passing identity authentication by a trusted 2) Metadata and Sharing Record Storage in the Vehicular
central authority, e.g., a government department of trans- Blockchain: Both metadata of raw data (as information index)
portation. A legitimate vehicle vi with the true identity and sharing records are stored in RSUs, and are shared among
IDi obtains its public & private keys and the correspond- vehicles using proof-of-work for data audit. More details of
ing certificates (i.e., {PKPIDk , SKPIDk , CertPIDk }νk=1 ) for metadata and sharing record storage are shown as follows.
i i i
encrypting sensing data. When vi carries out system ini- 1) Step 1 (Generating Information Index): Before vi
tialization, vi downloads the latest data information from uploading its Record, the vehicle generates a data
storage pools of nearby edge nodes in the vehicular information index about the Record as follows:
blockchain.
2) Step 2 (Uploading Shared Data): A vehicle vi first sends
an upload request to a local RSU of vehicular edge
cluster. Here, the RSU acts as a local data aggregator
(denoted as DAGj ). This request includes the current
4664 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 3, JUNE 2019

2) Step 2 (Building Information Blocks and Finding Proof- Protocol 1 Distributed Consensus Protocol for DAGs
of-Work): vi sends its data information index to a nearby 1. The leader broadcasts block data to all DAGs in the vehicular
blockchain for verification and audit.
DAG (e.g., DAGj ). DAGj collects all local information DAGj → All : Record = (Block_data||Block_hash
(e.g., indexes) during a certain period, and then encrypts ||CertBSj ||SigDAGj ||timestamp),
and digitally signs these indexes to guarantee authentic- where Block_hash = Hash(Block_Data||timestamp),
ity and accuracy. Fig. 1 shows that the index records SigDAGj = SignSKDAG (Block_data||Block_hash).
j
are structured into blocks. For traceability and verifica- 2. The DAGs broadcast their own audit results to each other for
tion, each block contains a cryptographic hash to the mutual supervision and verification, and then send their replies back
to the leader.
prior blocks in the vehicular blockchain. Similar to that DAGl → DAGj : Reply = EPKDAG (Data_2
in Bitcoin, the DAGs try to find their own valid proof- j
||CertDAGl ||SigDAGl ||timestamp),
of-work about data audit (i.e., a hash value meeting a where Data_2 = (my_result||Rece_results||Comparison),
certain level of difficulty). Each DAG calculates the hash SigDAGl = SignSKDAG (Data_2).
l
value of its block based on a random nonce value ϕ, the 3. The leader adds new block data into vehicular blockchain after
previous block hash value, timestamp, and data blocks’ verifying by DAGs, and broadcasts the block data to all DAGs for
merkel root, and so on (denoted as previousdata ) [21]. storage.
Namely, Hash(ϕ + previousdata ) < Difficulty. Here, DAGj → All : Data_block = (Data_3||SigDAGj ||
Difficulty can be adjusted by the system to control the timestamp),
speed of finding out the specific ϕ. After finding a valid where Data_3 = (Block_data||Block_hash||{CertDAGj }||
proof-of-work (i.e., ϕ), the fastest miner (DAG) broad- timestamp),
casts the block and the specific ϕ to other DAGs in the SigDAGj = SignSKDAG (Data_3).
j
vehicular blockchain. Other DAGs audit and verify the
records in the block and ϕ. If other DAGs agree on the
block, data information in this block will be added to
by information indexes. The data requestors choose their opti-
the vehicular blockchain by a linear, chronological order,
mal data providers according to reputation of providers. More
and the fastest miner (DAG) is awarded by vehicle coins.
details about the reputation calculation are given in Section IV.
3) Step 3 (Carrying Out a Consensus Process): The con-
For example, a data requestor vm sends a data sharing request
sensus process is carried out by authorized DAGs and
(Req) to a data provider vi . This request includes time, the
a leader acted by the fastest DAG with a valid proof-
usage of requested data, sharing times, etc.,
of-work. Fig. 3 shows that the leader broadcasts block
data Block_data with timestamp and its proof-of-work vm → vi : Req = EPKvi (Request||Certvm ||timestamp).
to other authorized DAGs for verification and audit.
For mutual supervision and verification, these DAGs 2) Step 2 (Data Sharing Authorization): After receiving the
audit the block data and broadcast their audit results request Req, vi verifies the identity of vm , and defines the data
with their signatures to each other. After receiving the access constraints based on the request from vm . After that,
audit results, each DAG compares its result with oth- vi sends the access constraints, pseudonyms’ private keys of
ers and sends a reply (Reply) back to the leader. This uploaded data, public key of the data requestor, and so on to
reply consists of the DAG’s audit result my_result, com- a nearby RSU, e.g., RSUj
parison result Comparison, signatures, and records of vi → RSUj : Message
received audit results Rece_results. The leader analyzes
= EPKRSUj (Constraints||SKPIDk ||PKvm ||timestamp||Certvi ).
the received replies from DAGs. If all the DAGs agree i
on the block data, the leader will send records including The ISSC is triggered by Message from vi . RSUs first verify
current audited block data and a corresponding signature the certificate of vi , and check the shared data information of
to all authorized DAGs for storage. After that, this block vi in the vehicular blockchain. The RSUs obtain and integrate
is stored in the vehicular blockchain, and the leader is the shared data stored in the vehicular blockchain according to
awarded by vehicle coins. More details about the con- the given pseudonyms’ private keys of shared data. The shared
sensus process are given in Protocol 1. If some DAGs data is encrypted with the public key of data requestor vm . If
do not agree on the block data, the leader will analyze vi and vm are at the same coverage of a local DAG, the shared
the audit results, and send the block data to these DAGs data will be sent to vm directly. Otherwise, the shared data
once again for audit if necessary [13]. will be sent to a DAG nearby vm
RSUj → RSUj+1 : Shared_data
= EPKRSUj+1 (Data_2||timestamp||CertRSUj )
C. Secure and Efficient Data Sharing Scheme Using ISSC
Data_2 = EPKvm (Data||Certvi ||CertRSUj ||timestamp).
The P2P data sharing process among vehicles using ISSC
consists of the following steps. 3) Step 3 (Recording and Generating Data Sharing Events
1) Step 1 (Uploading Data Sharing Requests): Data in the Vehicular Blockchain): After obtaining the shared data,
requestors first download the latest data blocks in the vehic- the data requestor pays for the provider using vehicle coins,
ular blockchain from DAGs, and search their data of interest and generates a record of the data sharing event, and adds this
KANG et al.: BLOCKCHAIN FOR SECURE AND EFFICIENT DATA SHARING IN VECONs 4665

record as a data block into vehicular blockchain similar to the expressed as

steps in Section III-B. Moreover, similar to [14], each vehi-
cle has a wallet account to store and manage personal vehicle Ti→j = bi→j + γ ui→j . (2)
coins. During a payment process, for privacy protection, we
Here, γ is a given constant given by vehicles, which indicates
use random pseudonyms as public keys of a vehicle’s wallet
the uncertainty effect level on reputation for vehicles. This
account, named wallet addresses, to replace the true address of
constant can be set as 0.5 by default [6].
the wallet account for privacy protection. The mapping rela-
tionships between the wallet account and the corresponding
wallet addresses are recorded in the trusted authority. B. Three-Weight Local Opinions for Subjective Logic
Traditional subjective logic (TSL) is evolved toward mul-
IV. S UBJECTIVE L OGIC FOR R EPUTATION tiweight subjective logic when considering weighting opera-
M ANAGEMENT IN VECON S tions. In VECONs, we consider different weights to formulate
local opinions. Compared with TSL models, the advantages of
In VECONs, vehicles may provide irrelevant data (informa- the TWSL model can obtain more accurate and reliable repu-
tion) to each other although they have similar “genomes” [4]. tation when taking the following weights into consideration.
The vehicles may share false information because of faulty 1) Interaction Frequency: It is known that the higher inter-
sensors, infection from computer viruses, or even selfish pur- action frequency means that the data requestor (vi ) has
pose [8], [15]. If a positive sharing interaction occurs between more prior knowledge about the data provider (vj ) lead-
two vehicles, namely, vm believes that the data shared by vi is ing to more accurate and reliable reputation calculation.
relevant and useful, the relationship from vm to vi is strength- The interaction frequency is the ratio of interaction times
ened and the reputation of vi is enhanced. Besides, previous between the data requestor and the data provider to
researches have also indicated that the higher reputation of average interaction number of the data requestor with
nodes brings higher data quality in mobile crowd sensing [16]. other data providers during a time window T, namely,
Vehicles choose the best data provider according to reputation IFi→j = ([Ni→j
values [8]. It is essential to design a mechanism to quantify ]/Ni ), where Ni→j = (αi + βi ), and
Ni = (1/|M|) m∈M Ni→m . M is the total number of
vehicles’ reputation based on their interactions [4]. We there- vehicle m that interacts with vehicle i (i.e., the data
fore propose to make use of a reputation-based sharing scheme requestor) during a time window. The higher interaction
with TWSL model for high-quality data sharing in this section. frequency brings higher reputation.
Subjective logic is utilized to formulate the individual eval- 2) Event Timeliness: In VECONs, a vehicle is not always
uation of reputation based on occurring interactions, which is trustable and reliable. Both the trustfulness between the
a framework for probabilistic information fusion operated on data provider and requestor, i.e., the reputation of vi to vj
subjective beliefs about the world. The subjective logic utilizes are changing over time. The recent events have a larger
the term opinion to denote the representation of a subjective impact on the local opinion of vi to vj , while the past
belief, and models positive statements, negative statements, events have less impact on this local opinion for more
and uncertainty. It also offers a wide range of logical operators accurate and reliable reputation calculation. The time
to combine and relate different opinions [6]. scale of recent events and past events is trecent , e.g., a
week. Moreover, the negative events have higher weight
A. Local Opinions for Subjective Logic on the local opinions of vehicles than that of the positive
Considering two vehicles vi and vj , the trustworthiness (i.e., events. Here, the weight of positive events is θ , and the
local opinion) of the data requestor vj to the data provider weight of negative events is τ . θ + τ = 1, θ < τ. ζ
vi in subjective logic can be formally described as a local represents the weight of recent events, σ is the weight
opinion vector ωi→j , i.e., ωi→j := {bi→j , di→j , ui→j }, where of past events. ζ + σ = 1, ζ > σ . The weights of
bi→j , di→j , ui→j represent the belief, distrust, and uncertainty, event-timeliness and negative/positive events are com-
respectively. bi→j , di→j , ui→j ∈ [0, 1] and bi→j + di→j + bined together to form a new interaction frequency as
ui→j = 1. Here, follows:
⎧ 
α αi = ζ θ α1i + σ θ α2i
⎪
⎨ bi→j = (1 − ui→j ) α+β (3)
β βi = ζ τβ1i + σ τβ2i
di→j = (1 − ui→j ) α+β (1)
⎪
⎩u
i→j = 1 − s i→j where the number of recent positive events and negative
events are α1i and β1i when the current time t satisfies
where α is the number of positive events, while β is the t ≤ trecent , respectively. When t > trecent , the number of
number of negative events. The uncertainty of local opinion positive and negative past events are α2i and β2i , respec-
vector ui→j depends on the quality of communication between tively. Therefore, the interaction frequency between the
vehicles i and j. The quality of communication si→j refers to data requestor and the data provider is updated as
the probability that data packets of data sharing requests are  
transmitted successfully during communication. According to Ni→j θ ζ α1i + σ α2i + τ ζβ1i + σβ2i
IFi→j = = 1 
. (4)
ωi→j , the reputation value Ti→j represents the expected belief Ni Ni→m
|M|
of vi that vj provides true and relevant data, which can be m∈M
4666 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 3, JUNE 2019

3) Trajectory Similarity: Data collected by vehicles is two trajectories Li and Lj . More specifically,
locally relevant for vehicles, and has spatial scope.
sin ϕ
To enable location awareness and improve data rele-  ,0 < ϕ ≤ π
direction Li , Lj = 1 2 |sin(ϕ+ π )| 2 π (9)
2 + , 2 < ϕ ≤ π.
2
vance, trajectory similarity is taken into consideration on 2
reputation calculation during data sharing among vehi-
Therefore, the overall weight of reputation for local
cles. The higher trajectory similarity means the sharing
opinions is
data from the data provider is more relevant leading 
to high-quality, more accurate and reliable data shar- δi→j = ρ1 IFi→j + ρ2 SIM Li , Lj (10)
ing [20]. The trajectory coefficients of vehicles are
where ρ1 + ρ2 = 1, and 0 < ρ1 ≤ 1, 0 < ρ2 ≤ 1.
represented by υ = {speed, location, direction}. The
weights of corresponding coefficients in υ are ψ1 , ψ2 ,
and ψ3 , and ψ1 + ψ2 + ψ3 = 1. The similarity C. Combining Recommended Opinions
degree of two trajectory segments (denoted as Li and After calculating the weights, the opinions are com-
Lj ) for vehicle i and vehicle j is SIM(Li , Lj ), which is bined into a common opinion in the form ωx→j rec :=
calculated as {bx→j , dx→j , ux→j }, where
rec rec rec

  ⎧ rec 
⎪ bx→j = δx→j bx→j
⎪ 1
SIM Li , Lj = 1 − DISS Li , Lj . (5) ⎪ δx→j
⎪
⎨ rec x∈X x∈X

Here, DISS(Li , Lj ) is the normalized dissimilarity dx→j =  1δ δx→j dx→j (11)
x→j
⎪
⎪ x∈X x∈X

of two trajectory segments Li and Lj , and is ⎪
⎪
⎩ ux→j = δx→j ux→j
rec  1
defined as δ x→j
x∈X x∈X
   where x ∈ X is a set of recommended vehicles that have
DISS Li , Lj = ψ1 speed Li , Lj + ψ2 location Li , Lj
 interacted with vj . Thus, the subjective opinions from different
+ ψ3 direction Li , Lj . (6)
recommenders (neighboring vehicles) are integrated into one
single opinion, which is named as the recommended opinion
We consider that DISS(Li , Lj ) depends on differences
according to each opinion’s weights [23].
of speed, location, and direction for two trajectory seg-
ments. The speed difference of two trajectory segments
can be expressed as D. Combining Local Opinions With Recommended Opinions


After obtaining shared data from data providers, a data

Vave (Li ) − Vave Lj
requestor has a subjective opinion (i.e., local opinion) for each
speed Li , Lj =    (7)
max V(Li ), V Lj data provider based on interaction histories. This local opinion
should still be considered while forming the final opinion to
where V(Li ) and V(Lj ) are the speeds of vehicles i and avoid cheating [23]. The final opinion of vi to vj is formed as
j during their trajectory segments, respectively. Vave (Li ) final := {bfinal , d final , ufinal }, where bfinal , d final , and ufinal are,
ωx→j x→j x→j x→j i→j i→j i→j
and Vave (Lj ) are the average speeds of these two vehi- respectively, calculated as
cles. We use location(Li , Lj ) to describe the location ⎧
⎪ bi→j urec +brec
x→j ui→j
difference of trajectory segments. The number of sam- ⎪
⎪ bfinal = ui→j +ux→j
⎪
⎨
i→j rec −urec u
x→j x→j i→j
ple points of Li and Lj are, respectively, denoted as e x→j +dx→j ui→j
di→j urec rec
di→j = ui→j +urec −urec ui→j
final (12)
and k during a time window T. The sets of sample ⎪
⎪ x→j x→j
points in chronological order are {Pi1 , Pi2 , . . . , Pie } and ⎪ final
⎪ urec ui→j
⎩ ui→j = x→j
.
ui→j +urec −urec ui→j
{Pj1 , Pj2 , . . . , Pjk }. We measure the similarity of the tra- x→j x→j

jectory segments by the longest common subsequence Similar to (2), the final reputation of vi to vj is
(LCS) that has been widely used in time series trajectory
clustering. The LCS is utilized to match two sequences
final
Ti→j = bfinal
i→j + γ ui→j .
final
(13)
by allowing them to stretch without rearranging the
sequence of the elements [22]. For trajectory segments E. Choosing the Optimal Data Provider for Data Sharing
Li and Lj , the LCS is described as lcs(Li , Lj ) = {Pie = For a data requestor, it chooses an optimal data provider
Pjk |e = k}, here, e ∈ {1, 2, . . . , E}, k ∈ {1, 2, . . . , K}. by comparing the final reputation values of data provider can-
Hence, the location difference of trajectory segments didates. There exists a candidate with the highest reputation
location(Li , Lj ) is given by value for each data requestor during a period of time. The
   optimal data provider can be found by
 max(e, k) − num lcs Li , Lj  
location Li , Lj = (8) v∗j = arg max Ti→j
final
. (14)
max(e, k) j∈M

where num[lcs(Li , Lj )] is the number of points in LCS As shown in Fig. 4, the operations of finding the optimal
for trajectory segments Li and Lj . The directory differ- data provider consist of the following steps.
ence of two trajectory segments is the angle between 1) Step 1: A data requestor vi first downloads the latest
two trajectory segments. Here, we use ϕ as the angle of data blocks on the vehicular blockchain. vi searches
KANG et al.: BLOCKCHAIN FOR SECURE AND EFFICIENT DATA SHARING IN VECONs 4667

and smart contract technologies to ensure security and pri-

vacy protection during data storage and sharing. Consortium
blockchain ensures data traceability, and the automatic exe-
cution of smart contracts protects data security sharing. For
data providers, the decentralized characteristic of vehicular
blockchain’s architecture defends against data security risks
brought by centralized data storage [9]. The transparency
characteristic of vehicular blockchain during data sharing
avoids second-hand sharing without authentication from data
providers. The anonymous operations using pseudonyms dur-
ing data storage and sharing bring privacy protection for data
providers and data requestors. We provide more details about
the blockchain-related security performances as follows [25].
1) Get Rid of a Trusted Intermediary: With the help of
the robust and scalable consortium blockchain, vehicles
Fig. 4. Operations of finding the optimal data provider.
can share data with others in a P2P manner without
involvement of a globally trusted intermediary [14].
2) Sharing Record Authentication: All sharing records are
data provider candidates through information indexes of publicly audited and authenticated by other entities. It is
shared data. Next, vi finds candidates’ local opinions impossible to compromise all entities due to overwhelm-
given by other vehicles (denoted as vx ) that had inter- ing cost. So the sharing records with errors can still be
acted with the candidates. The local opinion of vx for discovered and corrected before structuring into a block.
a certain candidate (e.g., vj ) includes local opinion vec- 3) Data Unforgeability: No adversary can act as vehicles
tor and the corresponding overall weight of reputation to corrupt the vehicular blockchain. It is because that
for its local opinion, i.e., ωx→j := {bx→j , dx→j , ux→j } the adversary cannot forge a digital signature of any
and δx→j . vehicle, or gain control over the majority of the net-
2) Step 2: vi combines these local opinions from vx to work’s resources [25]. An adversary only controlling a
calculate a common recommended opinion ωx→j rec :=
few of RSUs in the vehicular blockchain cannot learn
{bx→j , dx→j , ux→j }. For vi , it generates its local opinion
rec rec rec
anything about the raw data, as it is encrypted with keys
for vj according to its interaction history and received of vehicles.
safety messages during driving. For driving safety, vehi- 4) Secure Self-Execution: DSSC and ISSC, that are run on
cles are required to periodically broadcast safety mes- vehicular blockchain, are autonomous, self-executed and
sages (consisting of current positions, speeds, directions, self-maintained in the form of computer codes. These
and so on) to neighboring vehicles. These safety mes- smart contracts do not need mutual trust, and are com-
sages increase the awareness of vehicles about their pletely automatic. So no human factor is needed, and no
neighbors’ whereabouts and warn drivers of dangerous human factor can control these smart contracts once it
situations [24]. The local opinions of vehicles can be goes into effect [9].
calculated based on the received safety messages during
a period of time. Thus, vi obtains the final reputation
values for candidates using (13) in a time window. B. Numerical Results
3) Step 3: vi compares the final reputation values of can- We evaluate the performance of the proposed TWSL scheme
didates and uses (14) to find out the optimal data based on a real dataset from San Francisco Yellow Cabs [26].
provider. vi interacts with the data provider to request This dataset includes mobility traces of 536 urban taxis over
authorization of data sharing. a period of one month [27]. We take 100 taxis as exam-
4) Step 4: After verification, the data provider generates ples and choose an observation area, whose latitude is from
a smart contract for vi . More details about secure data 37.747 to 37.78, and the longitude is from −122.48 to
sharing using ISSC are given in Section III. −122.385, as shown in Fig. 5. The observed area is approxi-
5) Step 5: vi uploads its local opinion as a new reputation mately 8.34 × 3.67 km2 . The average time gap between two
opinion for its optimal data provider (e.g., vj ). Similar trace records is 43.34 s, i.e., data collection period of vehi-
to the data sharing events in Section III-C, this updated cles [27]. We set the update period of reputation is 15 min,
local opinion will be formed as a data block of vehicular and the observation time of our simulation is 240 min. In an
blockchain. urban area, the vehicles often take familiar routes in a spec-
ified time period, such as similar trajectories from home to
work in the day time [20]. So the vehicles would like to share
V. S ECURITY A NALYSIS AND N UMERICAL R ESULTS
data for obtaining rewards and promoting vehicular services.
A. Security Analysis for Vehicular Blockchain More parameters about our simulation are listed in Table II.
Unlike traditional communication security and privacy pro- A vehicle sends data requests to data providers with a
tection, our vehicular blockchain uses consortium blockchain high reputation. The quality of communication si→j affects
4668 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 3, JUNE 2019

Fig. 6. Reputation comparison of ten abnormal vehicles during 120 min.

Fig. 5. Spatial distribution of vehicles’ trajectories.
TABLE II
PARAMETER S ETTING IN THE S IMULATION

Fig. 7. Reputation changes of an abnormal vehicle over time.

the uncertainty of local opinion vector ui→j . Vehicles calcu-
late reputation based on their local opinions and recommended
opinions from other vehicles. Our proposed TWSL scheme lower than that of TSL scheme. It is because that all the
is used to calculate the reputation of vehicles according to reputation opinions are combined and weighted adequately
the interaction events between vehicles. We compare TWSL by considering prior knowledge (interaction frequency, event
scheme with a widely accepted TSL model using a lin- timeliness, and trajectory similarity). In this way, we pay more
ear function to calculate reputation [28]. The linear function attention to reputation opinions with better quality and avoid
is represented as: i→j = ωx→j ave + (1 − ω) las , where
i→j being misleading from reputation opinions with lower quality.
x→j
ave = bave + 0.5 ∗ uave and  las = blas + 0.5 ∗ ulas .
x→j x→j i→j i→j i→j As a result, highly accurate reputation computation is achieved
bave ave
x→j and ux→j are two average values of bx→j and ux→j from using TWSL scheme, thus ensuring high-quality data sharing
recommended opinions of other vehicles, respectively. blas i→j among vehicles.
and ulas
i→j are the latest parameters in local opinion of vehicle Fig. 7 shows the reputation changes of an abnormal vehicle
i for vehicle j. ω is the weight and can be set as 0.5 [6], [28]. over time. Here, the misbehavior probability of this abnormal
We set that all the abnormal vehicles initially pretend to vehicle is 70%. The abnormal vehicle first pretends to provide
behave normally within a short period of time (20 min.), that high-quality data to other vehicles for winning trust from a
provide high-quality data with high value of  , e.g.,  = 0.8. target vehicle. The initial reputation value for the target vehi-
Here,  is the probability that an abnormal vehicle behaves cle is w0 = [0.9, 0.05, 0.05]. Meanwhile, the abnormal vehicle
normally in order to hide its malicious intent. Their initialized randomly interacts with other normal vehicles. Due to the mis-
reputation values are represented by w0 = [0.64, 0.16, 0.2]. behavior events, the reputation value of the abnormal vehicle is
After the camouflage time, they do some misbehaviors and  decreased continuously over time, as illustrated in Fig. 7. The
becomes 0.2. To detect the misbehavior, the system updates reputation updated by TWSL scheme is much more accurate,
the reputation values in every time period. leading to lower reputation value for the abnormal vehicle.
As shown in Fig. 6, we randomly choose ten abnormal After 10 min, the reputation value is descended to 0.5, which
vehicles for reputation update during the observation time. is below than that of TSL scheme. This means that the abnor-
The abnormal vehicles randomly interact with normal vehicles mal vehicle has a higher probability for being detected when
[1, 4] times during 60 min. All the reputation values of the the threshold of trust is 0.5 in the TWSL scheme.
abnormal vehicles are lower than the initial reputation value For an abnormal vehicle, its reputation value is affected
w0 . The reputation values calculated by TWSL scheme are by different misbehavior probabilities. Fig. 8 shows the
KANG et al.: BLOCKCHAIN FOR SECURE AND EFFICIENT DATA SHARING IN VECONs 4669

consortium blockchain and smart contract technologies to

achieve secure and efficient data storage and data sharing.
These technologies efficiently prevent second-hand data shar-
ing without authorization. In addition, we have proposed a
reputation-based data sharing scheme with the TWSL model
considering interaction frequency, event timeliness, and tra-
jectory similarity. This scheme can achieve accurate reputation
management for high-quality data sharing among vehicles. The
vehicles can choose the optimal data providers with high-
quality data during sharing in VECONs. Security analysis
shows that our system ensures security of data storage and data
sharing. Numerical results indicate that the proposed TWSL
scheme has great advantages over the traditional reputation
Fig. 8. Reputation changes of an abnormal vehicle under different misbe- schemes in improving detection rate of abnormal vehicles to
havior probability. ensure security during data sharing.

R EFERENCES
[1] W. Xu et al., “Internet of Vehicles in big data era,” IEEE/CAA J.
Automatica Sinica, vol. 5, no. 1, pp. 19–35, Jan. 2018.
[2] J. Ni, A. Zhang, X. Lin, and X. S. Shen, “Security, privacy, and fairness
in fog-based vehicular crowdsensing,” IEEE Commun. Mag., vol. 55,
no. 6, pp. 146–152, Jun. 2017.
[3] Z. Su, Y. Hui, and Q. Yang, “The next generation vehicular networks:
A content-centric framework,” IEEE Wireless Commun., vol. 24, no. 1,
pp. 60–66, Feb. 2017.
[4] Q. Yang, B. Zhu, and S. Wu, “An architecture of cloud-assisted infor-
mation dissemination in vehicular networks,” IEEE Access, vol. 4,
pp. 2764–2770, 2016.
[5] X. Huang, R. Yu, J. Kang, Y. He, and Y. Zhang, “Exploring mobile edge
computing for 5G-enabled software defined vehicular networks,” IEEE
Wireless Commun., vol. 24, no. 6, pp. 55–63, Dec. 2017.
[6] X. Huang, R. Yu, J. Kang, and Y. Zhang, “Distributed reputation
management for secure and efficient vehicular edge computing and
Fig. 9. Detection rates under different trust thresholds. networks,” IEEE Access, vol. 5, pp. 25408–25420, 2017.
[7] D. Huang, S. Misra, M. Verma, and G. Xue, “PACP: An efficient
pseudonymous authentication-based conditional privacy protocol for
VANETs,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 3, pp. 736–746,
misbehavior probability impacts the reputation values. With Sep. 2011.
[8] Z. Yang, K. Yang, L. Lei, K. Zheng, and V. C. M. Leung,
the higher misbehavior probability, the average reputation “Blockchain-based decentralized trust management in vehicular
value of the abnormal vehicles using our TWSL scheme is networks,” IEEE Internet Things J., to be published. [Online].
lower than that of TSL scheme because of the considered Available: https://ieeexplore.ieee.org/abstract/document/8358773,
doi: 10.1109/JIOT.2018.2836144.
weights. For example, when the misbehavior probability is [9] L. Yue, H. Junqin, Q. Shengzhi, and W. Ruijin, “Big data model of
60%, our TWSL scheme is 38% lower than that of TSL security sharing based on blockchain,” in Proc. 3rd Int. Conf. Big Data
scheme. So our TWSL scheme is sensitive for the misbehav- Comput. Commun. (BIGCOM), 2017, pp. 117–121.
[10] Z. Xiong, Y. Zhang, D. Niyato, P. Wang, and Z. Han, “When mobile
ior, although malicious vehicles try to camouflage themselves. blockchain meets edge computing,” IEEE Commun. Mag., vol. 56, no. 8,
It is beneficial to detect and eliminate the misbehaviors timely pp. 33–39, Aug. 2018.
in VECONs. [11] Y. Yuan and F.-Y. Wang, “Towards blockchain-based intelligent trans-
portation systems,” in Proc. IEEE 19th Int. Conf. Intell. Transp. Syst.
We study detection rate of abnormal vehicles using TWSL (ITSC), Nov. 2016, pp. 2663–2668.
scheme and TSL scheme within 60 min. Fig. 9 shows the [12] M. Singh and S. Kim, “Trust bit: Reward-based intelligent vehicle com-
proposed TWSL scheme can distinguish much more abnor- mination using blockchain paper,” in Proc. IEEE 4th World Forum
Internet Things, Feb. 2018, pp. 62–67.
mal vehicles compared to TSL scheme. Note that, with higher [13] J. Kang et al., “Enabling localized peer-to-peer electricity trading among
threshold value of trust, more abnormal vehicles will be dis- plug-in hybrid electric vehicles using consortium blockchains,” IEEE
tinguished. When the threshold value of trust is 0.35, the Trans. Ind. Informat., vol. 13, no. 6, pp. 3154–3164, Dec. 2017.
[14] Z. Li et al., “Consortium blockchain for secure energy trading in indus-
recognition rate of abnormal vehicles in TWSL scheme has trial Internet of Things,” IEEE Trans. Ind. Informat., vol. 14, no. 8,
already been more than 100%, while that of TSL scheme pp. 3690–3700, Aug. 2018.
is only 50%. Due to higher detection rate in the proposed [15] Q. Yang and H. Wang, “Toward trustworthy vehicular social networks,”
IEEE Commun. Mag., vol. 53, no. 8, pp. 42–47, Aug. 2015.
TWSL scheme, potential security threats can be removed more [16] S. Delgado-Segura, C. Tanas, and J. Herrera-Joancomartí, “Reputation
effectively, which makes VECONs more secure. and reward: Two sides of the same bitcoin,” Sensors, vol. 16, no. 6,
pp. 1–23, 2016.
[17] Y. Zhang and J. Wen, “The IoT electric business model: Using
VI. C ONCLUSION blockchain technology for the Internet of Things,” Peer-to-Peer Netw.
Appl., vol. 10, no. 4, pp. 983–994, 2017.
In this paper, we have presented a secure P2P data shar- [18] S. Nakamoto. (2012). Bitcoin: A Peer-to-Peer Electronic Cash System,
ing system in vehicular computing and networks. We utilized 2009. [Online]. Available: http://www.bitcoin.org/bitcoin.pdf
4670 IEEE INTERNET OF THINGS JOURNAL, VOL. 6, NO. 3, JUNE 2019

[19] J. Kang et al., “Location privacy attacks and defenses in cloud-enabled Maoqiang Wu is currently pursuing the Ph.D.
Internet of Vehicles,” IEEE Wireless Commun., vol. 23, no. 5, pp. 52–59, degree in networked control systems at the
Oct. 2016. Guangdong University of Technology, Guangzhou,
[20] R. Yu et al., “MixGroup: Accumulative pseudonym exchanging for loca- China.
tion privacy enhancement in vehicular social networks,” IEEE Trans. His current research interests include blockchain,
Depend. Secure Comput., vol. 13, no. 1, pp. 93–105, Jan./Feb. 2016. resource management in wireless communications
[21] I. Alqassem and D. Svetinovic, “Towards reference architecture for cryp- and networking.
tocurrencies: Bitcoin architectural analysis,” in Proc. IEEE Int. Conf.
Internet Things (iThings) IEEE Green Comput. Commun. (GreenCom)
IEEE Cyber Phys. Soc. Comput. (CPSCom), 2014, pp. 436–443.
[22] L. Zheng, Q. Feng, W. Liu, and X. Zhao, “Discovering trip hot routes
using large scale taxi trajectory data,” in Proc. Int. Conf. Adv. Data Min.
Appl., 2016, pp. 534–546.
[23] Y. Liu, K. Li, Y. Jin, Y. Zhang, and W. Qu, “A novel reputation com-
putation model based on subjective logic for mobile ad hoc networks,”
Future Gener. Comput. Syst., vol. 27, no. 5, pp. 547–554, 2011. Sabita Maharjan (S’09–M’13) received the Ph.D.
[24] J. Kang, R. Yu, X. Huang, and Y. Zhang, “Privacy-preserved pseudonym degree in networks and distributed systems from
scheme for fog computing supported Internet of Vehicles,” IEEE Trans. the University of Oslo, Oslo, Norway, and Simula
Intell. Transp. Syst., vol. 19, no. 8, pp. 2627–2637, Aug. 2018. Research Laboratory, Fornebu, Norway, in 2013.
[25] G. Zyskind, O. Nathan, and A. Pentland, “Decentralizing privacy: Using She is currently a Senior Research Scientist
blockchain to protect personal data,” in Proc. IEEE Security Privacy with the Simula Metropolitan Center for Digital
Workshops (SPW), 2015, pp. 180–184. Engineering, Oslo, Norway, and an Associate
[26] C Projects. (2013). [Online]. Available: http://www.yellowcabsf.com/ Professor (adjunct position) with the University of
[27] M. A. Hoque, X. Hong, and B. Dixon, “Analysis of mobility patterns for Oslo. Her current research interests include wireless
urban taxi cabs,” in Proc. Int. Conf. Comput. Netw. Commun. (ICNC), networks, network security and resilience, smart grid
2012, pp. 756–760. communications, cyber-physical systems, machine-
[28] S. Zhong, J. Chen, and Y. R. Yang, “Sprite: A simple, cheat-proof, credit- to-machine communications, and software defined wireless networking.
based system for mobile ad-hoc networks,” in Proc. INFOCOM 22nd
Annu. Joint Conf. IEEE Comput. IEEE Commun. Soc., vol. 3, 2003,
pp. 1987–1997.

Shengli Xie (M’01–SM’02) received the M.S.

degree in mathematics from Central China Normal
Jiawen Kang received the M.S. and Ph.D. degrees
University, Wuhan, China, in 1992, and the Ph.D.
from the Guangdong University of Technology,
degree in control theory and applications from the
Guangzhou, China, in 2015 and 2018, respectively.
South China University of Technology, Guangzhou,
His current research interests include blockchain,
China, in 1997.
security and privacy protection in wireless commu-
He is currently the Director of the Laboratory
nications and networking.
for Intelligent Information Processing and a
Full Professor with the Faculty of Automation,
Guangdong University of Technology, Guangzhou.
He has authored 2 monographs and over 100 sci-
entific papers in journals and conference proceedings. He holds a dozen of
patents. His current research interests include automatic control and signal
processing, and focus on blind signal processing.
Rong Yu (M’08) received the Ph.D. degree from
Tsinghua University, Beijing, China, in 2007.
He was with the School of Electronic and
Information Engineering, South China University of
Technology, Guangzhou, China. In 2010, he joined
the Institute of Intelligent Information Processing,
Guangdong University of Technology, Guangzhou, Yan Zhang (SM’10) received the Ph.D. degree
where he is currently a Full Professor. His current from the School of Electrical and Electronics
research interests include wireless networking and Engineering, Nanyang Technological University,
mobile computing in featured environments, such Singapore.
as edge cloud, connected vehicles, smart grid, and He is currently a Full Professor with the
Internet of Things. He has co-invented over 30 patents and has authored or Department of Informatics, University of Oslo, Oslo,
co-authored over 100 international journal and conference papers. He was a Norway. His current research interests include next-
member of the Home Networking Standard Committee in China, where he generation wireless networks leading to 5G, green
led the standardization work of three standards. and secure cyber-physical systems, such as smart
grid, healthcare, and transport.
Dr. Zhang is an Associate Technical Editor of the
IEEE Communications Magazine, an Editor of IEEE Network Magazine, the
IEEE T RANSACTIONS ON G REEN C OMMUNICATIONS AND N ETWORKING,
IEEE C OMMUNICATIONS S URVEYS AND T UTORIALS, and the IEEE
Xumin Huang is currently pursuing the Ph.D.
I NTERNET OF T HINGS J OURNAL, and an Associate Editor of IEEE ACCESS.
degree in networked control systems at the
He has served as the Chair for a number of conferences, including IEEE
Guangdong University of Technology, Guangzhou,
GLOBECOM 2017, IEEE PIMRC 2016, IEEE CloudCom 2016, IEEE
China.
ICCC 2016, IEEE CCNC 2016, WCSP 2016, IEEE SmartGridComm 2015,
His current research interests include network per-
and IEEE CloudCom 2015. He serves as a TPC member for numerous
formance analysis, simulation and enhancement in
international conferences, including IEEE INFOCOM, IEEE ICC, IEEE
wireless communications and networking.
GLOBECOM, and IEEE WCNC. He is an IEEE Vehicular Technology
Society (VTS) Distinguished Lecturer. He is also a Senior Member of
the IEEE ComSoc, IEEE CS, IEEE PES, and IEEE VTS. He is a
Fellow of the IET.