This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
1

Blockchain-based Decentralized Trust Management

in Vehicular Networks
Zhe Yang, Kan Yang, Member, IEEE, Lei Lei, Member, IEEE, Kan Zheng, Senior Member, IEEE, and Victor C.
M. Leung, Fellow, IEEE

Abstract—Vehicular networks enable vehicles to generate and on purpose. For example, a malicious vehicle may broadcast
broadcast messages in order to improve traffic safety and a message claiming that the road is clear, while there is a
efficiency. However, due to the non-trusted environments, it is dif- traffic accident or congestion actually. These misbehaviors
ficult for vehicles to evaluate the credibilities of received messages.
In this paper, we propose a decentralized trust management can greatly endanger the traffic safety or efficiency of the
system in vehicular networks based on blockchain techniques. transportation system. Therefore, how to effectively evaluate
In this system, vehicles can validate the received messages from the trustworthiness of vehicles is an important problem in
neighboring vehicles using Bayesian Inference Model. Based on vehicular networks.
the validation result, the vehicle will generate a rating for each Trust management system enables vehicles to decide
message source vehicle. With the ratings uploaded from vehicles,
Roadside Units (RSUs) calculate the trust value offsets of involved whether the received message is trustworthy or not, and also
vehicles and pack these data into a “block”. Then, each RSU provides network operators the basis of rewards or punish-
will try to add their “blocks” to the trust blockchain which is ments on specific vehicles [8] [9]. Usually, the trust value of
maintained by all the RSUs. By employing the joint Proof-of- a certain vehicle can be calculated using ratings on its past
Work and Proof-of-Stake consensus mechanism, the more total behaviors, which are generated by relevant nodes. Existing
value of offsets (stake) is in the block, the easier RSU can find the
nonce for the hash function (proof-of-work). In this way, all RSUs trust management systems can be classified into two groups,
collaboratively maintain an updated, reliable, and consistent trust i.e., centralized and decentralized. In centralized trust manage-
blockchain. Simulation results reveal that the proposed system is ment systems [10] [11], all ratings are stored and processed
effective and feasible in collecting, calculating, and storing trust in a central server, e.g., cloud server. As vehicles usually have
values in vehicular networks. to make decisions in a quite short delay, these centralized
Index Terms—Blockchain, trust management, vehicular net- systems cannot always satisfy the rigorous Quality-of-Service
works, data credibility. (QoS) requirements for vehicular networks. In decentralized
trust management systems [13] [14] [15], trust management
I. I NTRODUCTION tasks are conducted in vehicle itself or in the Roadside Unit
(RSU). Local management of trust values may reduce the
Recently, vehicles have been given increasing autonomy interactions with network infrastructures. However, due to the
with the help of various on-board sensing, computation, and different capacities and conditions to observe and assess the
communication devices [1] [2]. All infrastructures and smart target events, ratings generated by a single vehicle cannot
vehicles constitute the vehicular network, which has become be always reliable. Moreover, the high variability of network
an important scenario of the fifth generation (5G) mobile topology makes it a big challenge to timely evaluate all
networks [3] [4] [5]. Vehicular networks provide a platform for vehicles it encounters. Some studies also utilize the RSU for
vehicles to share road-related messages with their neighbors, trust management [15]. However, RSUs are usually distributed
e.g., road conditions, traffic congestions, etc. These messages outside and are vulnerable to malfunctions and intrusions,
help vehicles timely be aware of traffic situations and hence which cannot provide reliable and consistent trust service for
improve the transportation safety and efficiency [6]. the whole vehicular network. Therefore, how to effectively
However, due to the high mobility and variability of ve- conduct trust management in vehicular networks is still a
hicular networks, neighboring vehicles are usually strangers problem remained to be solved urgently.
and cannot fully trust with each other. This problem becomes With this in mind, blockchain is considered as a feasible tool
more serious when there are malicious vehicles existing in the to cope with the problems above. Blockchain is initially known
network. These attackers may disseminate incredible messages as one of the disruptive technologies in financial industry,
which enables distributed nodes to trade with each other
Zhe Yang and Kan Zheng are with the Intelligent Computing and Commu-
nication (IC2 ) Lab, Key Laboratory of Universal Wireless Communications, and maintain a consistent and tamper-proof ledger without a
Ministry of Education, Beijing University of Posts and Telecommunications centralized bank [17] [24]. Besides, due to its high security
(BUPT), Beijing, 100876, China (E-mail: [email protected]). and reliability, blockchain has been widely studied and applied
Kan Yang is with the Department of Computer Science, University of
Memphis, Memphis, TN 38152, USA. in non-financial scenarios, e.g., content delivery [26], key
Lei Lei is with the School of Science and Engineering, James Cook management [18], decentralized storage [19] [20], etc. Based
University, Cairns, QLD, 4878, Australia. on the decentralization nature of blockchain, trust manage-
Victor C. M. Leung is with the Department of Electrical and Computer
Engineering, The University of British Columbia, Vancouver, BC V6T 1Z4, ment can be conducted among distributed RSUs, which can
Canada. effectively avoid the problems of centralization. Moreover,

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
2

blockchain enables RSUs to work together and maintain a When a malicious node deliberately drops relaying packets, it
consistent database. Even though a small portion of RSUs will be reported by the packet receivers and finally be evicted
are compromised by attackers, the block generation speed of from the network. In addition, a reputation system for reliable
attackers is much slower than that of benign RSUs. Therefore, cooperative downloading is designed in [11], which enables
the proposed system can effectively hold the trust management vehicles to securely download and forward packets for others.
tasks in vehicular networks, which enables vehicles to evaluate When a proxy vehicle honestly finishes the task, it will obtain
the trustworthiness of neighbors and assess the credibilities a virtual check from the packet receiver. Using this check, the
of received messages. The contributions of this paper can be reputation system is able to encourage cooperation and punish
summarized as follows: malicious vehicles.
1) We proposed a new decentralized trust management All these schemes utilize a fully-trusted central server for
scheme for vehicular networks based on the blockchain tech- trust management. However, with the rapid development of
nology, which not only enables all the RSUs to participate in intelligent transportation systems, it is not practical to cope
updating the trust values in a decentralized manner but also with large numbers of vehicles using a centralized node. Too
provides all RSUs the trust information of all the vehicles in many requests will probably bring about high latency or even
the vehicular networks. blocking, which may greatly decrease the QoS for users.
2) We proposed a joint Proof-of-Work and Proof-of-Stake Moreover, the single point of failure is also a big challenge
consensus mechanism which enables all the RSUs to compete for centralized networks.
to update the trust, i.e., add a trust block. This captures the
factor that the block containing the maximum total values of B. Decentralized trust management
trust offset needs to be added first, as it affects the whole trust
In order to cope with the problems of centralization men-
database significantly.
tioned above, decentralized systems are introduced for trust
3) We conduct the simulation to show that our proposed
management. In [12], the authors proposed a data-centric trust
blockchain-based decentralized trust management system is
management scheme in ad hoc networks. Once receiving data
efficient in practical vehicular networks.
from others, the node will firstly calculate the trust value
The remainder of this paper is organized as follows: Section
for each piece of data. Then, all these values are aggregated
II reviews the related works. The system model, adversary
using specific algorithms. If the aggregated value exceeds a
model and design goals are introduced in Section III. Section
threshold, the receiver will trust the content of the data. In
IV briefly describes the blockchain and the advantages of
[13], the authors also try to tackle with the similar problem.
applying blockchain in constructing trust management system.
This work firstly analyzes the key words of messages and
In Section V, we introduce our blockchain-based decentralized
divides all messages into groups. Then, the trust values of
trust management system in detail. Security analysis, perfor-
all messages are calculated based on the content similarity,
mance evaluation, and further discussion about the proposed
content conflict and routing path similarity. Besides, Li et al.
system are given in Section VI, VII, and VIII, respectively.
presented a mechanism which jointly considers privacy and
Finally, Section IX concludes this paper.
reputation issues [14]. Tasks of behavior evaluation, reputation
aggregation and manifestation are collectively performed on
II. R ELATED WORK each vehicle. The partially blind signature is adopted to
A. Centralized trust management preserve the vehicle privacy. All these schemes need individual
vehicle to manage the trust values by itself, which may be
Centralized trust management in vehicular networks has
inaccurate due to the limited observation conditions or possible
been widely studied recently. In these works [7] [10] [11], a
malfunctions.
central server is utilized to collect, calculate, and store the trust
In [15] and [16], RSU is employed for trust management,
values of all vehicles. The central server is usually assumed
where vehicles generate ratings for others and upload these
to be a fully-trusted entity, which cannot be compromised by
ratings into the nearby RSU. They can also send requests
attackers.
to RSU to query the trust values of neighboring vehicles.
In [7], a reputation-based announcement scheme is pre-
However, the trust information stored in the distributed RSU
sented in vehicular networks. In this paper, vehicles sense the
may not be complete and consistent. Therefore, it is critical
traffic-related events and publish announcements to neighbors.
but also challenging to develop a decentralized reliable and
The receivers need to evaluate the credibilities of messages
consistent trust management system in vehicular networks.
and generate feedback reports. All feedbacks are collected by
a centralized reputation server. Based on these data, the server
is able to update the reputation values and issue certificates C. Blockchain-based decentralized data management
for all vehicles in the network. Moreover, M. E. Mahmoud Blockchain has been given an increasing attention for decen-
et al. proposed a stimulation and punishment mechanism for tralized data management. For example, C. Cai et al. presented
mobile nodes [10]. In this mechanism, a “micropayment” is a encrypted decentralized storage system based on blockchain
used to stimulate nodes to relay packets from others. The techniques to handle the fraudulent behaviors of clients [19].
honest nodes can earn certain amounts of credits, which can In this system, important information about the stored files,
be spent when they have relay requirements. A reputation e.g., the digests, tokens, and metadata of integrity checking,
system is also designed to cope with the packet droppers. is stored in the blockchain, which provides fair judgments for

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
3

storage and search services. Moreover, in [20], the authors 2) Vehicles: Vehicles are equipped with on-board sensors,
designed a blockchain-based distributed storage and keyword computers, and communication devices, which are used for
search platform. In this paper, the blockchain is used to store data gathering, processing, and sharing.
the public keys of well-behaved nodes, which are confirmed by With the help of on-board devices, vehicles can automati-
the majority of the network. Therefore, due to the features of cally detect traffic-related events and send warning messages
decentralization, consistence, and tamper-proofing, blockchain to others using vehicle-to-vehicle communications standards,
can be a promising technique to help cope with the trust e.g., the Long Term Evolution Vehicle-to-vehicle (LTE-V2V)
management problems in vehicular networks. or dedicated short-range communications (DSRC) [22]. How-
ever, not all messages are useful. For example, if a vehicle has
already passed the location of a specific event, reports about
III. P ROBLEM DEFINITION this event will no longer be valuable for it. Therefore, each
vehicle needs to maintain a reference set, whose members are
A. System model
of high relevance to the traffic safety of the target vehicle.
In this system, the reference set of a certain vehicle is
composed of neighboring vehicles traveling in front within
certain distance, as illustrated in Fig. 1. Using messages sent
RSU from the reference set, vehicles can timely be aware of the
Trust traffic conditions and respond to possible events.
Values However, due to possible malfunctions or misbehaviors,
messages from the reference set are not always trustworthy.
Ratings Receivers need to aggregate all messages about a certain event
and figure out the credible ones. Specific models are used for
message aggregation, e.g., the majority rule. Then, the receiver
can generate ratings for messages based on the credibilities and
Vehicles Message 1: rating then upload these ratings into the RSU.
Message 2: rating
Event Messages Ă
Message n: rating B. Adversary model
Both vehicles and RSUs are vulnerable to probable at-
Messages tackers, which can severely interfere the operation of trust
management systems and thus endanger the traffic safety of
vehicles. Two adversaries are considered in this paper, i.e., the
Benign vehicle Malicious vehicle Reference set malicious vehicle and compromised RSU.
1) Malicious vehicle: Sometimes a number of malicious
Fig. 1. System model of decentralized trust management in vehicular vehicles may exist in vehicular networks. They usually have
networks. specific motivations and try to interfere with the normal
operation of the network. These misbehaviors can severely
As illustrated in Fig. 1, a decentralized trust management endanger the traffic safety or efficiency of benign vehicles.
system in vehicular networks mainly includes several con- In this paper, malicious vehicles mainly have two types of
nected RSUs and vehicles on the road. behaviors, i.e.,
1) RSU: Due to its resources and capabilities, RSU is a) Message spoofing attack: Attackers may deliberately
responsible for the major tasks, i.e., rating collection and trust broadcast fake messages in order to degrade the traffic safety
value management. or efficiency. For example, a malicious vehicle may detect a
• Rating collection: Ratings are generated by message traffic accident on the road, but broadcast a message claiming
receivers in order to evaluate the credibilities of messages. “The road is clear!” to nearby vehicles.
However, they cannot be stored and managed locally in b) Bad mouthing and ballot stuffing attack: In this
the long term, due to the fast changing traffic environ- system, the bad mouthing (ballot stuffing) attack means that
ments and limited capacity of on-board devices. There- vehicles generate and upload unfair negative (positive) ratings
fore, vehicles need to periodically upload their ratings on credible (incredible) messages. For example, after receiving
into the nearby RSUs, which serve as the collectors and a credible message, a malicious vehicle may deliberately
hosts for these data. generate a negative rating (e.g., -1) on this message and upload
• Trust value management: We assume that only RSUs this rating to the RSU.
are able to calculate the trust value for a certain ve- 2) Compromised RSU: RSUs are distributed along the road
hicle based on the collected ratings. Trust value is the and sometimes lack protection from the network operators.
aggregated opinion of a vehicle, which represents the Therefore, these entities are assumed to be semi-trusted, which
historical credibilities of messages sent by it. Once being may be compromised by the attackers. Once intruding into
calculated, trust values can be queried by other vehicles an RSU, the attacker is able to add, delete, and tamper the
if needed. data stored in it. However, the large-scale intrusion attacks

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
4

are highly unlikely due to the limited capacity of attackers. security and reliability, blockchain has been widely studied
Moreover, due to the periodical security check from the and applied recently.
network operators, the compromised RSU cannot be controlled
by attackers for a long time. Based on these facts, it is assumed Block
that attackers can only compromise a small portion of RSUs
Prev hash Nonce Prev hash Nonce Prev hash Nonce
during a short period of time.
TX list TX list TX list
C. Design goals
Blockchain
This paper focuses on assessing, recording, and dissemi-
nating the vehicle trustworthiness in vehicular networks. Thus Fig. 2. Typical structure of blockchain.
the main objectives are that all behaviors of vehicles can be
justly evaluated, and all vehicles are able to access the reliable As shown in Fig. 2, a blockchain is an ordered list of
trust values of neighbors if needed. The design of a trust blocks, where each block stores certain numbers of historical
management system should achieve the following goals: transactions (TXs). These TXs are generated by traders and
1) Decentralization: With the rapid increase of smart ve- are broadcasted through the entire network. Each block is
hicles, the centralized trust management schemes may be “chained” to the previous one, by keeping a digest (i.e.,
unpractical. Therefore, the trust management system needs the hash value) of the previous block. Thus, any change
to take full advantages of distributed nodes, i.e., RSUs and on a specific block would inevitably destroy the integrity
vehicles. Trust values are calculated by ratings uploaded from of the chain. In addition, a nonce is usually included in
message receivers and are stored in the RSU, which can ensure each block, which is the answer of a mathematical problem.
the reliability and scalability of the system. The node who solves the problem is elected as a temporary
2) Tamper-proofing: RSUs are usually distributed outside center node, i.e., miner and broadcasts its block to others.
and vulnerable to be compromised by attackers. If the data Several miner election schemes have been proposed in recent
stored in the compromised RSU are tampered, the reliability blockchain-based systems, e.g., the proof-of-work, proof-of-
of trust management will be affected. However, the large scale stake, and proof-of-capacity [21], under which nodes with
compromising of RSUs is unlikely due to the limited capacity higher computing power, capital, and storage capacity are
of attackers. Therefore, the trust management system should more likely to win the election. Consequently, blockchain has
be resistant to a small portion of compromised RSUs. provided a feasible way to keep data security and consistency
3) Consistency: Due to the high mobility feature, vehicles in decentralized networks.
usually need to travel across many RSUs. Under this scenario,
how to exchange trust data among RSUs and maintain a con- B. Design overview
sistent database becomes a challenging issue for decentralized
In this paper, we explore how the blockchain can be used
trust management in vehicular networks.
to keep the trust values of vehicles. Firstly, all vehicles assess
4) Timeliness: Trust value indicates the overall evaluation
the credibilities of received messages and then generate ratings
of a certain vehicle based on its historical behaviors. This
for them. The positive rating (e.g., +1) indicates a credible
value may change over time according to the credibilities of
message, while the negative rating (e.g., -1) represents an
messages sent by this vehicle recently. Hence, the trust values
incredible one. These ratings are uploaded into the RSU,
stored in RSUs need to be updated in time.
which plays a dominant role in trust management. Compared
5) Availability: Trust values stored in the RSU need to be
with the vehicles, RSUs usually have more stable network
available for vehicles if they want to know the trustworthiness
topology, more reliable communications channels (i.e., wired
of any neighboring vehicles. Thus, an application interface
link), and more powerful computing and storage capacities.
(API) is required for vehicles to send query requests and
These advantages make RSU a good choice for decentralized
receive corresponding trust values from RSUs.
trust management in vehicular networks.
Based on the ratings uploaded from vehicles, RSU needs to
IV. M ETHODOLOGY
calculate the offset of trust values for every involved vehicle
A. Blockchain using specific methods. In this system, trust value offset is
Blockchain is usually regarded as a series of techniques between −1 and +1, which is positively correlated with the
utilized in decentralized networks so as to maintain a consis- ratio of positive ratings. The sum of all offsets is the trust
tent database among all members. It is firstly proposed by value of this vehicle. Similar with the transactions mentioned
Satoshi Nakamoto in order to abstract the core techniques above, several offsets are packed into a candidate block by the
of the well-known digital currency, i.e., the Bitcoin [21]. RSU. Then the RSU tries to be elected as the miner and adds
Different from the centralized network structure, there are no this block into the blockchain.
fixed center nodes in blockchain-based networks. All members Miner election method is one of the most important parts
in the network have relatively equal positions and keep the in blockchain-based systems. In vehicular networks, the block
same copy of blockchain. Therefore, no one can change the containing the maximum total absolute values of trust offset
data recorded in the blockchain unless he has obtained strong needs to be added first, as it will affect the whole trust database
enough capacity to confuse the crowds. Due to the high significantly. In our design, we develop a joint proof-of-work

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
5

and proof-of-stake election scheme, which takes the sum of V. D ETAILED DESIGN OF THE PROPOSED SYSTEM
absolute values of offsets in the candidate block as the stake.
RSUs with larger stakes are more likely to be elected as the
miner. Therefore, the larger variations on trust values will be 4. Distributed consensus
more quickly to be reflected in the blockchain, which ensures
the timely update of the recorded data. By accumulating all Trust Blockchain maintained by RSUs
offsets of each vehicle stored in the blockchain, RSUs are able Hash Nonce Hash Nonce
to obtain the real-time trust values. Offset list Offset list
By contrast, the traditional proof-of-stake scheme lacks
randomization, which may cause that RSUs with large stakes
continuously win the election and restrict the trust value
Hash Nonce
updating of RSUs with low stakes. If the RSU with large
Offset list
stakes is compromised by the adversary, this adversary may 3. Miner election and VIN=1, offset= 0.8
compromise the entire trust system. In addition, the traditional block generation VIN=5, offset= -0.6
proof-of-work scheme only takes computing capacity as the ...
Trust value
basis of miner election, which cannot distinguish high stake Offset block
Offsets
RSUs and low stake ones. Therefore, large variations on trust
values cannot be fast reflected in the blockchain. 2. Calculation of trust
value offsets
C. Advantages of applying blockchain in trust management
RSU
The following advantages make blockchain a promising Ratings
solution for trust management in vehicular networks, which
is able to achieve the design goals listed in Section III-C.
1) Decentralization: Compared with the centralized net- 1. Ratings generation
Messages
work structure, blockchain enables distributed nodes to coop- and uploading
erate with each other and maintain a reliable database. Every
node in the network has a copy of the blockchain and is Vehicle
able to conduct specific operations on it, e.g., calculation and
data query. Hence, the system scalability is improved to a Fig. 3. System design of blockchain-based decentralized trust management.
large extent. Furthermore, the single point of failure can be
effectively mitigated once using blockchain, which is always
a challenging issue in centralized networks.
A. Main procedures
2) Tamper-proofing: Due to the chain structure, malicious
users attempting to modify one block stored in the blockchain, As illustrated in Fig. 3, main procedures of the blockchain-
have to rebuild the whole chain after it, which exponentially based decentralized trust management can be divided into the
increases the time cost of tampering. Moreover, all RSUs following steps, i.e., 1) Rating generation and uploading; 2)
compete to be the miner in order to add their blocks. Under Calculation of trust value offsets; 3) Miner election and block
the assumption that most of RSUs are benign, the block generation; 4) Distributed consensus.
generation speed of these RSUs is much larger than that of Step 1: Rating generation and uploading: This procedure
compromised ones. Therefore, the influence of small numbers is conducted on vehicles, more specifically, the message
of compromised RSUs on the blockchain is very limited. receivers. Due to the presence of possible malfunctions or
3) Consistency: Blockchain enables distributed RSUs to misbehaviors, messages from the reference set Ref are not al-
main a consistent trust database. All RSUs contribute to the ways credible. Therefore, specific rules are needed for message
blockchain and extract data from it. Therefore, they can return receivers to assess the credibilities of messages and generate
the same results once being queried by vehicles. ratings for them. Firstly, the receiver divides all messages into
4) Timeliness: Based on the idea of joint proof-of-work groups {M1 , M2 , ...Mj , ...}, where Mj represents the message
and proof-of-stake, RSUs with larger absolute offsets are more group reporting event ej , e.g., “There is a traffic accident at
easier to win the miner election and then add their blocks into road segment A!”. However, not all messages in a same group
the blockchain. This scheme makes larger changes on trust are of equal credibility. Messages sent by vehicles near the
values to be faster reflected and thus ensures the timeliness of event location are usually more trustworthy than these from
data stored in the blockchain. remote vehicles. Therefore, the credibility of a certain message
5) Availability: Using the data stored in the blockchain, is defined as follows,
RSUs are able to accumulate all offsets about a certain vehicle cjk = b + e−γ·dk ,
j
(1)
and get its current trust value. Once a vehicle wants to know
the trustworthiness of its neighbors, it can send request to the where cjk is the credibility of message in group Mj sent by
nearby RSU. After checking the identity of the requester, the vehicle k. djk is the distance between the message sender
RSU sends a response containing the trust value of the target and the event location. b and γ are two preset parameters,
vehicle. which control the lower bound and change rate of message

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
6

credibility, respectively. In addition, cjk = 0 if vehicle k does is usually used in blockchain-based systems, e.g., the Bitcoin.
not report this event. Using Eq. 1, the receiver can obtain In these systems, all nodes continuously change the nonce
a credibility set C j for event ej , where C j = {cj1 , cj2 , ...}. and then calculate the hash values of the block including the
Based on the credibility set C, the receiver is able to calculate nonce. The one getting the hash value lower than a threshold
the aggregated credibility of event e using Bayesian Inference is elected as the miner and is able to publish its block. All
[12]: nodes have the same threshold, which makes nodes with more
N powerful computation capacity easier to get the right nonce
Q
P (e) · P (ck /e) and win the election. Based on proof-of-work, the proof-of-
k=1 stake is proposed which makes different nodes have different
P (e/C) = N N
, (2)
P (e) ·
Q
P (ck /e) + P (ē) ·
Q
P (ck /ē) hash thresholds, and thus different generation speeds of blocks.
k=1 k=1 In this system, a joint proof-of-work and proof-of-stake
miner election method is designed, which takes the sum of
where ē is the complementary event of e. P (ck /e) = ck .
absolute offsets as stakes and the difficulty to complete the
P (ck /ē) = 1 − ck . P (e) is the prior probability of event e.
proof-of-work depends on the stake. RSUs with more stakes
P (e/C) ∈ [0, 1]. Once P (e/C) exceeds a preset threshold
can find the nonce and win the election easier (i.e., more
T hr, the receiver regards this event as true and generates
quickly to publish their blocks), which ensures the timely
positive ratings (i.e., +1) on messages correctly reporting this
update of data stored in the blockchain. The proposed miner
event. Otherwise, it will generate negative ratings (i.e., -1) on
election method is:
them.
Finally, vehicles periodically upload the ratings to the Hash(IDRSU , time, PreHash, nonce) 6 Si , (5)
RSU nearby for further use. The format of a rating is
(V INi , V INj , mk , rating), where V INi and V INj are the where Si is the hash threshold of RSU i. All RSUs con-
vehicle identity numbers of message receiver and sender, tinuously change the nonce and calculate the hash values
respectively; mk is the identifier of the message; rating according to Eq. 5. The one getting the nonce satisfying
is either −1 (for incredible messages) or +1 (for credible the above condition is elected as the miner. Si is positively
messages). correlated with Fi , which is defined as the sum of absolute
Step 2: Calculation of trust value offsets: The RSU may values of trust value offsets:
 
get conflicting ratings about a specific message, e.g., 7 positive X  j 
ratings and 3 negative ratings. The former is the majority group Fi = min  ok , Fmax  , (6)
and the latter is the minority group. In the proposed system, ojk ∈Oi
weighted aggregation is used on these ratings to obtain the
where Oi is the current offset set of RSU i. Therefore, the
offset of trust value. The offset is between −1 and +1, which
RSU with larger Fi is more likely to win the election and then
is positively correlated with the ratio of positive ratings on this
publish its block. In this way, large variation of trust values
message. The calculation of trust value offset is shown in Eq.
can be timely updated in the blockchain. Fmax is the upper
3.
θ1 · m − θ2 · n bound of Fi , which is used to avoid the situation that the RSU
ojk = , (3) with too large Fi continuously wins the election. Therefore,
m+n
the relative fairness is achieved among RSUs. Once the RSU
where ojk is the trust value offset of vehicle k based on message successfully adds the offset block into the blockchain, it will
j and ojk ∈ [−1, 1]. m and n are the number of positive clear the elements in Oi .
(+1) and negative (-1) ratings, whose weights are θ1 and θ2 , Construction of Si : Si is series of binary bits which starts
respectively. θ1 and θ2 are calculated using Eq. 4. from several continuous zeros. Si can greatly influence the
F (m) F (n) generation speed of offset blocks for each RSU. In the
θ1 = , θ2 = , (4) proposed system, the relationship between Si and Fi is defined
F (m) + F (n) F (m) + F (n)
as follow,
where F (.) controls the sensitivity to the minority group of  
ratings. For example, the aggregated offset with F (x) = x2 is Nz = int e−(η·Fi +µ) ,
less sensitive to the minority group of ratings compared with (7)
Si = 2Nm −Nz − 1,
F (x) = x. Under the assumption that attackers cannot control
a large portion of vehicles, the majority groups of ratings where int(.) returns the integral part of the value; Nz is the
are more likely to be fair ratings. Therefore, the proposed number of continuous zeros at the top of Si ; Nm is the bits
weighted aggregation is able to improve the reliability of trust of the hash value depending on the hash algorithm (e.g., 160
value offsets. Finally, the RSU puts all these offsets into the for SHA-1, 256 for SHA-256, etc.), as illustrated in Fig. 4.
set O and tries to add it into the blockchain. The format of an offset block is shown in Fig. 5, which
Step 3: Miner election and block generation: Due to the contains two parts, i.e., the header and body. The header stores
decentralized network structure, there is no constant center 1) The basic information about the block, such as block ID,
node to manage the blockchain. Therefore, a miner is pe- RSU ID, and generation time; 2) The hash of the previous
riodically elected from all RSUs in order to generate new rating block, which is used to chain this block to the existing
offset blocks. Miner election method based on proof-of-work blockchain; 3) Information to prove the validity of this block,

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
7

Nz 2) Warning and revocation: For the vehicles whose trust

values are relatively low, RSUs are able to take certain
actions, e.g., warning and revocation. Two thresholds, i.e.,
Si : 0 0 0 0 0 1 1 1 1 1 1 1 1 1... Rwarn and Rmin , are defined to punish these vehicles, where
Nm
Rwarn > Rmin . Vehicles with trust values lower than Rwarn
will be added into the warning list, which is broadcasted by
Fig. 4. Relationships among Nz , Nm , and Si . RSUs. The warned vehicles need to sense the environments
and broadcast credible messages actively, in order to improve
their trust values in time. More seriously, vehicles with trust
i.e., the nonce and hash threshold. The body mainly contains values lower than Rmin will be added into the revocation list.
the list of trust value offsets. Once being revoked, vehicles cannot receive any services from
the vehicular network.
Header
Generation VI. S ECURITY ANALYSIS
Block ID RSU ID
time
A. Defense against malicious vehicle
Hash
Prev Hash Nonce a) Message spoofing attack: A malicious vehicle may
threshold
broadcast fake messages to neighbors, which can cause severe
Body traffic accidents or congestions. In the proposed system, a
Bayesian Inference-based rating generation scheme is used to
Offset list VIN = 1, offset = 0.8 thwart this attack. The message receiver can comprehensively
analyze messages broadcasted by different reference vehicles
VIN = 5, offset = -0.6
about this event and decide which message is trustworthy.
... As the number of attackers is limited, vehicles are usually
able to know the credibilities of received messages. Moreover,
with the help of the application interface, vehicles can easily
Fig. 5. Format of offset block.
query the trust value of a specific neighbor, which is also an
important factor for message credibility assessment.
Step 4: Distributed consensus: Once receiving a block from b) Bad mouthing and ballot stuffing attack: Malicious
the miner, the RSU needs to check the validity of the nonce vehicles may generate unfair ratings and upload them into the
and then add it to its blockchain. Sometimes the RSU may RSU. However, in the proposed system, messages are broad-
receive more than one blocks at the same time. Under this casted to all neighbors and each receiver can only generate one
circumstances, the blockchain starts to fork. A distributed binary rating for a specific event. Due to the limited number of
consensus scheme is used to cope with this issue. Each RSU attackers, the unfair ratings can hardly change the aggregated
chooses one fork and continues to add new blocks after it. As trust values of vehicles.
time goes by, the fork acknowledged by the largest number
of RSUs grows faster than others. Finally, the longest one
becomes the distributed consensus of the network, while the B. Defense against compromised RSU
other forks are discarded. In addition, each RSU needs to In the proposed system, it is assumed that a small portion of
collect blocks generated by themselves in the discarded forks RSUs may be compromised during a short time period. Data
and try to add them to the blockchain in the future. In this stored in the RSU may be added, deleted, or tampered by
way, all RSUs store the same version of blockchain, which attackers. However, based on blockchain techniques, all RSUs
ensures the consistency of the network. store the same version of the blockchain and continuously add
new blocks on the current chain. Once getting rid of the control
from attackers, the compromised RSUs can easily detect their
B. Application interface
differences from others. This is because any changes to the
1) Trust value query: Once a vehicle wants to know the locally stored data will inevitably change the hash value of the
trust value of another, it needs to send a query request last block, which is different from the previous hash stored in
(QueryReq, V INi , V INj ) to the RSU nearby. V INi and the new coming block generated by benign RSUs.
V INj are the vehicle identity numbers of the request sender Moreover, the compromised RSUs may also generate and
and the target vehicle, respectively. After receiving the query broadcast fake blocks. However, they still need to compete
request, the RSU firstly checks the identity of the request with others for adding blocks into the blockchain. Due to
sender. Then, based on the offsets stored in the blockchain, the limited number of attackers and the short compromising
the RSU can accumulate all the offsets together and obtain time periods, the number of fake blocks generated by the
the current trust value of vehicle j. Finally, the RSU sends compromised RSUs remains small. In addition, the upper
response (V INj , En(T rj )pki ) to the requester. T rj is the bound of the stakes in each block (i.e., Fmax in Eq. 6) prevents
trust value of the target vehicle, which is encrypted using the attackers from generating fake blocks with too large stakes,
public key of the requester. which can also reduce the impacts of compromised RSUs.

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
8

TABLE I 1.0
K EY PARAMETERS
P(e)=0.1
P(e)=0.5
Parameters Values 0.8

Vehicle number 50

Ratio of unfair ratings

Distance Uniform distribution 0.6
between vehicles between 5 and 100 meters
β 0.5
0.4
γ 0.014
T hr 0.5
Vehicle number in Uniform distribution 0.2

reference set between 0 and 10

Hash algarithm SHA-256 0.0
0 20 40 60 80 100
η 0.01
Percentage of false messages
µ -3
Message: 800 bytes; Fig. 6. Percentage of unfair ratings versus false messages.
Packet size
Rating block: 8000 bytes

to unfair ratings when a small proportion of negative ratings

VII. P ERFORMANCE EVALUATION generated by the attackers are included in the rating set.
In order to validate the effectiveness and feasibility of the
1.0
proposed system, performance evaluations are conducted using F(x)=x
2
vehicular and blockchain simulation platform based on Matlab. F(x)=x
3
F(x)=x
The configurations of key parameters are listed in Table I. This 0.5 F(x)=e
x

section is divided into three parts. The first part studies the
Trust value offsets

calculation of ratings and trust value offsets. The generation

time intervals of blocks with the variation of absolute offsets 0.0
are provided in the second part. Finally, the third part analyzes
the communication overheads of this system.
-0.5

A. Calculation of trust value offsets

This part mainly shows the procedures from messages to -1.0
0 20 40 60 80 100
ratings, and from ratings to trust value offsets. Fig. 6 plots
Percentage of negative ratings
the relationship between unfair ratings and false messages.
In this test, attackers deliberately send fake messages to the Fig. 7. Trust value offset versus percentage of negative ratings.
receivers. Once being puzzled by the attackers, benign vehicles
may generate unfair ratings on the received messages. It is
evident that the percentage of unfair ratings is very low when
there are small numbers of false messages. This is because the B. Block generation
Bayesian-based model is able to find the truth using the reports After obtaining the trust value offsets of vehicles, RSUs
from the benign majority. However, the percentage of unfair try to become the miner and publish their blocks. The block
ratings grows gradually with the increase of false messages and generation time T of each RSU is mainly influenced by two
finally reaches a high value near 1. This is because the false parameters, i.e., the sum of absolute offsets Fi and the hash
messages gradually become the majority and severely mislead rate M . As shown in Fig. 8, T drops gradually with the
the model. In addition, the line with more prior knowledge increase of Fi . This is because a larger Fi indicates a larger
about the event probability (i.e., P (e) = 0.1) outperforms the hash threshold Si , which makes it easier to get the right nonce.
one with neutral prior probability (i.e., P (e) = 0.5), due to Moreover, T also drops with the increase of M . M represents
the effect of prior knowledge on decision making. the number of hash operations that an RSU is able to conduct
The relationship between trust value offsets and the per- per second, which is related to the computation capacity of
centage of negative ratings is shown in Fig. 7. Four functions this RSU. A larger M enables RSUs to try more nonces within
are tested for the calculation of the offsets based on Eq. 3 specific time period, and thus obtain the right nonce faster.
and Eq. 4. The offsets decrease gradually with the increase In comparison, the block generation time based on the
of negative ratings, whose lower and upper bounds are −1 proof-of-work (PoW) scheme is also evaluated. In this scheme,
and +1, respectively. In addition, different F (x) may have all nodes have the same hash threshold and the block genera-
different effects on offsets. For example, the blue line is above tion time only depends on the hash rate of each node. However,
the black line when the negative ratings are less than 50%, as RSUs usually have similar hash rate, the opportunities
which means that the offset with F (x) = x3 is less vulnerable of block generation among all RSUs are relatively equal.

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
9

Therefore, in this evaluation, it is assumed that Nz = 24 and 240

Message
M = 500 for all RSUs. From Fig. 8, it can be clearly seen Rating
that the block generation time of PoW does not change with 200

the stake, i.e., the Fi . Compared with the PoW, the proposed

Transmission latency (ms)

scheme is able to reflect the differences between high stake 160

RSUs and low stake RSUs, and thus faster to update the trust
values with larger variations. 120

10 80

M=100
M=500 40
8 M=2000
PoW
0
1 2 3 4 5 6 7 8
6
Arrival rate
log10(T)

4
Fig. 9. Transmission latency of messages and ratings.

2 where Rk is the trust value of the message sender; Ref is the

reference set of the receiver; α1 and α2 are two parameters to
0
control the weight of each part.
0 10 20 30 40 50
Sum of absolute offsets
B. Privacy issues
Fig. 8. Generation time of offset blocks.
Privacy is usually regarded as one of the most significant
issues in vehicular networks [25]. In the proposed system,
C. Communications overhead the vehicle identity number is used as the identifier of each
vehicle, which has the potential of privacy leak. For example,
Two kinds of data are transmitted using the wireless using the vehicle identity number, attackers may find out
channels in vehicular networks, i.e., messages and ratings. the real identity of the vehicle owner. A possible solution
Messages are triggered by certain road-related events and for this problem is using the public key as the identifier of
broadcasted by vehicles. The packet size of message is set each vehicle, which is just a meaningless string. Furthermore,
to be 800 bytes, which corresponds with the event-triggered vehicles can periodically change the public keys through
data defined in [23]. Ratings are non-safety data which are interacting with the RSUs. However, these tasks inevitably
generated by message receivers. A vehicle can accumulate increase the communication and computation overheads of the
several ratings within a certain time period, pack them together network. Therefore, how to jointly assure the privacy preser-
into a data packet, and upload them to the nearby RSU. vation and efficient trust management is still an interesting but
Therefore, the size of rating packet is usually larger than that challenging problem in vehicular networks.
of message. As shown in Fig. 9, it is evident that with the
increase of data arrival rate, the transmission latency of both
messages and ratings are increased. Moreover, the latency of IX. C ONCLUSION
rating packet is larger than that of the message, due to the In this paper, we proposed a blockchain-based decentralized
larger packet size. trust management system in vehicular networks. With the aid
of this system, vehicles are able to query the trust values
VIII. D ISCUSSION of neighbors and then assess the credibilities of received
A. Trust-based data credibility assessment messages. Trust values are aggregated in the RSU based on
ratings generated by messages receivers. Using blockchain
In the proposed system, the Bayesian Inference is utilized
techniques, all RSUs work together to maintain a reliable and
to generate message ratings. During the rating generation,
consistent database. A number of simulations are carried out
the distance between message sender and event location is
in order to evaluate the performance of the entire system.
considered as the indicator of message credibility. Apart from
Simulation results demonstrate that the proposed system is
this factor, the current trust value of the message sender is
effective and feasible for decentralized trust management.
also very important for the assessment of message credibility.
Further studies are still needed in the future. For example, how
Through querying the trust values from the RSU, the message
to jointly assure the trust management and privacy preservation
receiver is able to aggregate this information with the distance
is an open problem which needs to be studied in detail. It is
factor and obtain a more reliable result, as illustrated in Eq.
believed that a reliable decentralized trust management system
8.
can greatly help vehicles evaluate the credibilities of neighbors
 
j j R k
ck = α1 · e−γ·dk + α2 ·  , (8) and establish a safe and efficient intelligent transportation
max (Rm ) network.
m∈Re f

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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2018.2836144, IEEE Internet of
Things Journal
10

X. ACKNOWLEDGMENT [21] F. Tschorsch and B. Scheuermann, “Bitcoin and beyond: A technical

survey on decentralized digital currencies,” IEEE Communications Sur-
This work was supported by the China Natural Science veys & Tutorials, vol. 18, no. 3, pp. 2084-2123, Thirdquarter 2016.
Funding (NSF) under Grant 61731004, and the BUPT Ex- [22] J.B. Kenney, “Dedicated short-range communications (DSRC) standards
cellent Ph.D. Students Foundation under Grant CX2016208. in the United States,” Proceedings of the IEEE, vol. 99, no. 7, pp. 1162-
1182, July 2011.
[23] 3GPP TR 36.885 V1.0.0, “Study on LTE-based V2X services,” TSG
R EFERENCES RAN, 2016.
[24] Z. Yang, K. Zheng, K. Yang, and V. C. M. Leung, “A blockchain-based
[1] H. Zhou, B. Liu, T. Hao, H. Fen, L. Gui, Y. Li, Q. Yu, and X. reputation system for data credibility assessment in vehicular networks,”
Shen, “ChainCluster: Engineering a cooperative content distribution in Proc. IEEE 2017 Annual International Symposium on Personal,
framework for highways vehicular communications,” IEEE Transactions Indoor, and Mobile Radio Communication (PIMRC), Montreal, Canada,
on Intelligent Transportation Systems, vol. 15, no. 6, pp. 2644-2657, Oct. 2017, pp. 1-5.
Dec. 2014. [25] J. Petit, F. Schaub, M. Feiri, and F. Kargl, “Pseudonym schemes
[2] S. He, D. Shin, J. Zhang, J. Chen, and Y. Sun, “Full-view area in vehicular networks: a survey,” IEEE Communications Surveys &
coverage in camera sensor networks: Dimension reduction and near- Tutorials, vol. 17, no. 1, pp. 228-255, Firstquarter 2015.
optimal solutions,” IEEE Transactions on Vehicular Technology, vol. [26] N. Herbaut and N. Negru, “A model for collaborative blockchain-based
65, no. 9, pp. 7448-7461, Sep. 2016. video delivery relying on advanced network services chains,” IEEE
[3] K. Zheng, Q. Zheng, P. Chatzimisios, W. Xiang, and Y. Zhou, “Het- Communications Magazine, vol. 55, no. 9, pp. 70-76, Sep. 2017.
erogeneous vehicular networking: A survey on architecture, challenges,
and solutions,” IEEE Communications Surveys & Tutorials, vol. 17, no.
4, pp. 2377-2396, Fourthquarter 2015.
[4] A. Wasef, R. Lu, X. Lin, and X. Shen, “Complementing public key
infrastructure to secure vehicular ad hoc networks,” IEEE Wireless
Communications, vol. 17, no. 5, pp. 22-28, Oct. 2010.
[5] K. Zheng, F. Liu, L. Lei, C. Lin, and Y. Jiang, “Stochastic performance
analysis of a wireless finite-state Markov channel,” IEEE Transactions
on Wireless Communications, vol. 62, no. 4, pp. 1450-1458, Feb. 2013.
[6] K. Zhang, J. Ni, K. Yang, X. Liang, J. Ren, and S. Shen, “Security
and privacy in smart city applications: Challenges and solutions,” IEEE
Communications Magazine, vol. 55, no. 1, pp. 122-129, Jan. 2017.
[7] Q. Li, A. Malip, K. M. Martin, S. Ng, and J. Zhang, “A reputation-based
announcement scheme for VANETs,” IEEE Transactions on Vehicular
Technology, vol. 61, no. 9, pp. 4095-4108, Nov. 2012.
[8] T. Roosta, M. Meingast, and S. Sastry, “Distributed reputation system
for tracking applications in sensor networks,” in Proc. 3rd Annual
International Conference on Mobile and Ubiquitous Systems, San Jose,
USA, July 2006, pp. 17-21.
[9] S. Li and X. Wang, “Quickest attack detection in multi-agent reputation
systems,” IEEE Journal of Selected Topics in Signal Processing, vol. 8,
no. 4, pp. 653-666, Aug. 2014.
[10] M. Mahmoud and X. Shen, “An integrated stimulation and punishment
mechanism for thwarting packet dropping attack in multihop wireless
networks,” IEEE Transactions on Vehicular Technology, vol. 60, no. 8,
pp. 3947-3962, Oct. 2011.
[11] C. Lai, K. Zhang, N. Cheng, H. Li, and X. Shen, “SIRC: A secure
incentive scheme for reliable cooperative downloading in highway
VANETs,” IEEE Transactions on Intelligent Transportation Systems, vol.
18, no. 6, pp. 1559-1574, June 2017.
[12] M. Raya, P. Papadimitratos, V. D. Gligor, and J. P. Hubaux, “On data-
centric trust establishment in ephemeral ad hoc networks,” in Proc. IEEE
INFOCOM, Phoenix, USA, Apr. 2008, pp. 1238-1246.
[13] W. Li and H. Song, “Information-oriented trustworthiness evaluation
in vehicular ad-hoc networks,” in Proc. International Conference on
Network and System Security, Madrid, June 2013, pp. 94-108.
[14] Z. Li and C. T. Chigan, “On joint privacy and reputation assurance for
vehicular ad hoc networks,” IEEE Transactions on Mobile Computing,
vol. 13, no. 10, pp. 268-273, Oct. 2014.
[15] X. Huang, R. Yu, J. Kang, and Y. Zhang, “Distributed reputation
management for secure and efficient vehicular edge computing and
networks,” IEEE Access, vol. 5, pp. 25408-25420, Dec. 2017.
[16] X. Huang, R. Yu, J. Kang, and Y. Zhang, “A distributed reputa-
tion scheme for situation awareness in vehicular ad hoc networks
(VANETs),” in Proc. IEEE CogSIMA, San Diego, Mar. 2016, pp. 63-67.
[17] K. Christidis and M. Devetsikiotis, “Blockchains and smart contracts for
the Internet of Things,” IEEE Access, vol. 4, pp. 2292-2303, May 2016.
[18] A. Lei, H. Cruickshank, Y. Cao, P. Asuquo, C. P. A. Ogah, and Z.
Sun, “Blockchain-based dynamic key management for heterogeneous
intelligent transportation systems,” IEEE Internet of Things Journal, vol.
4, no. 6, pp. 1832-1843, Dec. 2017.
[19] C. Cai, X. Yuan, and C. Wang, “Towards trustworthy and private
keyword search in encrypted decentralized storage,” in Proc. IEEE 2017
International Conference on Communications (ICC), Paris, May 2017,
pp. 1-7.
[20] C. Cai, X. Yuan, and C. Wang, “Hardening distributed and encrypted
keyword search via blockchain,” in Proc. IEEE 2017 Symposium on
Privacy-Aware Computing (PAC), Washington, Aug. 2017, pp. 119-128.

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.