SPECIAL SECTION ON EMERGING TECHNOLOGIES FOR DEVICE TO DEVICE COMMUNICATIONS

Received July 31, 2018, accepted August 29, 2018, date of publication September 10, 2018, date of current version October 8, 2018.

Digital Object Identifier 10.1109/ACCESS.2018.2869575

D2D Data Privacy Protection Mechanism Basedon Reliability and Homomorphic EncryptionBIAO JIN 1,2, DONGSHUO JIANG1, JINBO XIONG 1, (Member, IEEE),LEI CHEN 3, (Member, IEEE), AND QI LI41College of Mathematics and Informatics, Fujian Normal University, Fuzhou 350108, China2College of Geographical Sciences, Fujian Normal University, Fuzhou 350007, China3College of Engineering and Computing, Georgia Southern University, Statesboro, GA 30458, USA4School of Computer Science, Nanjing University of Posts and Telecommunications, Nanjing 210023, China

Corresponding author: Jinbo Xiong (jbxiong@fjnu.edu.cn)

This work was supported by the National Natural Science Foundation of China under Grant 61872088, Grant 61872090, Grant 61872086,Grant 61502489, Grant 61402109, Grant 61502102, and Grant 61502103.

ABSTRACT Device-to-device (D2D) network utilizes various communication methods and regionalresource sharing mechanisms, which, despite the efficient and effective communications, may lead to avariety of security threats. Therefore, without the assistance from the base station, effective managingregional resources and protecting private data on mobile devices have become a major challenge in D2Dnetworks. In this paper, we propose a reliability-based central node election mechanism in a D2D network,where the attribute information is collected, normalized, and weight-summed to acquire the reliability ofeach mobile device. The central node can therefore be elected by sorting the reliability of all mobile devices.Furthermore, we propose a security protection mechanism of private data based on homomorphic encryptionin a D2D network, where homomorphic encryption is employed to implement secure data aggregate ofciphertext in the elected central node. Finally, theoretical analyses and simulation experiment verify thesuperior effectiveness and efficiency of the proposed schemes.

INDEX TERMS Device-to-device communication, 5G, privacy protection, homomorphic encryption.

I. INTRODUCTIONWith the continuous development and advancement ofmobilecommunication technologies, the number of mobile appli-cations and volume of mobile data have grown explosively.In order to meet the requirements of high bandwidth and lowlatency in mobile communications, the fifth-generation (5G)mobile communication network has developed rapidly inrecent years. Compared to its predecessor, the fourth gen-eration (4G) mobile communication network, 5G networksare characterized by its significantly faster transmission rates(with theoretical peak transmission rate up to tens of Gbps),shorter latency and lower power consumption [1]. Device-to-device communication (D2D), as a key technology of 5G,has broad prospects in the areas of local communications,emergency communications, and the enhanced Internet ofThings [1].

In D2D networks, data transmit and distribute directlybetween mobile devices, therefore significantly improvingnetwork throughput [2]. Given the fact that the device nodesare mostly mobile phones and tablet computers with very

limited computing power in a D2D network, it is a corerequirement to integrate network resources to improve thenetwork performance. At present, D2D networks have threemain operation modes [3], with the first one aiming to estab-lish a connection between the mobile devices guided by abase station which also performs resource allocation [3].In the second mode, where the mobile devices are outsidethe range of base station, mobile communication requiresdata routing through the mobile devices in the range andtransmits the data to the base station. The third mode hasa distinct difference from the other two modes in that thenetwork implements communication through data forwardingamong mobile devices in the area without a base station.The structure of the three modes is shown in Figure 1. In atraditional wireless network, all device nodes are connected tothe base station, which therefore usually function as the coreof the network. Since there is no base station serving as thecore in the third mode of a D2D network, a central node needsto be elected taking charge of performing network resourcemanagement for improved network performance. How to

511402169-3536 2018 IEEE. Translations and content mining are permitted for academic research only.

Personal use is also permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

VOLUME 6, 2018

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

FIGURE 1. Three operation modes of D2D network.

properly elect and assign the central node then becomes themain challenge in this mode. In addition, in this mode, allnodes are directly linked to other nodes with unknown reli-ability which inevitably leads to mobile data traveling in theD2D network possibly through unknown or un-trusted nodesand subsequently putting data privacy protection on risk.Currently, there are four main data protection methods forD2D network communications [4], including access control,data obfuscation, data anonymization, and data encryption.

Data obfuscation is to protect data privacy based on sum-marized information or provides false information to reducethe accuracy of data. For example, when publishing locationinformation ofmobile device nodes, only the state or provincewhere the mobile device node is located in would beadvertised. Data anonymization is implemented in a waythat the published private data contains a certain number ofpseudonyms, so that the mobile device nodes receiving suchdata are not capable of identifying the private data owner.These two protection methods often have an adverse effect ondata, so data encryption is most feasible for protecting privatedata in a D2D network. However, the traditional encryptionalgorithms have the challenge to data secure aggregationoperations that cannot be performed without decrypting theciphertext. Therefore, homomorphic encryption is introducedto perform data secure aggregation operation on the electedcentral node without decryption.

Combined with the central node election mechanism andhomomorphic encryption algorithm, this paper proposes aD2D data privacy protection mechanism based on reliabilityand homomorphic encryption. The main contributions of thispaper are as follows:

1) We propose a reliability-based central node electionmechanism in D2D communication networks, wherethe information about a mobile device node is inte-grated to calculate its reliability, and a device node is

elected as the central node of the D2D network basedon the sorting of the reliability among all nodes.

2) The Paillier homomorphic encryption system is used toimplement a private data protection model, where theintegration and allocation of computing resource anddata in a D2D network can also be achieved.

3) Simulating the real network environment ofD2D communication verifies the reliability and secu-rity of the proposed model, establishing a D2D dataprivacy protection mechanism based on reliability andhomomorphic encryption.

II. RELATED WORKThere are four main methods of privacy protection: accesscontrol, data obfuscation, data anonymization, and dataencryption, which have been widely used in wireless com-munications. The following section will discuss some exist-ing privacy protection methods that can be used in D2Dcommunication.

At present, there are mainly three models for accesscontrol, which are discretionary access control (DAC),mandatory access control (MAC), and role-based access con-trol (RBAC). The DAC means that the mobile device nodehas full control over the object it belongs to and the pro-gram it runs. For example, a mobile device node owns itslocation information. The mobile device node only allowsneighbor nodes to obtain this information, and the rest ofthe nodes have no right to reach it. The MAC refers to theadministrator to define the relevant device to stipulate whichdevice node can access the object. Because of its policy-based, any unauthorized operation cannot be performed, andthis access control model can set the different security levels.The RBAC means that the administrator can define a seriesof roles and assign suitable mobile device nodes to them.Therefore, the mobile device nodes with corresponding rolescan access the classified objects. Behrooz and Devlic [5]proposed a model for controlling context access throughDAC systems, which uses context awareness and mobiledevice relationships to solve complex situations in com-munications, enabling mobile users to control who canaccess their context by specifying their context-aware privacyrules. Another SensorSafe access control model proposedby Chakraborty et al. [6] is designed to protect personalsensor data. The level of data access is determined by theadministrator based on the credibility between users.

Data obfuscation is mainly to protect the private data bysummarizing information and providing false information.Wishart obfuscates the data based on the content of the infor-mation and the surrounding environment, the users can setthe obfuscation level applied to the data based on the currentsituation [7]. They can also control who is the informationpublished to,and set the policy on publishing the informationand the level of detail of the published information. Anothermechanism is proposed by Franz to divide the devices in thenetwork into different groups. The members of the groupnegotiate the private data policy to determine which private

VOLUME 6, 2018 51141

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

data can be published and the degree of obfuscation of theprivate data [8]. This mechanism is mainly applied to socialnetworks.

In the aspect of data anonymization, since theD2D communication needs to be authenticated when itis established, it is necessary to consider the specialty ofusing data anonymization in the D2D network to protectdata privacy. In the D2D network, it is necessary to com-bine anonymous technology and reliability mechanism toestablish trust relationships and trust connections betweendevice nodes anonymously [9]. In this way, even if thecontent is shared with strange device nodes, their identitieswill not be revealed, and more sensitive information can beshared. Christin et al. [10] proposed a method to protectthe privacy of device sensor information in the network bygiving the pseudonyms credibility. Any operation of devicenode using a pseudonym will affect its credibility.If thedevice changes a pseudonym, the credibility of the previouspseudonym is passed to the next pseudonym. In addition,Boutsis and Kalogeraki [11] share user’s location informa-tion on mobile devices. Each user only knows part of thelocation information, so they cannot identify the informationsource.

In the aspect of data encryption, the focus of related workis on lightweight encryption mechanisms because of thelimitations of mobile device’s computing power and theirenergy consumption. Currently, the standard method widelyused in wireless communication is the Public Key Infrastruc-ture (PKI). Also, due to D2D communication, it should usemulti-party and distributed cryptographic protocols. In 2007,Kate proposed an Identity-based Cryptography (IBC) system,where each device node can create a public key using locallyavailable information [12]. They do not need to go to thecentral node to verify the public key, so the certificates can bedirectly exchanged with other mobile device nodes. In addi-tion, Searchable Encryption (SE) can also be applied to D2Dnetworks. It allows users to search for keywords in ciphertext.

III. RELIABILITY-BASED CENTRAL NODE ELECTIONMECHANISMIn the data security protection mechanism based on homo-morphic encryption, a central node is required to collectthe encrypted ciphertext of each mobile device, performsecure data aggregation operation and send it to the devicethat requests the data in the D2D network. At the sametime, the network also needs the central node to manage theresources in the entire network. In a traditional communi-cation network, this core device is usually a base station.However, in the D2D network, there is often no base stationas a core device. Therefore, it is necessary to elect a device asa central node in the D2D network. Due to the complexity ofthe D2D network environment, the assessment of eachmobiledevice in the network should consider its internal and externalfactors. In order to meet the above requirements, a reliability-based central node election mechanism is described in thissection.

A. RELATED TERMS1) Term 1: Min-max Normalization [13]. The origi-

nal data is transformed linearly so that the resultis between 0 and 1. The conversion function isXnorm = (X − Xmin)/(Xmax − Xmin), where Xnorm is anormalized data and Xmin and Xmax are a minimum anda maximum value of the original data set, respectively.

2) Term 2: Device Node Density [14]. One of the electionfactors of the reliability-based central node electionmechanism refers to the density of device nodes arounda device node in the D2D network, which is used todescribe the degree of mobile device nodes located atthe center of the D2D network.

3) Term 3: Service Success Rate. One of the election fac-tors of the reliability-based central node election mech-anism is the inspection of the device node providingservices to the network and representing the ratio of thenumber of success service times and the total numberof services provided by the device node historically.

4) Term 4: Service Hours. One of the election factors ofthe reliability-based central node election mechanismis the inspection of the device node providing servicesto the network and showing the time when the devicenode historically provides services.

B. MAIN IDEATo elect the central node for secure data aggregation andmanagement of network resources in a D2D network, a reli-ability mechanism is proposed to evaluate and calculate thereliability of device nodes from different aspects of mobiledevice nodes. In this mechanism, we firstly collect informa-tion of each mobile device node in the D2D network, and theinformation is then normalized and weighted summation toget the reliability of each mobile device. Finally, accordingto this reliability, the central node of data aggregation andmanagement network resources is elected in the current D2Dnetwork. The process of the election mechanism is shownin Figure 2. The detailed description is as follows.

FIGURE 2. The process of the election mechanism.

51142 VOLUME 6, 2018

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

1) The information of eachmobile device node in the D2Dnetwork is collected in a quantized form, including:service hours, calculation performance, service suc-cess rate, device node density and transmission perfor-mance. The reliability of each mobile device node isobtained by using these quantized data.

2) According to theD2Dnetwork environment, theweightof each election factor is obtained through repeatingexperiments, and each factor is weighted to obtain thereliability of each mobile device.

3) The reliability of all devices in the D2D network issorted, the most trusted mobile device node is electedto perform the central node of secure data aggregation.

C. QUANTIFICATION OF DEVICE NODE INFORMATIONTo calculate the reliability of each mobile device node, all thefactors need to be quantified to get the election factor. Theabove five factors are quantified as follows.

1) Quantification of service hours. The history servicehour of each device in the D2D network is mi. Thegreater the value of mi, the longer the service hours ofthe mobile device.

2) Quantification of device node density. Broadcastinghello information to the entire network, the ratio De ofthe number of neighboring mobile device nodes to thetotal number of mobile device nodes in the network iscalculated. The specific process is as follows.

1: sendHelloMessage(HelloBean)//Broadcast Hello messages to the entire network

2: list<HelloBean>=recieveHelloMessage()//Receive Hello messages from the entire network

3: N ← list.length+ 1//N is the number of device nodes in the entirenetwork

4: ni//ni is used to count the number of neighbor nodes

5: for i = 1→ list.length do6: if list[i].jump == 1 then7: ni+ = 1

//Device node with hop count 18: end if9: end for10: Dei = ni/N

//Calculate the node density of the device,Dei ∈ [0, 1)

3) Quantification of calculation performance. The numberof milliseconds required for each mobile device toperform 1Kb data encryption in the D2D network is xi.The lower the value of xi, the better the computingperformance of the mobile device node.

4) Quantification of transmission performance. Eachmobile device in the D2D network requests 1Mb ofdata to each mobile device on the entire network.

The number of mobile device nodes in the entire net-work is N . The number of milliseconds between eachdevice’s request and the time it takes to receive a totalof NMb’s data is yi. The lower the value of yi, the bet-ter the transmission performance of the mobile devicenode.

5) Quantification of service success rate. The ratio ofhistorical service times to successful service times ofdevice nodes is Sui. The process is as follows.

1: Z ← getServiceTimes()//Z is the total number of services in the device node

history2: Zi← getSuccessServiceTimes()

//Zi is the number of service successes in the devicenode history

3: Sui = Zi/Z//Sui is the service success rate of the device,

Sui ∈ [0, 1]

Quantifying the above five participating factors, theabstract description of the network environment and devicenode performance can be visually displayed in a digital way.At the same time, it can quantitatively express the differencesbetween node and node factors and provide basis for thecalculation of reliability and the election of central nodes.

D. ELECTION FACTOR NORMALIZATIONTo make the value of the finally calculated reliability at [0,1],the quantified election factors need to be normalized. Thenode density and the service success rate of mobile devicein the five election factors are already at [0,1], so only theservice hours, calculation performance, and transmission per-formance need to be normalized. Using the min-max nor-malization algorithm to normalize service hours, as shown inAlgorithm 1. However, the calculation performance and thetransmission performance are inversely proportional to theirquantified values. Therefore, the original min-max normal-ization algorithm needs to be simply modified so that the nor-malized value can react the mobile device node’s calculationand transmission performance, as shown in Algorithm 2 andAlgorithm 3.

The Algorithm 1 is to obtain the value of the node servicehour election factor through the service hours of the node.Firstly, we obtain the dataset of historical service hours of allnodes in the entire D2D network, themaximum andminimumvalues in the dataset, that is, the longest service hours and theshortest service hours, and calculate the difference betweenthem. Then, we calculate the difference between the servicehours and the shortest service hours of each node. Finally,we find the ratio of two differences as the service hour elec-tion factor. As can be seen from the expression, the longer theservice hours of the node, the greater the value of the servicehour factor, and the better the node performing on the item.

The Algorithm 2 is to obtain the value of the node calcula-tion performance election factor through the encryption time

VOLUME 6, 2018 51143

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

Algorithm 1 Normalize the Service TimeInput: the serviceTime dataset M , the number of

serviceTime miOutput: weight of serviceTime Sei, Sei ∈ [0, 1]1: for i = 0→ M .length do2: if mmax < m[i] then3: mmax = m[i]4: end if5: if mmin > m[i] then6: mmin = m[i]7: end if8: end for9: return Sei = (mi − mmin)/(mmax − mmin)

Algorithm 2 Normalize the Calculation PerformanceInput: the calculationPerformance dataset X , the number of

calculationPerformance xiOutput: weight of calculationPerformance Cai,

Cai ∈ [0, 1]1: for i = 0→ X .length do2: if xmax < x[i] then3: xmax = x[i]4: end if5: if xmin > x[i] then6: xmin = x[i]7: end if8: end for9: return Cai = (xmax − xi)/(xmax − xmin)

of the node. Firstly, we obtain the dataset of encryption timeof all nodes in the entire network, the maximum and mini-mum values in the dataset, that is, the longest encryption timeand the shortest encryption time, and calculate the differencebetween them. Then, we calculate the difference between thelongest encryption time and the shortest encryption time ofeach node. Finally, we find the ratio of two differences as thecalculation performance election factor. As can be seen fromthe expression, the shorter the encryption time of the node,the greater the value of the calculation performance factor,and the better the node performs on the item.

The algorithm 3 is to obtain the value of the node trans-mission performance election factor through the encryptiontime of the node. Firstly, we obtain the dataset of transmissiontime of all nodes in the entire network, the maximum andminimum values in the dataset, that is, the longest transmis-sion time and the shortest transmission time, and calculate thedifference between them. Then, we calculate the differencebetween the longest transmission time and the shortest trans-mission time of each node. Finally, we find the ratio of twodifferences as the transmission performance election factor.As can be seen from the expression, the shorter the transmis-sion time of the node, the greater the value of the transmissionperformance factor, and the better the node performing on theitem.

Algorithm 3 Normalize the Transmission PerformanceInput: the transmissionPerformance dataset Y , the number

of transmissionPerformance yiOutput: weight of transmissionPerformance Tri,Tri ∈

[0, 1]1: for i = 0→ Y .length do2: if ymax < y[i] then3: ymax = y[i]4: end if5: if ymin > y[i] then6: ymin = y[i]7: end if8: end for9: return Tri = (ymax − yi)/(ymax − ymin)

E. ELECTION OF A CENTRAL NODE THROUGHRELIABILITYAfter obtained the normalized election factors, the reliabilityof each mobile device node needs to be obtained throughthe weighted summation of the election factors. The weightsof the above five factors are α, β, γ , δ, ε and the weightssatisfy the expression α + β + γ + δ + ε = 1. Due to thecomplexity of the D2D network, the weights of the factorsare based on actual application scenarios and experiments.The reliability of the weighted summation obtained by Rei,as shown in Algorithm 4.

Algorithm 4 Calculate the ReliabilityInput: the node density Dei, success rate of service Sui,

weight of serviceTime Sei, weight of calculatedPer-formance Cai, weight of transmissionPerformance Tri,weight of five factors α, β, γ, δ, ε

Output: reliability Rei,Rei ∈ [0, 1]1: return Rei = α∗Dei+β ∗Sei+γ ∗Cai+δ∗Tri+ε∗Sui

The higher the reliability value, the better themobile devicenode is as the center node. After obtaining the reliability ofeach mobile device node in the D2D network, a fast sort algo-rithm is used to sort the reliability of each node in the D2Dnetwork. According to the result of quick sorting, the mobiledevice node with the highest reliability is selected as thecentral node, and the mobile device node with the secondhighest reliability is used as the backup central node. Whenthe central node cannot provide services normally, the centralnode is replaced by the backup central node.

IV. PRIVATE DATA PROTECTION MECHANISM BASEDON PAILLIER HOMOMORPHIC ENCRYPTIONIn the D2D network, mobile device nodes often need torequest network resources in the entire network, and theseresources often contain some private information. If the rele-vant data is directly sent to the requesting device, it is easy tocause the privacy disclosure and the insecure D2D network.Therefore, the private data in the D2D network needs to be

51144 VOLUME 6, 2018

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

encrypted and then sent to the device that requested data.To achieve the above requirements, the Paillier homomorphicencryption model [15] is introduced into the D2D network,and the proposed private data protection mechanism basedon homomorphic encryption is described as follows.

A. RELATED TERMS1) Term 1: Homomorphic Encryption. The multiple

encrypted data is calculated directly to obtain an inte-grated ciphertext, which is the same as that obtained byprocessing the unencrypted original data in the sameway.

2) Term 2: Additive homomorphism. There is an algo-rithm

⊕that makes E(x+y) = E(x)

⊕E(y) or x+y =

D(E(x)⊕

E(y)) true without leaking x and y.3) Term 3: Paillier encryption. Paillier encryption system,

a probabilistic public key encryption system inventedby Paillier [15] in 1999. The encryption algorithm isa homomorphic encryption that satisfies addition andmultiplication homomorphism.

B. MAIN IDEAWhen a mobile device node in the D2D network requestsresources from the entire network, a data privacy pro-tection mechanism is needed to meet this requirement.Figure 3 employs homomorphic encryption algorithm [15] toencrypt the data of each mobile device node and sends it tothe requesting device after secure data aggregation. Firstly,each mobile device node encrypts the data after receivingthe request. Then the ciphertext is sent to the elected centralnode, which aggregates the received ciphertext and sends itto the requesting device. Finally, the requesting device nodedecrypts the ciphertext to obtain the resource.

FIGURE 3. Architecture of private data protection mechanism based onhomomorphic encryption.

C. KEY GENERATIONIn the key generation phase, each mobile device node needsto generate its own public key and private key and sends theirpublic key to the elected central node. Each node in D2Dnetwork performs Paillier public and private key generationalgorithm. k mobile devices generate k sets of public keys(n, g) and private keys (λ,µ). hey send the public keys tothe central node, and the private keys are saved locally.

Paillier public and private key generation algorithm, theprocess is as follows [15].

1: Select prime numbers p and q randomly to satisfygcd(pq, (p− 1)(q− 1)) = 1

2: Let n = pq, λ = lcm(p− 1, q− 1)3: Defining the function L is L(µ) = (µ− 1)/n4: Select random number g(g ∈ Z∗

n2) and satisfy

gcd(L(gλmodn2), n) = 1,µ = (L(gλmodn2))−1modn

5: The public key is (n, g) and the private key is (λ,µ).

Private data protection mechanism based on Paillier homo-morphic encryption allows new mobile devices to join theD2D network. When a new mobile device requests to join thenetwork, it firstly sends a request to the central node. Afterpassed the authenticates of the central node device, the newmobile device node will join the D2D network.When the newnode is added, the Paillier public and private key generationalgorithm needs to be executed. The public key of the newdevice node generated by the algorithm is sent to the centralnode, and the private key is locally saved.

D. ENCRYPTION PROCESSWhen a mobile device node in the D2D network needs torequest resources from the entire network, the requestingdevice firstly sends a request to the central node, whichbroadcasts a request to the entire network and publishes thepublic key of the requesting device. The other nodes in theD2D network need to encrypt the provided private data afterreceiving the request to prevent privacy leakage. Each nodeencrypts according to the obtained request and the public key,as shown in Algorithm 5 [15]. Finally, the encrypted cipher-text is sent to the central node, and the requesting device nodeneeds to be shielded in the routing process to prevent therequesting device node from decrypting the ciphertext.

Algorithm 5 Paillier EncryptionInput: the public key of request node (n, g), the clear-text

m(m ∈ Zn)Output: the cipher-text Ci1: r ← randome()r ∈ Z∗n2: return Ci = gm ∗ rnmodn2

After collecting the ciphertext of each node in the D2Dnetwork, the central node prevents the data of each node frombeing leaked by the requesting device node, the secure aggre-gated operation needs to be performed on the received data,as shown in Algorithm 6. Finally, the processed ciphertextis sent to the requesting node, so the requesting node devicecan only obtain the requested data, not the data from a singlenode device, which protects the private data of each nodedevice [16].

After receiving the ciphertext processed by the homo-morphic encryption, the requesting node device uses the

VOLUME 6, 2018 51145

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

Algorithm 6 Paillier Homomorphic ProcessInput: the cipher-text dataset COutput: the cipher-text after homomorphic process C∗

1: for i = 0→ C .length do2: C∗∗ = C[i]3: end for4: return C∗

private key saved locally to perform a decryption algorithmon the ciphertext to obtain the returned result, as shown inAlgorithm 7 .

Algorithm 7 Paillier DecryptionInput: the cipher-text after homomorphic process C∗,

the public key of request node (n, g), the private key ofrequest node (λ,µ)

Output: the return result m∗

1: return m∗ = L(Cλmodn2) ∗ µmodn

E. HOMOMORPHIC VERIFICATIONThe Paillier encryption algorithm [15] is an encryption algo-rithm with additive homomorphism, which is processed asfollows. Let message m1 and m2 exist, and encrypt them toget:

E(m1) = gm1xn1modn2 (1)

E(m1) = gm2xn2modn2 (2)

(1)*(2) to get:

E(m1) ∗ E(m2) = gm1xn1 ∗ gm2xn2modn

2= E(m1 + m2) (3)

V. THEORETICAL ANALYSIS AND PERFORMANCEEVALUATIONThe theoretical analysis section firstly verifies the D2D dataprivacy protection mechanism based on reliability and homo-morphic encryption proposed in this paper satisfying thepreset target. Based on this, the merits and rationality of themechanism are discussed. The analysis is based on the com-prehensiveness of the election mechanism, the complexity ofthe algorithm, and the efficiency of homomorphic encryption.The part of performance evaluation is based on the analysis ofthe experimental results to verify the reliability of the mech-anism and the reasonableness of the reliability calculationmethod of the mobile device node.

A. THEORETICAL ANALYSISBased on the D2D communication technology and networkstructure, this paper proposes a reliability-based centralnode election mechanism and employs Paillier homomorphicencryption system to construct a private data protectionmechanism in the D2D networks. It ensures the security ofprivate data in D2D networks while reasonably performsD2D network resource allocation. In the D2D network,

mobile device node’s reliability calculation process, the his-torical service hours, historical service success rate, devicenode density, calculation performance and transmission per-formance of device nodes are taken into consideration, andthe reliability of the device node is fully and reliably eval-uated. The program firstly calculates the credibility of theinternal factors of the mobile device node, including thenode’s calculation performance, historical service hours, andhistorical service success rate, and evaluates whether the nodedevice can allocate and manage D2D network resources andperform data aggregation. Secondly, the proposedmechanismevaluates the network environment of themobile device node,including the density of nodes around the mobile device nodeand the surrounding network transmission performance. If theperformance of the mobile device node is poor, the entireD2D network resource allocation may not be timely andthe security of private data may be reduced. To ensure thecredibility of the reliability, when calculating the reliability,it is not simply to sum up each election factor, but to usea weighted summation. In practice, the proportion of eachfactor’s weight may be adjusted according to the applicationof the network to select the most reasonable mobile devicenode as a central node in the D2D network.

In the same type of network node election mechanism,the paper [17] proposed a trusted center node selection algo-rithm, which firstly uses the history service success rate of theelection node as a direct trust value. Then, after the electionnode processing the service provided by the specified node,the recommendation trust value of the election node to thespecified node is obtained, and the recommendation trustvalues of all nodes in the network are added up and summedto obtain the recommendation trust value of the election node.Finally, according to the weights of the two parts, the compre-hensive trust value is calculated. The mechanism considersonly the node density of the mobile device node and the nodeservice, and lacks the consideration of the calculation andtransmission performance of the node.

The paper [18] proposed an algorithm that user require-ments adapt to the node selection. Firstly, the user putsforward requirements for nodes, including factors such ascalculation performance, bandwidth, and number of neighbornodes. These parameters are formed into a matrix, which isprocessed to obtain a criteria level. Secondly, user-definedweights of various parts constitute a matrix, which is pro-cessed to be a decision level. Finally, according to the ratioof the importance of the decision and the criteria, a decisionmatrix is obtained, and the values in the matrix are sortedto obtain the decision goal which is the election node. Thismechanism only considers the performance of the node itselfand the surrounding network, and the user must give thequantized parameters, and lacks the calculation method ofrelated parameters. The following is a summary of the abovethree mechanisms, as shown in Table.1.

In this paper, the algorithm complexity of the reliability-based central node election mechanism is O(n3 log2 n).Because each node in the network only needs to fetch data

51146 VOLUME 6, 2018

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

TABLE 1. Node selection algorithms comparison.

from the other n − 1 nodes, the number of execution timesof the entire network is n ∗ (n− 1). Then, by multiplying thecomplexity of quick sorting (O(n∗ log2 n)), we can obtain thecomplexity of our algorithm O(n3 log2 n). As a comparison,the complexity of the other two algorithms is O(n4 log2 n)with the number of execution times of the entire network in[17] and [18] being n3. Therefore, themechanism proposed inthis paper has a lower algorithm complexity. The comparisonof the algorithm complexity is shown in Table.2.

TABLE 2. Algorithm complexity comparison.

B. ELECTION PERFORMANCE EVALUATIONThe feasibility of the reliability-based central node electionmechanism is verified through repeated simulation experi-ments. Under the corresponding D2D communication net-work, the simulation experiment is used to verify thatthe nodes in the network can obtain the reliability under theactual situation, and elects the central node according to thereliability. The experimental environment is configured withan Intel Core i7-4710HQ CPU, 4G RAM, and Ubuntu 16.04Linux operating system (64-bit) Network simulation soft-ware adopts NS2, and simulation data processing softwareadopts gawk. The reason for employing NS2 over NS3 is thatNS2 has superior stability compared to Ns3.

Firstly, we employ NS2 to establish a D2D networktopology as shown in Figure 4. The bandwidth between thenode 2 and the node 3 is 100 Mbps and the delay is 10 ms,and the bandwidth between the other nodes is 500 Mbps andthe delay is 2 ms.

Secondly, we set relevant parameters for each node asfollows. Node 0, Node 1 and Node 2 are original nodes andthe service hours are 100 hours, 110 hours and 120 hoursrespectively. The service success rates are 85%, 90% and 95%respectively. Node 3, Node 4 and Node 5 are newly addednodes and the service hours are 10 hours and the service

FIGURE 4. NS2 network topology.

success rate is 100%. Node 2 and Node 3 are portablecomputer, and the key length is 1024 bits. According tothe encryption performance analysis, the encryption timeis 12.94ms. Node 0, Node 1, Node 4, and Node 5 areAndroid devices. According to experiments, the encryptiontime is about 20.85ms. The parameters of the node are shownin Table.3.

TABLE 3. Node parameters.

Finally, the network simulation software is used to simulatethe data transmission process to obtain the network perfor-mance parameters around each node. The network parametersof the nodes are shown in Table.4.

TABLE 4. Network parameters.

After all the election parameters of the node are obtained,five election factors are obtained by using the optimizednormalization algorithm in the election mechanism. Theobtained values of the reliability factors are shown in Table.5.

Getting the weight matrix of each election factor,we here set the same for each election factor,

VOLUME 6, 2018 51147

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

TABLE 5. The reliability values of election factors.

{α, β, γ, δ, ε} = {0.2, 0.2, 0.2, 0.2, 0.2}. Finally, accordingto the weights of each election factor, the weighted summa-tion of the reliability to each node is obtained. The reliabilityof each node is shown in Table.6.

TABLE 6. Node reliability.

As shown in the table above, Node 2 and Node 3 arelocated in the network center. Compared with other nodes,these two nodes directly connect with more other devicenodes, and the surrounding network bandwidth is conse-quently higher. Therefore, the central node should be selectedfrom Node 2 and Node 3. Compared to Node 3, Node 2 hassuperior calculation performance, transmission performanceand longer service hours. For these reasons, Node 2 is moresuitable for serving as a network center node for managingnetwork resources. The experimental results are in line withexpectations.

C. ENCRYPTION PERFORMANCE EVALUATIONThe feasibility of Paillier’s homomorphic encryption algo-rithm [15] is verified through repeated experiments. In thetraditional network, the central node is often a base stationwith good performance. In the third mode of D2D network,there is no base station as the core of the network. There-fore, the elected central node may be a portable computeror even a smart phone. The experimental environment wasconfigured with an Intel Core i7-4710HQ CPU, 8G memory,Windows 10 Ultimate operating system (64-bit), and encryp-tion algorithm implementation software adopts MyEclipse.

Firstly, we verify the relationship between the data encryp-tion time and the key length under the experimental envi-ronment. Generally, the longer the key length is, the lowerthe probability of private data leakage is, and the higher thesecurity of data, while the encryption time of the unit datawill also increase. Therefore, the key length needs to beadjusted according to the size and importance of the data. Theexperiment employs the Paillier encryption algorithm [15]to encrypt 8-byte-size bigInt data types for different lengthsof keys and to calculate the time. Since the key generationprocess includes the generation of a random prime number,the time of the encryption process is related to the value of

the prime number. Therefore, ten keys of the same data sizeare generated for encryption, and the average value of theencryption time is taken as the encryption time of the lengthkey. The relationship between key length and encryption timeis shown in Figure 5. From this figure, the encryption timeof data increases exponentially with the increase of the keylength, which is in line with expectations.

FIGURE 5. Key length and encryption time.

Secondly, we verify the relationship between the dataaggregation time and the number of aggregated data andkey length under the experimental environment. Generally,the greater the number of data, the longer the key length,the longer the data aggregation time, and also the longer thenetwork responds to the request. The data aggregation time isalso related to the homomorphic algorithm. The experimentuses the Paillier homomorphism algorithm to perform dataaggregation on the ciphertext of bigInt data when the keysare the same. The data aggregation time of different dataamounts is counted. Then, we use different key lengths toaggregate data for 10 bigInt data types and calculate the time.Since the key generation process includes the generation ofa random prime number, the time of the encryption processis related to the value of the prime number, so 10 groups ofkeys with the same data size are generated for encryption,and the time average time value of the data aggregation istaken as the aggregation time of the length key. The relation-ship between the number of aggregated data and aggregationtime is shown in Figure 6, and the relationship between keylength and aggregation time is shown in Figure 7. From thesefigures, we can see that the aggregation time of data growslinearly with the increase of aggregated data quantity, andexponentially increases with the increase of key length, whichis in line with expectations.

Finally, we verify the relationship between data encryptiontime and key length under the experimental environment.Generally, the longer the key length, the longer the decryp-tion time of the unit data. The experiment uses the Paillierdecryption algorithm to decrypt the ciphertext aggregatedby 10 times of 8-byte bigInt data types using different keylengths and to calculate the time. Since the key generationprocess includes the generation of a random prime number,the time of the decryption process is related to the value of theprime number, so ten keys of the same data size are generated

51148 VOLUME 6, 2018

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

FIGURE 6. Data amounts and aggregation time.

FIGURE 7. Key length and aggregation time.

for encryption, and the average value of the decryption timeis taken as the decryption time of the length key. The rela-tionship between key length and decryption time is shownin Figure 8. From this figure, the decryption time increasesexponentially with the increase in key length, which is in linewith expectations.

FIGURE 8. Key length and decryption time.

VI. CONCLUSIONWith the rapid development of the fifth-generation (5G)communication, D2D communication technology will deeplyaffect everyone’s daily work and life. For this reason, the data

security and privacy of D2D communications becomes a criti-cal issue andmajor challenge in D2D communications. In thispaper, we introduce the homomorphic encrypted private dataprotection mode in D2D communication and solved the issueof lacking reliable devices to perform secure data aggregationoperation without a base station. The proposed mechanismimproves the security and anti-attack ability of D2D networksand optimizes the resources allocation in D2D network.

Our future work will be focused on the following aspects:1) delving more deeply into the connection and influence fac-tors between election factors and the impact of the dynamicsof wireless devices on D2D networks; 2) the D2D networkis dynamic in reality: a) the distance and bandwidth amongnodes, the performance of nodes and other factors are con-tinuously changing; b) new nodes can join the network whileexisting nodes can leave at arbitrary moments. How to electcentral nodes in the real network environment needs to befurther studied.

REFERENCES[1] R. I. Ansari et al., ‘‘5G D2D networks: techniques, challenges, and future

prospects,’’ IEEE Syst. J., to be published, [Online]. Available: https://ieeexplore.ieee.org/abstract/document/8214255/, doi: 10.1109/JSYST.2017.2773633.

[2] D. Wu, J. Yan, H. Wang, D. Wu, and R. Wang, ‘‘Social attribute awareincentive mechanism for device-to-device video distribution,’’ IEEE Trans.Multimedia, vol. 19, no. 8, pp. 1908–1920, Aug. 2017.

[3] D. Huo, Z. Zhang, Q. Zhou, Q. Tang, and Z. Huo, ‘‘Ergodic secrecysum rate maximization in downlink sparse code multiple access network,’’presented at the Int. Symp.Wireless Pers. Multimedia Commun. (WPMC),Bali, Indonesia, Dec. 2017.

[4] M. Haus, M. Waqas, A. Y. Ding, Y. Li, S. Tarkoma, and J. Ott, ‘‘Securityand privacy in device-to-device (D2D) communication: A review,’’ IEEECommun. Surveys Tuts., vol. 19, no. 2, pp. 1054–1079, 2nd Quart., 2017,doi: 10.1109/COMST.2017.2649687.

[5] A. Behrooz and A. Devlic, ‘‘A context-aware privacy policy language forcontrolling access to context information of mobile users,’’ in Securityand Privacy in Mobile Information and Communication Systems. Berlin,Germany: Springer, 2011, pp. 25–39.

[6] S. Chakraborty, Z. Charbiwala, H. Choi, K. R. Raghavan, andM. B. Srivastava, ‘‘Balancing behavioral privacy and informationutility in sensory data flows,’’ Pervasive Mobile Comput., vol. 8, no. 3,pp. 331–345, Jun. 2012. [Online]. Available:https://link.springer.com/chapter/10.1007/978-3-540-73549-6_91,doi: 10.1016/j.pmcj.2012.03.002.

[7] R. Wishart, K. Henricksen, and J. Indulska, ‘‘Context privacy and obfus-cation supported by dynamic context source discovery and processing in acontext management system,’’ in Proc. Int. Conf. Ubiquitous Intell. Com-put. (UIC), in Lecture Notes in Computer Science, vol. 4611. Hong Kong:Springer, pp. 929–940, 11 Jul. 2007, doi: 10.1007/978-3-540-73549-6_91.

[8] E. Franz, T. Springer, and N. Harder, ‘‘Enhancing privacy in social applica-tions with the notion of group context,’’ inProc. Int. Conf. Internet Technol.Secured Trans., London, U.K., Dec. 2012, pp. 112–118.

[9] D. Wu, S. Si, S. Wu, and R. Wang, ‘‘Dynamic trust relationships awaredata privacy protection in mobile crowd-sensing,’’ IEEE Internet Things J.,vol. 5, no. 4, pp. 2958–2970, Aug. 2018, doi: 10.1109/JIOT.2017.2768073.

[10] D. Christin, C. Roßkopf, M. Hollick, L. A. Martucci, and S. S. Kanhere,‘‘IncogniSense: An anonymity-preserving reputation framework for partic-ipatory sensing applications,’’ presented at the IEEE Int. Conf. PervasiveComput. Commun., Lugano, Switzerland, Mar. 2012.

[11] I. Boutsis and V. Kalogeraki, ‘‘Privacy preservation for participatory sens-ing data,’’ in Proc. IEEE Int. Conf. Pervasive Comput. Commun. (PerCom)San Diego, CA, USA, Mar. 2013, pp. 103–113.

[12] A. Kate, G. M. Zaverucha, and U. Hengartner, ‘‘Anonymity andsecurity in delay tolerant networks,’’ presented at the 3rd Int. Conf.Secur. Privacy Commun. Netw. Workshops (SecureComm), Nice, France,Sep. 2007.

VOLUME 6, 2018 51149

B. Jin et al.: D2D Data Privacy Protection Mechanism Based on Reliability and Homomorphic Encryption

[13] Y. K. Jain and S. K. Bhandare, ‘‘Min max normalization based dataperturbation method for privacy protection,’’ Int. J. Comput. Commun.Technol., vol. 2, no. 8, pp. 45–50, Oct. 2011.

[14] J.-F. Qiao, S.-Y. Liu, and X.-Y. Cao, ‘‘Density-based clustering protocolfor wireless sensor networks,’’ Comput. Sci., vol. 36, no. 12, pp. 46–49,Dec. 2009.

[15] P. Paillier, ‘‘Public-key cryptosystems based on composite degree residuos-ity classes,’’ in Proc. Int. Conf. Theory Appl. Cryptograph. Techn., Prague,Czech Republic, Mar. 1999, pp. 223–238.

[16] S.M. Anggriane, S.M. Nasution, and F. Azmi, ‘‘Advanced e-voting systemusing Paillier homomorphic encryption algorithm,’’ presented at the Int.Conf. Inform. Comput. (ICIC), Mataram, Indonesia, Oct. 2016.

[17] S. Xiong, ‘‘The study of trust node interaction on the multiple socialnetwork,’’ M.S. thesis, Shijiazhuang Tiedao Univ., Shijiazhuang, China,2016. [Online]. Available: http://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=CMFD&dbname=CMFD201701&filename=1016769831.nh&v=MjE2MjIxTHV4WVM3RGgxVDNxVHJXTTFGckNVUkxLZmJ1ZHZGQ2pnVmJ2SlZGMjZHTFMrRjluUHJwRWJQSVI4ZVg=

[18] T. Xu, ‘‘Research on the selection of super node in peer-to-peernetwork,’’ M.S. thesis, Univ. Sci. Technol. China, Hefei, China, 2008.[Online]. Available: http://kns.cnki.net/KCMS/detail/detail.aspx?dbcode=CMFD&dbname=CMFD2010&filename=2010019665.nh&v=Mjc4ODhidWR2RkNqaFVMcklWMTI2SHJPNUY5ZktxcEViUElSOGVYMUx1eFlTN0RoMVQzcVRyV00xRnJDVVJMS2Y=

[19] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan,‘‘An application-specific protocol architecture for wireless microsensornetworks,’’ IEEE Trans. Wireless Commun., vol. 1, no. 4, pp. 660–670,Oct. 2002, doi: 10.1109/TWC.2002.804190.

BIAO JIN received the M.S. degree in soft-ware engineering from the Faculty of Software,Fujian Normal University, China, in 2011, wherehe is currently pursuing the Ph.D. degree. He isalso a Lecturer with Fujian Normal University.His research interests include computer network,geographic information, and data mining.

DONGSHUO JIANG received the B.Eng. degreein software engineering from the Faculty of Soft-ware, Fujian Normal University, China, in 2018.He is currently pursuing the M.S. degree with theUniversity of New South Wales, Australia.

JINBO XIONG (M’13) received the M.S. degreein communication and information systems fromthe Chongqing University of Posts and Telecom-munications, China, in 2006, and the Ph.D. degreein computer system architecture from XidianUniversity, China, in 2013. He is currentlyan Associate Professor with the College ofMathematics and Informatics, Fujian Normal Uni-versity. His research interests include cloud datasecurity, privacy protection, and mobile Internet

security. He has published more than 40 publications and one monographand holds eight patents in these fields.

LEI CHEN (M’05) received the B.Eng. degreein computer science and applications from theNanjing University of Technology, China, in 2000,and the Ph.D. degree in computer science andsoftware engineering from Auburn University,USA, in 2007. He is currently an Associate Pro-fessor, the Interim Department Chair, and theGraduate Program Director with the Departmentof Information Technology, Georgia SouthernUniversity, USA. He has authored or co-authored

over 100 peer-reviewed scholarly works. His research interests focus onnetwork, information, cloud, and big data security, digital forensics, andmobile, handheld, and wireless security.

QI LI received the Ph.D. degree in computer sys-tem architecture from Xidian University, Xi’an,China, in 2014. He is currently a Lecturer withthe School of Computer Science, Nanjing Univer-sity of Posts and Telecommunications, China. Hisresearch interests include cloud security, informa-tion security, and applied cryptography.

51150 VOLUME 6, 2018