1. Introduction
Future fifth-generation (5G) networks are expected to support a huge number of heterogeneous connected devices in manifold usage scenarios. Autonomous driving, tactile Internet, personal cloud, disaster alert, video streaming and downloading are among the most challenging 5G use cases. Some technologies are expected to play a key role in helping to satisfy the demanding requirements of the foreseen use cases. Device-to-device (D2D) communications and multicasting are certainly among these. The former, because of its capability to offload cellular data traffic, enhances spectrum efficiency and extends cell coverage [
1]. The latter, because of its capability to answer to the increasing user demand for multicast/broadcast multimedia services (mobile TV, IP radio broadcasting, and video streaming [
2]).
Regarding 5G in general, in the 3rd Generation Partnership Project (3GPP), the 5G radio access roadmap foresees two tracks: one is based on the evolution of Long-Term Evolution (eLTE), and the other on the design of the New Radio (NR) access. Thus, the core enhancements/changes of the 5G paradigm are (i) the standardization of a new radio interface (i.e., called New Radio-NR in 3GPP and other standardization bodies) and (ii) the empowering of the existing LTE systems in order to handle use cases (e.g., mMTC, cMTC, V2X or Sidelink) for which NR is still not mature yet (i.e., the full specification of NR, it was agreed just on the 15th of June during the 3GPP RAN#80 plenary). Consequently, in relation to the former case, it is good to point out that LTE constitutes an essential piece of the 5G puzzle and can be considered as a 5G-ready technology due to the variety of enhancements and new features already introduced in Rel-14 and Rel-15 timeframes in 3GPP. Therefore, even if the research efforts on D2D (i.e., later called Sidelink) and multicast/broadcast services started way before the concept of “5G” gained momentum, it is fairly evident that these technologies have evolved over the years to meet requirements more and more challenging that nowadays are very close (if not the same) to those of the future 5G systems.
Multicast transmission is an effective means for delivering group-oriented services since users can be fed through a single point-to-multipoint (PtM) transmission by exploiting the broadcast nature of the radio channel. In order to handle multicast and broadcast services over cellular networks, the 3rd Generation Partnership Project (3GPP) has standardized the evolved Multimedia Broadcast Multicast Service (eMBMS) [
3]. Although it represents the current standard to support group-oriented services over mobile networks, significant work must be done in order to effectively support multicast/broadcast traffic in 5G networks. The approach for delivering multicast traffic in cellular networks, the so-called Conventional Multicast Scheme (CMS) [
4], is known to suffer from poor spectral efficiency. According to it, all users belonging to the multicast group are served with the same data rate imposed by the user with the worst channel conditions. Despite this approach guaranteeing fairness because all users are always served and receive the same treatment, it suffers from poor performances because users with good channel conditions are constrained to the low data rates that cell-edge users can sustain.
In our earlier work in [
5], we proposed a D2D-aided radio resource management policy for eMBMS, called D2D-enhanced CMS with Single Frequency (D2D-SF), with the aim to increase the aggregate data rate of CMS while maintaining the CMS short-term fairness. This protocol includes a first step, in which users with the best channel conditions are served via CMS directly by the eNodeB. Then, the served nodes forward the received data to the excluded users over D2D links. The peculiarity of this approach is that the forwarding devices send data simultaneously on the same frequency to their D2D receivers. Even though the good results of the protocol D2D-SF have been widely demonstrated, security in communications is not taken into account.
The need to look at aspects related to security stems from the fact that the 5G system is designed with security requirements in mind. In particular, the following properties make 5G a trustworthy multi-service platform: resilience, communication security, identity management, privacy and security assurance [
6]. In addition to the magnified risk of security threats due to the huge number of 5G connected devices, D2D communications cause further problems due to connections happening directly between devices in proximity [
7].
Several works in the literature deal with security in D2D communications. A classification and discussion of solutions to secure D2D communication can be found in [
8,
9], wherein a comparison of the different approaches is also performed in terms of their ability to satisfy D2D security requirements, such as confidentiality and integrity, authentication, privacy, non-repudiation, and so on. Among these compared approaches, the work in [
10] is identified as the more effective, since it is able to satisfy most of the security requirements posed by future 5G networks [
8,
9].
The purpose of this work is to present an algorithm that improves the CMS performance by exploiting secure D2D communications. The proposed enhanced CMS with secure D2D communications (eCMS-sD2D) can efficiently and securely deliver multicast traffic in 5G networks. As in the cited [
5], D2D clusters are formed in order to forward data towards users with the worst channel conditions, which are excluded from a first multicast transmission by the eNodeB (eNB). In this paper, we enhance the work in [
5] by selecting the relay node (i.e., the RN, which is in charge to forward data sent by the eNB towards users unable to directly receive the multicast transmission) in each D2D cluster on the basis of its trustworthiness measured, by means of security mechanisms, as their capability of reliably acting as data forwarders. Thus, our proposed scheme can be seen as a clear security enhancement with respect to [
5] in that it takes into account, in the D2D communications’ procedure, security aspects previously not considered at all. In particular, security procedures inspired from [
10] are implemented in order to guarantee confidentiality and integrity of data transmitted in D2D communications, and D2D users’ privacy protection. Security mechanisms as encryption, keyed-hash message authentication code (HMAC), and signature are used to manage the message exchange between the peers. Data encryption is realized through a symmetric encryption algorithm, with the same private key used to encrypt and decrypt data. The private key is generated through an enhanced version of the Diffie–Hellman key exchange (DHKE) protocol, in which the public key exchange is intermediated by the eNodeB, representing the trusted third party. In addition, message authentication helps to avoid the man-in-the-middle attack, which represents the primary vulnerability of the DHKE algorithm. Thanks to the designed security mechanisms, many typical D2D attacks can be avoided. Among these, eavesdropping, impersonation and masquerading, besides the already cited man-in-the-middle. The detection of any security attack reduces the reliability of the node and, consequently, the probability of being selected as data forwarder.
The remainder of the paper is organized as follows.
Section 2 and
Section 3 present, respectively, the related works in this field and the basics of the proposed eCMS-sD2D.
Section 4 describes the designed algorithm in details. Results from our simulative analysis are shown in
Section 5. Conclusive remarks are given in the last section.
2. Related Work
In the literature, many works deal with the candidate supporting technologies for enabling future 5G networks [
11,
12]. D2D communications is considered among these because of its capability to improve network performance in terms of delay, throughput, energy efficiency, and spectral efficiency.
The D2D taxonomy presented in [
13] distinguishes between inband and outband D2D. Inband D2D communications use the cellular (i.e., licensed) spectrum and can be further categorized as underlay or overlay when the radio resources are shared with cellular users or are dedicated to D2D links. Differently, outband D2D exploits an unlicensed spectrum. To this aim, an extra interface must be used and controlled by the cellular network (i.e., controlled mode) or by the users (i.e., autonomous mode). Usually, the underlay inband D2D is preferred to the other modes because it better utilizes the spectrum and is suitable for all types of devices. As concerns the resource allocation in D2D communications, several solutions can be found in the literature related to D2D [
14] or in closely related fields [
15].
An interesting usage scenario for the D2D technology is multicasting. Multicast and broadcast services are enabled in 3GPP Release 8 networks through the evolved Multimedia Broadcast Multicast Service (eMBMS) architecture. According to [
16], the group-oriented services will be protagonists in 5G networks, and multicasting will be the effective means to offer these services. Establishing direct communications between devices in proximity has been shown to improve the multicasting performance. In [
5], the D2D-enhanced CMS with Single Frequency (D2D-SF) approach was designed to improve the performance of the Conventional Multicast Scheme (CMS), counting on D2D communications occurring simultaneously at the same frequency. This paradigm has shown encouraging results in Long-Term Evolution-Advanced (LTE-A) systems.
A further work focusing on content sharing by exploiting both D2D and multicasting technologies can be found in [
17]. In this article, the authors present a device-to-device multicast (D2MD) scheme for content sharing in cellular networks by taking into account social and physical attributes in D2MD cluster formation, and jointly optimizing power and channel allocation among D2MD clusters. However, trust and security aspects are still overlooked even if they play a fundamental role when involving social and physical relationships. Another work that addresses a similar topic is presented in [
18]. Here, the authors considered a scenario where D2D users may demand multicast data at various rates and, in return, they offer different profits (revenue) to the telecom operator. However, it is shown that satisfying the user requests to maximize the profit is an NP-hard when the resource blocks are limited, and thus they propose a greedy heuristic algorithm to solve this problem. An interesting analysis of content dissemination scenarios is presented in [
19], where two game theoretic medium access strategies, based on energy-aware utility functions, are proposed. The impact of cellular network characteristics on D2D communication is analyzed in [
20], where the authors also exploit the benefits of network coding (NC) in the design of an adaptive cooperative protocol for the D2D data exchange.
No attention has been given so far to security that represents an important aspect for future 5G networks; this is the focus of this paper. According to [
21], privacy and security issues must be faced in order to definitely make D2D a successful technology. In [
8], a thorough analysis on these problems is conducted. First, the difference between security and privacy concepts is defined. Then, the requirements to be satisfied in order to guarantee both security and privacy in D2D communications are listed, and the possible attacks are identified. Finally, related works and proposed solutions are described. Among these, Zhang et al. describe in [
10] a secure data sharing strategy able to guarantee D2D privacy and security in LTE-A networks. The basic idea is to encrypt data transmitted in the D2D link using a symmetric encryption algorithm and generating the private key, for data encryption and decryption, following the Diffie–Hellman key exchange protocol. The strength of this strategy lies in the intervention of the eNodeB that represents a trusted third party and protects against malicious behaviors such as the man-in-the-middle attack. In [
9], security solutions proposed improving D2D in 5G networks are analyzed. In addition to reporting D2D security requirements, threats and solutions, this work hints at the role that the social relationships could play in improving the D2D security. This is an interesting starting point for future research aimed at exploiting concepts, such as social trust, to evaluate how much a network node can be reliable. In this work, we aim to add the security dimension to the D2D-SF solution we proposed in [
5] in order to make it suitable to 5G networks and cope with typical D2D vulnerability attacks.
4. The Proposed eCMS-sD2D Protocol
The reference scenario of this work is composed of a set of devices interested in downloading data over the 5G network. This generic scenario is suitable to several types of applications, from software update to video downloading, and to both human-oriented and machine-oriented communications. Data transmission is accomplished over a multicast transmission by the eNB. The considered architecture is composed of all the nodes foreseen in the eMBMS standard architecture, but procedures must be improved to make multicast transmission more suitable for future 5G networks. Our proposed solution called enhanced CMS with secure D2D communications (eCMS-sD2D) aims to enhance performance and security of a multicast CMS transmission. All steps of the proposed eCMS-sD2D solution are depicted in
Figure 2.
Our proposal is tailored to cellular 5G networks where the eNodeB acts as a central coordinator of both D2D and traditional communications. In fact, we focus on network-assisted D2D communications in a licensed spectrum, which is the preferred solution because of the expected gains it offers [
1]. In such a scenario, it is a natural choice to also make the eNodeB work a trusted third party for the implementation of the security mechanisms. Differently, in a distributed scenario (that is not the case analyzed in this paper and that is not the preferred option for D2D communications), the selection of the node that will act as a trusted third party is a tricky choice that needs to be carefully investigated.
Step 1—Multicast service delivery notification
Through this first step, the eNB invites users interested in the service to form the multicast group (MG) by registering with the network.
Step 2—CQI collection
Users belonging to the MG send their CQI values to the eNB. Each user equipment (UE) has to communicate not only the cellular CQI value for its link towards the eNB, but also those for the D2D links to its neighbors. The eNB stores the received D2D CQI values in a matrix.
Step 3—Multicast and D2D configuration selection
Thanks to the collected information, the eNB can establish: (i) the set of registered UE to serve through the multicast transmission; (ii) the MCS to use for the multicast transmission in the CMS coverage area; (iii) the users served in a multicast that can act as relay nodes by forwarding data received directly by the eNB towards the cell-edge users; (iv) the D2D pair to establish; and (v) the MCSs to use in each D2D communication.
The sD2D-eCMS approach foresees the formation of some D2D clusters. The cluster formation and resource allocation are important factors for the D2D-based multicast performance. This is true in general and not only for the eCMS-sD2D solution specifically presented in the current paper. However, in our previous research on the topic, we have investigated these aspects in detail. Since in this paper these aspects are not in focus, we rely on the results of our previous research. In particular, for the eCMS-sD2D proposal, we implement the best performing solution obtained in Reference [
5]. The resulting configuration of each cluster is a cluster composed of the selected RN and the associated users that it has to serve through D2D communications. Users inside the cluster receive the multicast content by the RN by means of unicast D2D communications, i.e., the RN will transmit the same content to each D2D receiver belonging to its cluster by setting the appropriate MCS to each D2D link.
The multicast and D2D configuration selection is accomplished through an iterative procedure. First of all, the eNB orders the received cellular CQI values from the lowest to the highest. For every CQI value, it determines the subset of UE that can decode data transmitted with the correspondent MCS and the subset of UE, which, on the contrary, are in the worst channel conditions and must be served through D2D communications. Among all the eligible configurations (i.e., those in which all UE, belonging to the MG, receive all data), the eNB finally selects the one that guarantees the maximization of system data rate. Please refer to the D2D-enhanced CMS with single frequency solution presented in [
5] for more details on the implementation of this step.
Once the multicast and D2D configuration is defined, the operations aimed at securing D2D communications are carried out. Security mechanisms described in the following are inspired by [
10].
Step 4—Keys distribution
In order to protect privacy of registered UE, the eNB computes a pseudo-identity (PID) for each of them. For each that registered with the network using its real ID (), the eNB computes a pseudo-identity as: , where is a secure hash function, chosen and published by the eNB. In order to protect the privacy of the users and not increase the total overhead, the eNB never distributes the PIDs to the users, but each of them has to compute its PID autonomously, using the same function of the eNB (i.e., ). Then, through a secure control channel, the eNB sends to the D2D users their private and public keys. It obtains the first key by choosing (i.e., a set of integers with a prime number q of elements) and computes the second key as , where g is the fixed primitive element of used as generator/base. Note that , q, g, , and the symmetric encryption algorithm (the last two will be mentioned in the future steps) are all security parameters chosen and published by the eNB.
Step 5—D2D initialization
After receiving the keys, the D2D receiver (in the following indicated by
) sends an initialization message to the eNB. This message is composed by:
where:
is the index of the portion of data that the user requires to receive. Indeed, too large data can be divided into multiple portions, each identified by a specific index. The eNB keeps track of the portions of data sent to each user to avoid data retransmissions.
is the operator used to concatenate strings;
z is the first public key for generating the secret key that will be used for data encryption and decryption. It is computed as , where is randomly chosen by ;
is the
. Generally, the
is used to guarantee the integrity and authentication of the message
m. It is based on the use of any cryptographic hash function
h applied to a combination of the original message
m and the secret key
k. In Equation (
7),
is the key padded out to size,
opad and
ipad are specified padding constants. In the remainder of the paper,
will be expressed as
, where • denotes the message attached by the HMAC and
k is the secret key hashed together with the message. Note that
is only known by the sender
and the receiver eNB. In all future steps, the verification of the HMAC will always be performed by the recipients of the messages to verify message integrity and authentication, hence, from here on, this procedure will be omitted.
Step 6—D2D pair announcement
After receiving the initialization message, the eNB authenticates the requesting user in the normal cellular communication mode, obtaining its RID and checking if it is registered. In the positive case, the eNB has to inform both D2D devices involved in the direct communication of their imminent communication. Thus, it randomly selects
and computes
as the first public key for generating the secret key
to use in the exchange of private messages with the selected D2D transmitter (i.e., the RN). To communicate to
, which has been chosen as RN of the D2D communication, the eNB sends to it the following message:
Simultaneously, to acknowledge the reception of the initialization message, the eNB sends to
a response message with PID and public key of the selected RN:
Step 7—Data transmission
First of all, the eNB must sign data to send to devices with
:
where
is a secure hash function,
is the private key of the eNB, and
M is data to be transmitted. After that, it performs multicast transmission to users with the best channel conditions, using a Conventional Multicast Scheme (CMS).
, which has received data sent by the eNB, has to forward them to the previously notified D2D receiver. Then, it carries out all the operations aimed at the secure D2D communication. First of all, to allow the receiver to generate the secret key
, it randomly selects
and computes
as the second public key for
. It does not send
y directly to the
, but it sends it to the eNB, randomly choosing
, generating the secret key
and using it to encrypt the public key
y. Then,
must encrypt data, so it generates the communication key
and uses it to encrypt the data
M. After computing
,
signs the message calculating:
where
is the timestamp used against the replay attack. Thus, the secure D2D communication takes place when the RN sends the following message to the receiver:
Step 8—Public key report
In order to allow the eNB to generate the secret key
used to encrypt the public key
y,
computes
as the second public key for
, using
chosen in the previous step. Thus, it sends to the eNB a report:
Step 9—Data check
After receiving data,
first has to verify the identity of the transmitter. To this aim, it compares the
reported on the message received by
with that communicated by the eNB and, if the two do not match, the packet is dropped, otherwise it proceeds with the next steps. Thus, it checks the signature of the transmitter
and, if it is valid, data are considered sent by the entity corresponding to
. Once the identity of the sender is verified,
needs to generate the decryption key
to obtain the plaintext. To do this, it sends a public key request message to the eNB:
After receiving this message, the eNB decrypts the
, first generating the decryption key
, and sends the response message to
:
where
is employed to record the feedback time.
It is important to underline that, in the traditional DHKE algorithm, public keys are exchanged directly between the direct-communicating users. Instead, in sD2D-eCMS, the public keys’ exchange is mediated by the eNB, which represents the trusted third party. This can help to avoid the man-in-the-middle attack.
Thanks to the reception of the public key
y,
can get the communication key by computing
. Thus, it can decrypt the message
to obtain the original data
M. To verify the origin of data, it also checks the signature
and, if it is valid, data are accepted. Otherwise, it is possible that data may have corrupted. In this case,
must send to eNB a
beacon in order to report the fabrication of the original data and to allow it to identify the attacker:
The beacon must be sent within the timestamp , which satisfies the condition , where is the time interval, starting with , where the eNB is willing to wait for feedback from .
Thanks to Equation (
16), the eNB must keep track of any malicious behavior of users. If beacon arrives during the time interval
, the eNB first checks the validity of
and, if it is invalid, it is judged that data are not the original ones and may be fabricated by the transmitter. Thus, the eNB also verifies the validity of
to ensure that the fake message comes from the entity corresponding to
. A malicious behavior amount (MBA) counter is stored by the eNB for each user that does not transmit data correctly in the D2D communication. Thus, in the case of malicious behavior of
, the eNB increments by one its MBA counter. When a user’s counter exceeds a given threshold, called the maliciousness threshold, the user is punished by the network, i.e., it is excluded from future communications. The maliciousness threshold belongs to the interval
. Its value is of utmost importance in the evaluation of the performance of the proposed protocol, as it determines its degree of selectivity. A user
(where
N is the set of registered users) can be chosen as RN only if its MBA value is less than or equal to the maliciousness threshold, i.e., the following condition must be verified:
This means that the value of the threshold represents the maximum number of malicious behaviors tolerated to allow a user playing the role of transmitter (i.e., RN) in D2D communication. When it is set to zero, only users with an MBA value equal to zero can be selected as RN, which means that selecting a malicious user is only possible if the user has never had malicious behavior in the past. As the value of the threshold increases, the algorithm is always less selective; this means that more users that, in the past, have already behaved maliciously can be chosen as RN.
We highlight that the MBA counter is incremented only in the case that a certain malicious behavior is detected (i.e., in case the eNodeB receives a beacon message from a D2D receiver alerting a malicious behavior from RN and the eNodeB verifies that has actually served as a data transmitter). Thus, an increment in the MBA counter can not be due to misdetection of a malicious behavior. Thus, for low values of the maliciousness threshold, the goodput increase is not at the expense of maliciousness misdetection. Setting the maliciousness threshold to a value higher than zero (i.e., choosing a not too selective approach in the relay election procedure) could be useful in case a node turns its malicious nature to non-malicious. In fact, in case a node stops exhibiting a malicious behavior, it will no longer be selected anyway as a relay node in case the maliciousness threshold with a goodput increase that can be relevant in case the node is in good channel conditions and can serve as relay for a high number of neighbors.
5. Performance Evaluation
A simulative analysis in MATLAB (MathWorks, Natick, MA, USA) has been performed to study the performance of the proposed eCMS-sD2D protocol. The considered scenario consists of a variable number of users uniformly distributed in a single LTE-A cell of dimensions 100 m × 100 m, with available bandwidth of 20 MHz, which corresponds to 100 RBs. A Time Division Duplex (TDD) LTE frame type 2 configuration 3 is used. Each slot (or Transmission Time Interval, TTI) in the frame lasts 1 ms, so the entire frame has a duration of 10 ms. The inband D2D mode is chosen, so uplink slots are reserved to D2D communications. In the downlink slots, a multicast transmission allows for sending data to in-coverage devices.
The performance of the proposed eCMS-sD2D protocol is evaluated on the basis of the following metrics:
Data loss in D2D communications because of an unreliable transmitter;
Mean Throughput, measured as the mean data rate value experienced by the D2D receivers. It is independent of the nature of the communication transmitter (i.e., if it is malicious or not). In particular, even if the data transmitter is a malicious user, the amount of data it delivers is counted in the throughput computation;
Mean Goodput, measured as the mean useful data rate value experienced by the D2D receivers (i.e., rate of data that has been forwarded by non malicious transmitters). This metric takes into account the reputation of the user. In fact, if an RN behaves maliciously, the amount of data that it delivers does not cause a goodput increase;
Aggregate Data Rate (ADR), computed as the sum of the data rates experienced by the D2D receivers;
Good Aggregate Data Rate (ADR), computed as the sum of the useful data rates experienced by the D2D receivers;
Mean number of malicious nodes accounting for malicious D2D receivers that have been served and selected relay nodes;
Mean number of malicious relays that have been selected to transmit data in D2D communications.
In the following simulation results, we will compare our proposed eCMS-sD2D with respect to the existing protocol presented in [
5], in which the security of D2D communications is not guaranteed. Under eCMS-sD2D, we analyze its performance with three different values of the maliciousness threshold (equal to 0, 40, and 80). We assign to each node a nature (malicious or non-malicious) and we assume that it does not change over time. Thus, our analysis can be seen as a sort of worst case analysis.
The results in terms of mean throughput and goodput are presented in
Figure 3a,b, respectively, under increasing percentage of malicious UE. By looking at
Figure 3a, we can notice that curves are almost overlapping because throughput does not depend on security. In fact, D2D-SF and our proposed eCMS-sD2D achieved really similar throughput values and the experimented data rate is independent from the percentage of malicious users. Differently, the results in terms of goodput depicted in
Figure 3b are substantially different. In fact, D2D-SF exhibits significantly lower performance with respect to eCMS-sD2D that achieves goodput values as close to the throughput as the maliciousness threshold is lower. In detail, as the threshold increases, eCMS-sD2D becomes less selective, therefore more similar to D2D-SF, in which no selection is applied in the choice of RNs. On the contrary, when the maliciousness threshold is set to zero, eCMS-sD2D shows the better results since nodes that behaved maliciously just once are also not selected as RN.
A similar reasoning holds for the ADR metric. In fact, we can appreciate in
Figure 4 that, while the ADR is almost identical for eCMS-sD2D (under all maliciousness thresholds under analysis) and D2D-SF, the good ADR is significantly lower than ADR when no security mechanism is implemented (see the last pair of bars). Furthermore, we highlight that the increment of the maliciousness threshold causes a performance degradation since security mechanisms are less stringent.
This significant degradation of the goodput/good ADR is due to the high amount of data loss, as confirmed by
Figure 5. Thanks to the implementation of the security mechanisms foreseen in eCMS-sD2D, data loss can be reduced by about 70%. In particular, when the maliciousness threshold is set to zero, thanks to the better selection of RNs, eCMS-sD2D shows the best results. Nevertheless, data loss is not zero, due to the fact that malicious RN can be identified only after the eNodeB detects an incorrect behavior. As the threshold value increases, the efficiency of eCMS-sD2D in terms of selection of RNs is always lower, so the results are increasingly similar to D2D-SF.
Because of security, as shown in
Figure 6, the mean number of malicious D2D UE that gains service is reduced (see the first bar with respect to the fourth bar). However, despite this metric showing an improvement of about 10%, it takes into account both malicious users served by the CMS transmission (on which the security mechanisms do not have impact) and selected malicious RNs. This is the reason why the performance improvement of eCMS-sD2D with respect to eCMS-D2D is limited. Going more in depth in the analysis, we show in
Figure 7 the mean number of selected malicious RNs under increasing percentage of malicious users. We can appreciate that the reduction in the number of selected malicious RNs is of more than the 50% also in case of a high number of malicious nodes (75% of malicious nodes).