Resource Allocation for 5G D2D Multicast Content ... - IEEE Xplore

POINT-TO-MULTIPOINT COMMUNICATIONS AND BROADCASTING IN 5G

Resource Allocation for 5G D2D Multicast Content Sharing in Social-Aware Cellular Networks Lei Feng, Pan Zhao, Fanqin Zhou, Mengjun Yin, Peng Yu, Wenjing Li, and Xuesong Qiu The authors present a D2D multicast 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. Simulation results validate the significant throughput gain for D2MD-based content sharing in 5G cellular networks.

Abstract As the traffic of local content sharing grows rapidly, device-to-device multicast (D2MD) is introduced into 5G cellular networks, enabling traffic offloading from base stations to device direct transmissions and improved energy and spectrum efficiency. In the content sharing context, the social attributes of mobile users have special concerns to generate effective D2MD groups and D2MD links, and due to reuse of licensed cellular spectrum, the interference between D2MD groups and cellular users should be carefully mitigated. This article presents a 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. Simulation results validate the significant throughput gain for D2MD-based content sharing in 5G cellular networks.

Introduction

Due to the interest-sharing nature of human beings, it is quite common for people to share interests with others in their close vicinity. The rapid evolution of mobile communication technologies, the proliferation of smart devices (phones, tablets, laptops, etc.) and mobile applications, such as WeChat, Facebook, and Dropbox, make it fast and convenient for mobile users to generate and disseminate various types of contents [1], which pushes the sharing habits online. It is noticeable that people are sharing almost everything anytime and anywhere with their familiar groups nearby or even strangers. Because of the pervasive wireless access in cellular networks and currently available tariff plans for unlimited bandwidth, mobile users are becoming more inclined to share the content in which they are interested through online-based approaches over cellular networks. Figure 1 illustrates some content sharing cases in our daily life. In the figure, for example: • Before classes, teachers distribute course materials to students. • At a sports event, excited fans are keen to share their snapped video clips to their peers. • In the workplace, pieces of collaborative work items need to stay synchronous among team members. Digital Object Identifier: 10.1109/MCOM.2018.1700667

112

• Police share traffic control information. • In the body area, emerging wearable devices call for constant data sharing. According to whether the content to be shared is stored on the local device or not, content sharing can be categorized into two cases, local-direct sharing and local-request sharing. In the former case, the disseminator has the content locally and is able to share it directly with others. In the latter case, a member of a content-sharing group just initiates the sharing task, and a nearby base station (BS) has to help retrieve the requested content from a remote server and delivering it to other group members. Conventionally, in both cases, content sharing in cellular networks is performed through online approaches, and a BS nearby has to push the shared content to target receivers either through different unicast transmissions or through multicast transmission. In the unicast case, dedicated unicast links established between the BS and the target receivers cause a waste of access network bandwidth. The duplicated transmissions pose redundant traffic load to the BS in a cellular network. By leveraging the broadcasting nature of wireless transmission, multicast delivers the shared content to multiple receivers simultaneously. The waste of bandwidth and redundant traffic load produced in the unicast case can be alleviated. However, multicast links from BS to mobile devices are severely limited by the worst channel users, while due to limited licensed spectrum and the continuously increasing content sharing demands, cellular access networks would suffer congestion in the near future [2]. To better achieve content sharing in cellular networks, endeavors are made on various aspects, and the main directions are discussed in the following section.

Approaches to Improve Content Sharing in Cellular Networks

The location of shared contents, the length of transmit paths from disseminator to receivers, and the number of simultaneous transmissions really count in deciding the efficiency of content sharing. Generally, the shared contents are often popular contents viewed and saved by a disseminator in advance. There is usually more than one receiver, and the members in a content sharing group are possibly in close proximity. By exploiting these

The authors are with Beijing University of Posts and Telecommunications.

0163-6804/18/$25.00 © 2018 IEEE

IEEE Communications Magazine • March 2018

D2MD transmission Company

can efficiently balance

High-density environment

5G network

the traffic load of central BSs, leading to improved spectral efficiency, since content

(c)

Campus

sharing traffic usually shows significant

(d)

redundancy caused by, for example, a single (a)

(e)

Shopping center

hot TV show/drama or sports highlights content requested by multiple users at the same time.

(b) Stadium

D2D link Cellular link

Figure 1. Illustration of different applications of device-to-device multicast content sharing. intrinsic properties, a number of approaches are designed for more effective content sharing, and these approaches can be categorized into two main directions. Get content to users as close as possible: In this direction, different approaches are trying to cache or store shared content at the network edge, for example, the caching entity in BS, or even other nearby devices owned by other mobile users. The heterogeneous network also contributes much in this direction, because the intensively deployed low-power nodes go much closer to mobile users.With a deliberately designed content selection and caching strategy, the greatly shortened transmission paths benefit content sharing a lot, especially in the local-request sharing case. Use short-range communications: With shared contents stored in user devices, taking the local-direct sharing as an example, short-range communication technologies such as WiFi Direct, Bluetooth, ZigBee, and cellular device-to-device (D2D) can be used to perform device-direct communications. They reduce the reliance of content sharing procedures on cellular base stations, offload traffic from cellular networks, and save radio resources. Mobile users can also benefit from device-direct approaches by enjoying much lower transmission delay and energy consumption due to the short communication range. Intuitively, utilizing device-direct transmission via actively established user WiFi Direct, Bluetooth, or ZigBee links would be a welcome option for both mobile users and network operators in the proximity-based sharing scenario. However, these methods are less flexible, as tedious user-involved configuration and coordination are required before actual transmissions [3]. Besides, these approaches are mainly for local point-to-


point communication, so they are very applicable in the local direct sharing case, but help little with local request sharing. With affordable tariff plans available for unlimited bandwidth in mobile networks and 4K videos becoming popular, the local-request sharing case will be more common than ever. At the same time, local-direct sharing is still in great demand. These call for a new paradigm for content sharing that supports device-direct transmission and at the same time is interoperable with mobile networks. Cellular D2D communication makes user devices establish data connections with each other directly in the licensed spectrum, thus offloading the traffic generated from BS-to-user links. Meanwhile, D2D communication is under the control of a cellular BS. A D2D transmitter would retrieve remote files from online servers and at the same time retransmit the retrieved pieces of data direct to a D2D receiver [4]. For our focused content sharing case in which a group of users in geographically close proximity request the same content, it is reasonable to leverage the broadcasting nature of wireless transmissions by pushing or sharing the same data to requesters from the local (original or temporary) content holder. D2D multicast (D2MD) transmission is thus proposed. D2MD transmission can efficiently balance the traffic load of central BSs, leading to improved spectral efficiency, since content sharing traffic usually shows significant redundancy caused by, for example, a single hot TV show/drama or sports highlights content requested by multiple users at the same time. Exploiting proximity direct communication, D2MD transmission is expected to offer a number of impressive advantages including higher energy efficiency, shorter latency, offloading traffic, and improved user experience.

113

Social domain

Community B2

Community B3 Community B1

C1

D2 f1&f2

D3 f4

D1 f3

C2

Data signal

C4

Mapping Social relation relation

Physical domain

C3

Interference signal

Figure 2. Factors in the social and physical domains affect the formation of D2D multicast clusters.

D2D Multicast Social-Aware Content Sharing

In practical networks, certain individuals have strong social relationships, including neighbors, co-workers, classmates, and so on. They are not only trustful for each other, but easily meet data transmission requirements of short-range communications. When some of them have the desired contents, they are willing to share contents with others in a D2D manner. However, people unfamiliar with each other are seldom interested in sharing anything. In addition, mobile users will be discouraged from using D2D communication if they perceive poor D2(M)D sharing services, which is probably caused by overly long communication distance or strong interference. From these observations, we know that both social and physical factors should be taken into account for forming D2MD groups. Aiming to ensure the efficacy of D2MD, the following characteristics are considered in D2MD content sharing. An illustration of these characteristics is given in Fig. 2. Community and Social Ties: In a content sharing context, communities are used to identify different content sharing groups. As the precondition for allocating resources, community information can help form effective D2MD groups. Social ties reflect the strength of connections among mobile users, which can be built between humans through friendship, kinship, colleague relationships, and so on. In general, direct communication links are more likely to be established between users with strong social ties. It should be noted that all members in a community are not necessarily D2D users, and some users will be still served by cellular BSs due to physical and tie constraints. Physical Distance and Interference: Since D2D communication has short distance requirements, not all users with high social ties can build

114

D2D links. The physical distance of a peer-to-peer link is also important to establish a D2D connection. Besides, D2D transmitters reuse the cellular spectrum to deliver content to the receivers. Thus, efficient resource assignment approaches are required to mitigate the impact of co-channel interference and ensure quality of service (QoS). Hence, knowledge of information for both the physical and social domains strongly affects the forming of D2MD groups, D2D content provider selection, and resource allocation, which should be fully leveraged to establish stable and reliable D2D multicast links, and effective resource allocation.

Related Work

There are mainly two procedures in D2MD content sharing, D2D cluster discovery and content delivery. Existing work related to the first procedure mainly focus on issues (e.g., reasonable D2MD cluster formation and optimal transmitter selection). In content delivery, resource allocation and interference are the main concerns. Forming appropriate clusters can improve D2MD performance in D2D underlay cellular networks. Factors like physical position, social attributes, and content preference impose significant impacts on D2MD cluster formation. In [5], physical distance played a pivot role in cluster formation for D2D and other short-range technologies. Interests of users in specific mobile applications are utilized in [6] to realize D2D discovery. The joint effect of physical distance and social attributes were considered in [7]. Cluster head selection also affects the performance of cluster formation and multicast performance. According to existing works, an elected cluster head can be e.g., content holder [8], user with the most potential D2D links [9]. The effect of users’ willingness to perform D2D communication on cluster head selection was investigated in [10]. Since D2D content delivery is achieved through reusing cellular spectral resources, the resource allocation should be carefully addressed to mitigate the co-channel interference and guarantee QoS. For instance, to optimally coordinate co-channel interference, a channel allocation approach was proposed in [11]. In addition, more and more research works pay attention to socialaware D2D communication networks. In [12], C. Xu et al. exploited D2D communication for social content delivery through matching theory, and a two-stage coalition gaming was proposed in [13] to deal with social-aware D2D resource allocation. Existing research works on resource allocation in D2D underlay cellular networks mainly elaborate on D2D pairs. A few of them are interested in D2MD, but with seldom or insufficient social-factor considerations. Moreover, existing work commonly adopts the single channel reuse mode, and seldom considers assigning multiple channels for individual D2DM clusters, with the fairness issue of allocated resources to each cluster still open. After an overview of background and related work of content sharing in D2D, the rest of the article aims to present an approach utilizing both physical and social information to build efficacious D2D links and form D2MD clusters. Thus,


immense content sharing traffic can be offloaded from BS to D2MD clusters. To boost the performance of D2MD content sharing in social-aware cellular networks, the article further and mainly proposes a resource allocation scheme to cope with interference caused by channel reusing via power control and channel assignment.

System Model

Let us consider a typical single-cell scenario consisting of one BS and multiple user equipments (UEs) in Fig. 2. UEs interested in the same content form a content sharing community, B 1. In each community, some UEs may have strong social relationships and appear in close proximity, so they are willing and able to share files in a device-direct manner. The BS guides the process, making UEs in a community form D2MD clusters (a D2D pair is a special case of D2MD cluster, and we see them without discrimination). UEs are then in two types, cellular UEs (CUs) and D2D UEs (DUs). For example, in community B1, UEs that grouped into D2MD cluster D1 and D2D pair D3 are DUs, and cellular UE C1 is a CU. We assume CUs occupy equal pieces of the licensed spectrum. Due to the exclusive resource occupation among CUs, we see these pieces the same as channels. In a D2MD cluster, the transmitter reuses some channels to deliver a shared file. This causes interference to CUs using the same channels. For example, D1 can use channels occupied by C1, C2, C3, and C4. If D1 reuses channels of C2, great interference will be incurred by the C2–BS link. To mitigate the interference, D1 has to lower the transmit power or choose better channels (e.g. C1s and C3s). Power control and channel assignment will both alleviate interference between CU and DU. Besides, D2MD clusters call for fair resource allocation to ensure the uniform service perception of mobile users. In the following part, we first design the models to characterize the social and physical factors; for utilizing them, we propose the D2MD cluster formation scheme and resource allocation scheme.

Modeling Social and Physical Features

Interested in similar contents, people share their interests and frequently contact each other within a group; a community therefore comes into being. In the content sharing context, we suppose there are K users in a community, and a symmetric matrix S = {s mn} KK is used to represent the strength of social ties between UEs. The element smn, ranging from 0 to 1, represents the social tie strength between} UE m and UE n. By taking into consideration the security and privacy issues, a threshold sth is set, and only when the social tie strength smn is bigger than sth are UE m and UE n mutually trusted enough to establish a D2D link; if not, no D2D link is established. Since D2D communication has a short distance requirement, not all users with strong social ties can build D2D links. In the physical domain, whether a D2D link is built between two UEs is decided by their physical distance. We use a symmetric matrix L = {lmn}KK to indicate the existence of physical connections between UEs in a community, where lmn indicates the existence of the physical connection between UE m and n, and its


value is decided by the following observations. If the physical distance is below a certain threshold lth for direct communication, lmn = 1; otherwise, lmn = 0.

Forming D2D Multicast Clusters

In this section, we utilize both the social domain and physical domain knowledge presented above in a joint manner to form effective D2MD clusters for content sharing, considering both the local-direct sharing and local-request sharing cases. The main difference in these two cases lies in the selection of a D2MD cluster head, as in local-direct sharing the content holder is more likely to be selected, while in the local-request sharing, a UE with more potential D2D connections is more suitable for selection to perform the D2D multicast content sharing. We identify social communities by the content interest shared among the members within each community. In a community of total K UEs, a symmetric matrix E = {emn}KK can be obtained by letting emn = dmn ⋅ smn ⋅ lmn, where dmn ranges from 0 to 1, and represents the ratio of the amount of UE n desired content stored in UE m. Here, we use a trick to avoid the all-zero instance of E by setting an arbitrary small e > 0 as the sublimit of dm. Thus, matrix E reflects not only the joint effect of physical and social factors, but also the contributing effect of existing contents. Matrix E is useful in selecting cluster heads and forming D2DM clusters. The sum of elements in the mth row stands for the total social closeness degree of user m to those in the community able to build D2D links with UE m. A larger sum implies that more efficacious D2D links can be established, and the corresponding UE has priority to be selected as a D2MD cluster. After calculating the sum of each row of E, we select the UE with the largest sum as a cluster head and UEs able to build efficacious D2D links to it as its peer members. These UEs form a D2MD cluster and appear in no other cluster. Thus, we make the corresponding rows and columns of E obsolete, and repeating this process until no such clusters can be formed, we get other D2MD clusters.

In a content sharing context, communities are used to identify different content sharing groups. As the precondition for allocating resources, community information can help form effective D2MD groups. Social ties reflect the strength of connections among mobile users, which can be built between humans through friendship, kinship, colleague relationships, and so on.

Allocating Power and Channel Resources

Our aim is to design a resource allocation method to get a maximal transmission rate for D2D multicast content sharing in a social-aware cellular network. For example, there are two local video sharing communities, B1 and B 2, served by the same BS. Even though they have different multicast spectrum efficiency caused by varied wireless channel conditions, the BS will try its best to guarantee a uniform service of content sharing for both B1 and B2 while pursuing the largest total cell throughput. This can be accomplished via power control and channel assignment.

Power Control via Geometric Programming

Aiming at throughput improvement, it is important to allocate reasonable power to cellular and D2D users. As shown in Fig. 2, D1 reuses resources of C 3, and their simultaneous transmission will cause interference to each other. Lowering

115

their transmit power can alleviate their mutual interference; however, it brings about lower received power, and further reduces signal-to-interference-plus-noise ratio (SINR) and transmission rate. To ensure QoS of both D1 and C3, the SINR at the target multicast receivers in D1 and BS should be above certain thresholds. Besides, the transmit power of UE is always constrained by the maximal transmit power. Thus, the optimized power for CUs and DUs that maximizes the overall cell throughput is able to be found in a feasible region, which is determined by the constraints of maximal Tx power and minimum SINR for D2D and cellular users, respectively. We use geometric programming to find the optimal solution in the feasible region. As Fig. 3 shows, the feasible region, namely the shaded area, varies obviously due to different channel ld Pcij

lc

Pcmax

ld

Pcij Pcmax

0

PDmax Pdij

Channel Assignment via Bipartite Matching

lc

Y1

Y0

0

Y3

Y2

PDmax Pdij

(a)

(c)

Pcij Pcmax

Y4

Y0

0

PDmax Pdij

(b)

Y2

lc

Y1

0

lc

Pcmax

Y5 Y0

ld

Pcij

ld

PDmax Pdij

(d)

Figure 3. Possible shapes of feasible regions under different wireless channel conditions. PCmax and PDmax stand for the maximum transmit power of CU and DU, respectively. lc and ld represent the SINR constraints for the transmit power of CU and DU.

SD

N

Ri,j

Channel

SC

SD

D1 F2 F4

F1

C1

D2

F2

C2

D3

F3

C3

D1

F4

C4

D2

F5

D3

F6

∂N

(a)

C5 C6 M

N M M=∂N

Ri,j

Channel

SC

F1

C1

D2

F2

C2

D3

F3

C3

F4

C4

F5

C5

F6

C6 N-h M+N-h

D1

F5

D1 D2

R1,5

R1,5

D3 (∂-1)N

(b)

M M=∂N+h

Figure 4. Examples of extended bipartite graphs of D2D clusters and CUs in different cases: a) in the M –= ∂N case, add N (∂ – 1) virtual vertices in the set – SD with in the M = ∂N + h case, add N – h virtual vertices in the set SC with zero weighted edges.

116

gains. Lines ld and lc represent the SINR thresholds of a CU and the corresponding DU, respectively. On the right side of ld, the D2MD group meets its minimum SINR requirement, while the CU meets its minimum SINR in the region over lc. The square area represents the feasible region of transmit power of the CU and D2MD transmitter. According to [14], it is known that either a CU or its corresponding D2MD transmitter has optimal transmit power equal to the maximum power; then the optimal power pair for CU Ci and D2MD cluster Dj resides on one of the corner points of the feasible region (i.e. Y 1, Y 2, Y 3, Y 4, and Y 5 in Fig. 3). With at most four points to calculate, we can quickly get the optimal power allocation for a given channel reuse pair, such as (C1, D1), where D1 reuses C1’s channel. Similarly, we can easily get the optimal power allocation for other channel reuse pairs. Moreover, the optimal throughput increment corresponding to the optimal power is obtained according to path loss and Shannon’s theorem. Exploiting the power control results, we are capable of getting all feasible mappings of resource reuse between a D2MD cluster and the corresponding cellular users. As one D2MD cluster is allowed to reuse the resources of multiple CUs, but one CU’s resources are reused exclusively by one D2MD cluster, it is necessary to search all sets of feasible mappings for the optimal channel allocation scheme. To obtain the optimal solution, the maximal weighted bipartite matching algorithm developed in graph theory provides a reasonable way. As shown in Fig. 4, bipartite graph G is constructed where two sets of vertices SD and SC represent the group of D2MD clusters and that of CUs, respectively. If Ci is a candidate resource provider of D2MD cluster D j, the edge is established between the corresponding vertices in G having the weight DRi,j. Otherwise, there is a zero-weight edge. To solve the bipartite matching problems of this type, the Hungarian algorithm is usually adopted. However, it requires symmetric bipartite graphs. To make the Hungarian algorithm applicable, the original non-symmetric bipartite graph G is turned into a symmetric one, as illustrated in Fig. 4. The advantage brought by this transform is that we can minimize the gap between allocated resources for each D2MD cluster due to the duplicates of vertices in S D . After this transform, the Hungarian algorithm can be directly employed to derive the solution for the original problem. For instance, in Fig. 4 a cluster reuses C 2’s channel F 2 with maximum weight gain, while the duplicate of D1 reuses F4. Hence, the optimal choice for D1 is to reuse F2 and F4. However, it cannot guarantee that other clusters do not reuse the same channels. Hence, another key point of the Hungarian algorithm is to avoid such collisions. In each matching iteration, we either add an edge so that the objective value increases, or replace an existing edge to obtain a matching with more edges. It terminates when the perfect matching is found. Therefore, we can get the optimal solution for channel assignment with guaranteed resource fairness among different D2MD clusters.


Illustrative Results

Conclusion

In this article, we investigate the background and research issues of D2MD content sharing in social-aware cellular networks, and present our D2MD content sharing scheme. In our scheme, both physical and social domain factors are


100

s = 0.3, M s = 0.5, M s = 0.7, M s = 0.3, N s = 0.5, N s = 0.7, N

90

The number of CUs or D2D clusters

80 70 60 50 40 30 20 10 0

10

30

20

40 50 60 70 D2D physical proximity threshold (m)

80

90

100

Figure 5. The effects of physical proximity threshold on the cluster results. 12,000

s = 0.3 s = 0.5 s = 0.7 Before

11,000

The D2D clusters throughput (Mb/s)

We consider a single cell case in a 0.5  0.5 km simulation area with the macro station surrounded by three communities without position overlapping. We distribute the earliest users randomly. The social tie matrix is set up in each community for the users. The social closeness degree between users is generated randomly within the range 0 and 1.As to other physical parameters, the maximum transmit power of BS and UE is 42 dBm and 24 dBm, respectively. A simplified path loss model is used to characterize cellular links by 128.1 + 37.6 log(d[km]) and D2D side links by 148 + 40log(d[km]), individually. All CUs identically share a total of 5 MHz spectrum and reach minimal QoS prior to resource allocation. Since the primary goal of forming D2MD clusters is to push as many users into D2D communication as possible, we set the number of CUs M and that of D2MD clusters N as metrics to evaluate clustering performance with regard to the thresholds. d stands for the physical proximity threshold and s for the social closeness threshold. As resource allocation is mainly to improve the overall throughput performance, throughput increment is adopted. Figure 5 shows the effects of physical proximity threshold d on the cluster formation results when the social closeness threshold s is 0.3, 0.5, and 0.7, respectively. The number of CUs M decreases as d increases, and smaller s produces fewer CUs. This is consistent with our intuition. Smaller s allows D2D links to be established for users with weaker social closeness, thereby making more UEs go into D2D mode. Both s and d affect the cluster forming with varied importance. When s = 0.7, N increases rapidly because high social closeness requirement strictly limits cluster sizes, and more UEs become potential D2D users as d increases. It is not hard to see that the N curves first stretch up and then down when s = 0.3 and 0.5. Figure 6 depicts the overall throughput increments, compared to the initial system throughput of about 5100 Mb/s without D2D communication, for different d when s is 0.3, 0.5, and 0.7, respectively. In the figure, throughput increments in all three cases go up as the physical proximity threshold rises because larger d means more original users turn into D2D users, and cluster size also grows. As a result, more spectrum resources can be reused by each D2MD cluster, resulting in larger spectrum multiplexing gain. These facts jointly lead to system throughput improvement. Moreover, social closeness threshold s also affects throughput improvement, and s progressively dominates the constraining effect on throughput when d gets larger. Therefore, the margin between small s and large s gets larger with the increase of d, and the throughput gain of smaller s is always larger using the same d.

10,000 9000 8000 7000 6000 5000 4000

0

20

30

40 50 60 D2D physical proximity threshold (m)

70

80

90

Figure 6. The effects of physical proximity threshold on the overall throughput. exploited to produce efficacious clusters, and geometry programming and bipartite matching are utilized to achieve optimal power control and channel assignment for shared content delivery. Considerable throughput increments can be observed in the simulation results. In the future, as content retrieval is usually resource and time consuming, it is of great interest to consider content caching and sharing mode selection in D2MD content sharing. Moreover, to take full advantage of D2MD communication, incentive mechanisms to encourage mobile users to share contents through D2MD deserve more attention.

Acknowledgment

This work has been supported by the 863 Program (2015AA01A705) and the National Natural Science Foundation of China (61271187).

117

References

[1] Y. Wang et al., “A Locality-Based Mobile Caching Policy for D2D-Based Content Sharing Network,” IEEE GLOBECOM, 2016, pp. 1–6. [2] Y. Zhao et al., “Survey on Social-Aware Data Dissemination over Mobile Wireless Networks,” IEEE Access, vol. 5, 2017, pp. 6049–59. [3] D. J. Dubois et al., “Supporting Heterogeneous Networks and Pervasive Storage in Mobile Content-Sharing Middleware,” IEEE CCNC, 2015, pp. 841–47. [4] Y. Cao et al., “Social-Aware Video Multicast Based on Device-to-Device Communications,” IEEE Trans. Mobile Comp., vol. 15, no. 6, 2016, pp. 1528–39. [5] L. Militano et al., “Wi-Fi Cooperation or D2D-Based Multicast Content Distribution in LTE-A: A Comparative Analysis,” IEEE ICC Wksps., 2014, pp. 296–301. [6] K. W. Choi et al., “Discovering Mobile Applications in Cellular Device-to-Device Communications: Hash Function and Bloom Filter-Based Approach,” IEEE Trans. Mobile Comp., vol. 15, no. 2, 2016, pp. 336–49. [7] J. Liu et al., “Modeling Multicast Group in Wireless Social Networks: A Combination of Geographic and Non-Geographic Perspective,” IEEE Trans. Wireless Comp., vol. 16, no. 6, 2017, pp. 4023–37. [8] B. P. et al., “Cluster-Based Multicast Transmission for Deviceto-Device (d2d) Communication,” VTC-Fall, Sept. 2013, pp. 1–5. [9] G. Z. et al., “Socially Aware Cluster Formation and Radio Resource Allocation in d2d Networks,” IEEE Wireless Commun., vol. 23, no. 4, Aug, 2016, pp. 68–73. [10] L. Wang et al., “Device-to-Device Users Clustering Based on Physical and Social Characteristics,” SAGE IJDSN, vol. 11, 2015, pp. 1–14. [11] A. Bhardwaj et al., “A Resource Allocation Scheme for Device-to-Device Multicast in Cellular Networks,” IEEE PIMRC, 2015, pp. 1498–1502. [12] C. X. et al., “Social Network-Based Content Delivery in Device-to-Device Underlay Cellular Networks Using Matching Theory,” IEEE Access, vol. 5, 2017, pp. 924– 37. [13] F. Wang et al., “Social-Community-Aware Resource Allocation for D2D Communications Underlaying Cellular Networks,” IEEE Wireless Commun. Lett., vol. 4, no. 3, 2015, pp. 293–96.

118

[14] P. Zhao et al., “A Social-Aware Resource Allocation for 5G Device-to-Device Multicast Communication,” IEEE Access, vol. 14, no. 12, 2017, pp. 6621–34.

Biographies

Lei Feng obtained his B.Eng. and Ph.D. degrees in communication and information systems from Beijing University of Posts and Telecommunications (BUPT) in 2009 and 2015. He is a lecturer in the State Key Laboratory of Networking and Switching Technology, BUPT. His research interests are resources management in wireless networks and smart grid. Pan Zhao obtained her B.Eng. and M.S. degrees from Henan Polytechnic University in 2009 and Chongqing University of Posts and Telecommunications in 2012, respectively. She is working toward a Ph.D. degree at BUPT. Her research interests include resource management in D2D and heterogeneous networks. Fanqin Zhou obtained his B.Eng. degree from BUPT in 2012. He is working toward a Ph.D. degree at BUPT. His research interests include resource allocation and mobility load balancing in multi-RAT heterogeneous networks. Mengjun Yin obtained her B.Eng. degree from BUPT in 2013. She is working toward a Ph.D. degree at BUPT. Her research interests are wireless network management and automatic healing in SONs. Peng Yu obtained his B.Eng. and Ph.D. degrees from BUPT in 2008 and 2013, respectively. He is an associate professor at the State Key Laboratory of Networking and Switching Technology, BUPT. His research interests are autonomic management and hybrid energy allocation in GreenNet. Wenjing Li s a professor at BUPT and serves as a director in the Key Laboratory of Network Management Rresearch Center. Meanwhile, she is the leader of TC7/WG1 in the China Communications Standards Association (CCSA). Her research interests are wireless network management and automatic healing in SONs. Xuesong Qiu is a professor at BUPT and serves as the vice-president of the Institute of Network Technology (INT) at BUPT. He has hosted a series of state research projects on network management. He has earned more than 13 China state-level and provincial and ministerial-level science and technology prizes.
