5 Device-to-device (D2D)

communications
Zexian Li, Fernando Sanchez Moya, Gabor Fodor, Jose Mairton B. Da Silva Jr.,
and Konstantinos Koufos

Direct Device-to-Device (D2D) communication, which refers to direct communication

between devices (i.e. users) without data traffic going through any infrastructure node,
has been widely foreseen to be an important cornerstone to improve system performance
and support new services beyond 2020 in the future fifth generation (5G) system. In
general, the benefits resulting from D2D operation include, among others, highly
increased spectral efficiency, improved typical user data rate and capacity per area,
extended coverage, reduced latency, and enhanced cost and power efficiency. These
benefits are resulting from the proximity of the users employing D2D communication
(proximity gain), an increased spatial reuse of time and frequency resources (reuse gain)
and from using a single link in the D2D mode rather than using both an uplink and a
downlink resource when communicating via the base station in the cellular mode (hop
gain). The chapter starts with an overview of the fourth generation (4G) D2D develop-
ment. Afterward, the challenges to be addressed in the context of 5G D2D and related key
enablers are discussed. In particular, this chapter covers Radio Resource Management
(RRM) for mobile broadband applications, multi-hop D2D communication, especially for
public safety and emergency services, and multi-operator D2D communication.

5.1 D2D: from 4G to 5G

In the future 5G system, it is predicted that network-controlled direct D2D commu-

nication offers the opportunity for local management of short-distance communica-
tion links and allows separating local traffic from the global network (i.e. local traffic
offloading). By doing this, it will not only remove the load burden on the backhaul
and core network caused by data transfer and related signaling, but also reduce the
necessary effort for managing traffic at central network nodes. Direct D2D commu-
nication therefore extends the idea of distributed network management by incorpor-
ating the end devices into the network management concept. In this way, the wireless
user device with D2D capability can have a dual role: either acting as an infrastructure
node and/or as an end-user device in a similar way as a traditional device.

5G Mobile and Wireless Communications Technology, ed. A. Osseiran, J. F. Monserrat, and P. Marsch.
Published by Cambridge University Press. © Cambridge University Press 2016.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 108 Zexian Li et al.

Relaying
Discovery (Range extension)

Single or multi-hop Local data

local proximity sharing
communication (offloading)

Figure 5.1 Typical use cases of D2D communication in cellular networks.

Further, direct D2D facilitates low-latency communication due to the local commu-
nication link between users in proximity. In fact, direct D2D has been seen as one of
the necessary features to support real-time services in the future 5G system [1][2].
Another important aspect is reliability, where an additional D2D link can be employed
to increase reliability through a larger extent of diversity. Moreover, due to the short-
distance transmission, the device power consumption can be reduced significantly.
Figure 5.1 illustrates typical use cases of D2D communication. A more detailed
discussion on different 5G use cases can be found in Chapter 2. Four D2D scenarios
are shown. The first one is about local data sharing where data caching in one device
can be shared with other devices in proximity. In the second scenario, called relaying,
D2D communication can play a key role to improve network availability (i.e. to extend
the coverage area) via a D2D based relay. This is especially important for the use cases
related to public safety and those including both indoor and outdoor users. The third
scenario, called single or multi-hop local proximity communication, is the one con-
sidered in the 3rd Generation Partnership Project (3GPP) Release 12. In this scenario,
the devices within proximity can set up a peer-to-peer link or multicast link that does
not use the cellular network infrastructure. One of the particular applications is the
public safety service. The last scenario is D2D discovery (considered in 3GPP Release
12 as well), which refers to a process that identifies whether a UE is in proximity of
another UE.
Considering D2D air interface design, it is usually assumed that the air interface for
D2D communication is derived from the cellular air interface in order to simplify the
design and implementation. For example in 3GPP Release 12, Single-Carrier Frequency
Division Multiple Access (SC-FDMA) based D2D signaling is employed for all data-
carrying physical channels, and the structure of the Physical Uplink Shared CHannel
(PUSCH), as defined in 3GPP, is re-used (with limited changes) for the D2D communica-
tion channel as well. Regarding spectrum usage, D2D can operate, depending on the
scenario, in licensed spectrum and/or unlicensed spectrum.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 109

Table 5.1 D2D scope in LTE Releases 12 and 13.

Within LTE network coverage Outside LTE network coverage

Discovery Non-public safety & public safety requirements Public safety

Direct Communication At least public safety requirements Public safety

When talking about cellular network-controlled D2D, it is necessary to mention the

standardization progress especially in 3GPP on Long Term Evolution (LTE) D2D (also
known as ProSe: Proximity Services). It is worthwhile to note that Wi-Fi Direct and Wi-
Fi Aware are relevant as well, although they are not addressed here since the focus of this
chapter is on cellular technology-based D2D.
In the following, the current D2D development in 4G LTE is examined. Thereafter, the
5G D2D concept is introduced in order to have a full picture on D2D concept development.

5.1.1 D2D standardization: 4G LTE D2D

Although, in principle, D2D can offer various promising benefits as discussed previously,
in 3GPP LTE D2D work, the main driver is public safety in Releases 12 and 13 [3]. In
addition, commercial discovery is supported as well, as can be seen from Table 5.1.
LTE D2D can be seen as an add-on feature in a 4G LTE system, hence allowing legacy
cellular User Equipment (UE) to operate on the same carrier. In LTE, D2D is operated in
a synchronous way, where the synchronization source can be an eNode-B1 (in case of
UEs being under network coverage) or a UE (in case at least one of the UEs is not under
network coverage or in case of inter-cell operation). Either uplink (UL) spectrum (in case
of Frequency Division Duplexing (FDD)) or UL subframes (in case of Time Division
Duplexing (TDD)) can be used for D2D transmission. One interesting feature is the
interference management among D2D links and cellular links. This feature has not been
discussed in 3GPP, since in practice it is assumed that D2D is running within a dedicated
resource pool (i.e. certain physical resource blocks in specific subframes), where the
D2D-enabled UEs will get the resource pool configuration information from the
eNode-B. In addition, the transmission signals are based on the UL signal design to
avoid introducing a new transmitter at the UE side. Further compared to OFDM
signaling, SC-FDMA can provide better coverage due to the lower Peak to Average
Power Ratio (PAPR). The major features of the 4G LTE D2D concept are listed in the
following, where it should be pointed out that the D2D link is referred to as sidelink in
the 3GPP Radio Access Network (RAN) Working Groups (WGs).

5.1.1.1 D2D synchronization

The sidelink synchronization signal (i.e. D2D synchronization signal), which is transmitted
by the D2D synchronization source (either eNode-B or UE), is used for time and frequency
synchronization to facilitate synchronous D2D operation. In order to achieve
1
The eNode-B term is used to refer to a base station when talking about LTE specific aspects.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 110 Zexian Li et al.

synchronization, at least the following issues need to be solved: synchronization signal

design, entities acting as synchronization source, and criteria to select/re-select the syn-
chronization source.
The sidelink synchronization signal is composed of the primary sidelink synchro-
nization signal and the secondary sidelink synchronization signal. Assuming the UEs
have network coverage, then the eNode-B transmits primary and secondary synchro-
nization signals (specified in LTE Release 8) that are reused for D2D synchronization.
New sidelink synchronization sequences, which are transmitted by a UE acting as
synchronization source (such UE can be in or out of network coverage), have been
specified in 3GPP as well.
Both eNode-B and UEs can act as synchronization sources. It is easy to understand that
the eNode-B can act as a synchronization source. However, in order to facilitate inter-cell
D2D operation, under certain conditions, for example at the cell edge, UEs with network
coverage can transmit synchronization signals as well. In case of partial coverage where
some D2D UEs are with network coverage and the rest are without network coverage,
synchronization signals transmitted by UEs within network coverage can also help the out-
of-coverage synchronization by aligning the out-of-coverage transmission to cellular net-
work timing. In this way, the possible interference from D2D transmission to cellular links
can be reduced.
In order to solve the potential issue of synchronization source selection and re-
selection, different types of synchronization sources are specified with different priority
levels. The eNode-B has the highest priority order followed by in-coverage UEs, and
then out-of-coverage UEs that are synchronized to in-coverage UEs. The out-of-cover-
age UEs not synchronized to any in-coverage UEs have the lowest priority.

5.1.1.2 D2D communication

In LTE Release 12, D2D communication is based on physical layer broadcast commu-
nication, i.e. a physical layer broadcast solution is used to support broadcast, multicast
and unicast services at application layer. In order to support multicast or unicast, the
targeted group ID (for multicast) or user ID (for unicast) is indicated in the higher layer
message. Since by construction it is a broadcasted information, no physical layer closed
control loop exists, i.e., no physical layer feedback, no link adaption, and no HARQ is
supported for D2D links. The air interface is based on the Uu interface and the UL
channel structure is extended to D2D communication. In particular, for D2D data
communication related physical channels, the PUSCH structure (as defined in [4]) is
reused whenever possible. Considering resource usage, D2D communication is based on
a resource pool concept as illustrated in Figure 5.2, where certain time/frequency
resources (called resource pool) are configured for D2D usage. The D2D resource
pool is configurable within one cell and there are separate resources for D2D control
information transmission and D2D data transmission. The resource pool information is
carried over broadcast messages, i.e. SystemInformationBlockType18.
Before the D2D data transmission, every transmitter sends out a control signal with
information on the data transmission format and the occupied resource. This applies to
the scenario where the network is assigning resources to the D2D transmitter and the

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 111

Resource pool for D2D control information transmission

frequency

Resource pool for D2D data transmission

time
LTE resources
Figure 5.2 D2D resource pool.

scenario where the transmitter selects the resource by itself. At the receiving side, it is
not necessary to listen to the cellular control channels in order to find out where the D2D
data is located. Just based on the content of the D2D control channel, the receiving
devices can find out the right location of the relevant resources. As to the resource usage
for D2D communication, two different modes were specified:
• Mode 1: An eNode-B or relay node schedules the exact resources used by a UE to
transmit D2D data and D2D control information. Obviously, Mode 1 can be only
applied to the scenarios where the transmitting UEs are within network coverage.
• Mode 2: A UE by itself selects resources from the configured resource pools to
transmit D2D data and D2D control information. Mode 2 can be applied no matter
whether the transmitting UE has network coverage or not.

5.1.1.3 D2D discovery

In LTE Release 12, discovery is applicable only to the UEs with network coverage.
The concerned UEs can be in either RRC_IDLE state or RRC_CONNECTED state.
Similar to the resources for D2D communication, the D2D discovery resources are
arranged as resource pools as well, which are indicated by the eNode-B via
SystemInformationBlockType19. The resource pools are defined with the parameters
including discoveryPeriod, discoveryOffsetIndicator and subframeBitmap. The
frequency resources within a D2D subframe are given by the parameters startPRB,
endPRB and numPRB. There are two ways specified for a transmitting UE to get the
resources for discovery message transmission:
• Type 1: The UE selects autonomously the resource for transmission from the
discovery pools (independent of the UE RRC state).
• Type 2B: The UE transmits on resources allocated for it by the network (only
applicable to RRC_CONNECTED UEs).

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 112 Zexian Li et al.

The 3GPP RAN WGs specified further D2D enhancements in Release 13 as described in
[5]. These enhancements, for the public safety use cases, aimed to solve out-of-coverage
discovery, layer 3 based UE-to-network relays, enhancement of D2D communication to
support group priorities and group call functionality. However, these are different from the
challenges of 5G D2D that aims to address a wider range of use cases.

5.1.2 D2D in 5G: research challenges

Since in 4G LTE D2D communication the focus is on public safety, the potential
improvements that can be provided by D2D operation are not fully exploited. In the
5G system, such restriction does not exist anymore, and it is predicted that D2D
operation will be natively integrated as part of the future 5G system. Main potential
gains that can be achieved include:

• Capacity/throughput gain: Because the involved devices are in close proximity with
potentially better propagation conditions comparing to the propagation conditions
toward the Base Station (BS), link throughput can be improved due to e.g. better
Modulation and Coding Scheme (MCS) level. In addition, there is the possibility of
sharing the same radio resources among cellular users and D2D users, which can
improve the overall spectrum usage. System capacity can be improved due to off-
loading and local content sharing gain from D2D communication.
• Latency gain: The End-to-End (E2E) latency may be reduced due to a short distance
with less propagation delay, and no involvement of infrastructure network entities
resulting in reduced transport delay and processing delay.
• Availability and reliability gain: D2D can be used to extend network coverage with
one hop or multi-hop. Network coding and cooperative diversity via D2D can be used
to enhance link quality as well. Furthermore, a D2D ad-hoc network can provide a fall
back solution in case of a failure of the infrastructure or in case the infrastructure
cannot be easily established.
• Enabling new services: Full-blown D2D has great potential to enable new services
and applications not only in the telecommunication area, but also in vertical industries,
as for example Vehicle-to-X (V2X) communication as discussed in Chapters 2 and 4.
The extension of D2D solutions for Vehicle-to-Vehicle (V2V) communication is part
of LTE Release 14.
However, as discussed in [6][7], fully utilizing potential D2D gains poses new challenges
in terms of device discovery, communication mode selection, co-existence, interference
management, efficient multi-hop communication support and multi-operator support
among others.

• Device discovery: Efficient network-assisted D2D discovery, which is used to deter-

mine the proximity between devices and the potential to establish a direct D2D link, is
a key element in order to enable D2D communication and possible new applications.
• Communication mode selection: Mode selection is another core function that con-
trols whether two devices will communicate to each other in direct D2D mode or in

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 113

regular cellular mode (i.e. via a BS). In direct D2D mode, the devices can take
advantage of their proximity and may reuse cellular resources for the direct
communication link. In cellular mode, the devices communicate through a com-
mon or separate serving BS by means of regular cellular links in orthogonal
resources with cellular users. How to select the most appropriate communication
mode in different scenarios is an important issue to be solved, as discussed in
Sections 5.2.3, 5.3.4 and 5.4.2.
• Co-existence and interference management: Considering co-existence and related
interference issues, at least two different aspects should be taken into account:
(1) co-existence among a large number of D2D links, and (2) co-existence among
D2D links and regular cellular links. Efficient schemes to handle the interference
are of importance in order to achieve the potential D2D benefits.
• Multi-operator or inter-operator D2D operation: Inter-operator D2D is a clear
requirement resulting from e.g. V2X communication, and supporting inter-opera-
tor D2D operation is essential for the 5G D2D concept. Without multi-operator
D2D support, the applicability of the future D2D solution to e.g. Cooperative
Intelligent Traffic Systems will be quite limited. Considering inter-operator D2D
operation, issues to be solved include, for example, spectrum usage and how to
control and coordinate UEs in D2D communication across multiple operators’
networks.
Clearly, the above bullets are only a subset of the challenges related to D2D operation. In
this chapter, the focus is on the challenges related to radio resource management with the
proposal of one example of a 5G RRM concept in Section 5.2 followed by multi-hop
D2D operation in Section 5.3. Finally, in Section 5.4, multi-operator D2D is addressed,
including discovery support, distributed mode selection and spectrum for multi-operator
D2D.

5.2 Radio resource management for mobile broadband D2D

In this section, the key aspects related to D2D RRM both from a state of the art and
future research perspective are covered. The focus is on mobile broadband D2D
scenarios, i.e. scenarios with typically low mobility where offloading of the cellular
network, enhancement of system capacity and improvement of user experience in
terms of reduced latency and increased data rates play a dominant role [8]. The focus
will be on in-band underlay D2D, in which D2D communication uses the same
spectrum and resources as cellular communication.
The section is structured as follows. Firstly, a brief overview of RRM techniques for
mobile broadband D2D is presented. It is followed by some of the most significant RRM
and system design challenges to be solved in order to make D2D a native and efficient
technology in 5G systems. Finally, an example of a 5G RRM concept based on flexible
TDD is described and performance numbers illustrating the user experience are
provided.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 114 Zexian Li et al.

5.2.1 RRM techniques for mobile broadband D2D

The addition of the D2D layer as an underlay to cellular networks poses new challenges
in terms of interference management in comparison with traditional cellular commu-
nication. These challenges come from the reuse of resources between cellular and D2D
users, which creates intra-cell interference [9][10]. Therefore, in order to exploit the
benefits of D2D communication and achieve an improved system performance over
baseline cellular-only systems, careful resource management that takes into account
both cellular and D2D users is essential.
RRM algorithms and techniques for D2D underlay communications can be classified
depending on the optimization metric and the tools used to achieve that optimized or
improved performance. The most common objectives or optimization metrics of RRM
algorithms and techniques are spectral efficiency, power minimization and performance
with Quality of Service (QoS) constraints [11]. The basic toolbox of available RRM
techniques commonly agreed in the literature, such as mode selection, resource alloca-
tion and power control [12][13], is described in the following.

• Mode Selection (MoS): Several factors influence the MoS decision such as distance
between devices, path loss and shadowing, interference conditions, network load, etc.
and the time scale on which MoS should be operated. A MoS decision can be made
before or after D2D link establishment, while operating on a slow time scale, e.g. based
on distance or large-scale channel parameters [14]. Further, a MoS can be done on a faster
time scale [15][16], based on changing interference conditions coupled with the resource
allocation phase.
• Resource Allocation (ReA): ReA determines which particular time and frequency
resources should be assigned to each D2D pair and cellular link [9][17]. ReA algorithms
can be broadly classified according to the degree of network control, e.g. centralized
versus distributed, and the degree of coordination between cells, e.g. single-cell (unco-
ordinated) versus multi-cell (coordinated).
• Power Control (PC): In addition to MoS and ReA, PC is another key technique
to deal with the interference, both intra- and inter-cell, that results from underlay
D2D operation [18][19]. The focus is mostly on limiting the interference from
D2D to cellular transmission, in order to improve the overall system performance
while ensuring that the cellular user experience is not degraded. The applicability
of LTE power control mechanisms to efficiently support D2D, and optimizations
that rely on a practical distributed scheme, have been extensively studied in [20].
It is worth mentioning that the different algorithms do not rely on just one RRM
component or isolated technique, but normally combine several of them to achieve
better performance [19].

5.2.2 RRM and system design for D2D

Complementing evolved legacy standards with non-backward compatible radio inter-
faces in 5G will allow designing a radio technology that natively and efficiently supports

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 115

D2D from the onset. In Section 5.1.2, some of the general challenges to support D2D in
5G systems, with its broad scope of use cases and scenarios, were highlighted. The
focus here is to specifically address some of the fundamental RRM and system design
questions to be answered for an efficient support of mobile broadband D2D, for
instance:

• How valuable is the usage of D2D across multiple cells, and does this justify the
additional coordination and signaling burden introduced? Enabling inter-cell D2D
requires some kind of basic conflict prevention of RRM decisions between the serving
BSs of the devices involved in D2D communication, even if not targeting optimally
coordinated resource allocation. It could be the case that, in a half-duplex system (e.g.
a 5G system with flexible TDD optimized for dense scenarios), one BS schedules one
of its assigned D2D users for UL transmission (cellular mode selection) while
another BS schedules a direct D2D transmission toward the same user, violating the
half-duplex constraint. Solutions to prevent this issue may include: exchange of
scheduling information between BSs (or via a centralized coordination entity); pro-
tocol-level solutions that orchestrate the order of the transmissions; or simply dis-
abling inter-cell D2D, i.e. only allowing intra-cell D2D and routing the inter-cell D2D
traffic through the infrastructure to avoid the coordination burden.
• Does sophisticated D2D (e.g. fast joint MoS and ReA with flexible TDD) require
centralized radio resource management, or can this be done in a decentralized or
distributed manner? Apart from the multi-cell D2D aspect, it is possible to question
whether centralized RRM can bring substantial benefits to the challenging interfer-
ence conditions of D2D scenarios at a reasonable signaling and computational com-
plexity cost.
• How should MoS between D2D communication and device-infrastructure-device
(DID) be performed, and on which time scale should this be conducted? The possibility
to make use of fast, instantaneous SINR-based MoS against a simpler path-loss based
slow MoS will have a major impact on the protocol stack design. It is needed to carefully
evaluate the trade-off between achievable gains, complexity and signaling overhead.
• Is instantaneous Channel State Information (CSI) of all potentially interfering cellular
and D2D links needed for scheduling purposes, or is the statistical CSI knowledge
enough? In general, D2D communication requires information on the channel gain of
D2D pairs (i.e. the quality of the direct links), the channel gain among D2D pairs (i.e.
generated/received interference to/from other D2D pairs), the channel gain between
D2D transmitters and cellular UEs, and the channel gain between cellular transmitters
and D2D receivers, in addition to the CSI information of cellular-only systems. The
exchange of such extra channel information can become an intolerable overhead to the
system if instantaneous CSI feedback is needed.

5.2.3 5G D2D RRM concept: an example

In this section, an example of a 5G D2D RRM concept in the context of a flexible TDD
air interface is described. The seamless integration of D2D in the flexible UL/DL TDD

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 116 Zexian Li et al.

frame structure is presented and the joint multi-cell D2D and cellular resource alloca-
tion is explained for the case of centralized and decentralized schedulers. Afterward,
adequate mode selection schemes for D2D are analyzed. Finally, some performance
numbers showing the gains of D2D with flexible TDD over fixed TDD and centralized
over decentralized scheduling are provided. The performance and the implementation
implications of two MoS algorithms that operate on different time scales are also
compared.

5.2.3.1 Flexible uplink and downlink TDD concept for D2D

The UL and DL dynamic TDD concept for D2D is based on a MIMO-OFDMA air
interface, similar to the proposal in [21]. The TDD optimized radio has a flexible
frame structure that enables fast TDD access and fully flexible UL/DL switching, in
addition to support for non-conventional type of communications such as D2D and
self-backhauling (cf. Chapter 7 and [21][22]). Each cell can flexibly switch the data
frames to UL or DL within a scheduling slot based on short-term traffic requirements,
without requiring clustered TDD.
D2D communication is natively integrated into the flexible TDD frame by consider-
ing the D2D users in addition to the cellular users. The scheduler decides among UL, DL
and D2D (with simultaneous reuse of resources between cellular and D2D users
allowed) for that cell, taking into account both favorable transmission conditions and
user fairness [23].
Figure 5.3 illustrates the challenges and opportunities presented by multi-cell D2D
communication in scenarios with flexible TDD. The focus is on a specific scheduling
slot and resource block, assuming that resource reuse between D2D and cellular users is
allowed. Further, D2D communication (from UE2 to UE3 and from UE4 to UE5) may
take place at the same time as a UL transmission in Cell1 (from UE1 to BS1) and DL
transmission in Cell2 (from BS2 to UE6). A variety of challenging cross-interference
situations arise such as:

• DL-to-UL interference from BS2 to BS1

• DL-to-D2D interference from BS2 to UE5
• D2D-to-UL interference generated by D2D transmitters like UE2 and UE4 toward
BS1
• D2D-to-D2D interference from D2D transmitters like UE4 to D2D receivers like UE3

Cell 1 Cell 2

BS1 BS2

UE4 UE5
UE1
Interference
UE2 D2D Signal
UE6
D2D UE3

Figure 5.3 Multi-cell D2D in the context of a flexible UL/DL/D2D air interface.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 117

The management of the rapidly changing interference conditions created by flexible

TDD and multi-cell D2D is challenging from the scheduler perspective, but it also
creates the opportunity for joint fast mode selection and resource allocation based on
instantaneous channel conditions, i.e. the scheduling of direct D2D or DID communica-
tion depending on the current signal and interference conditions and network load.

5.2.3.2 Decentralized and centralized schedulers

Centralized (coordinated) or decentralized (uncoordinated) resource allocation approaches
are considered, leading to two different architecture alternatives (cf. Chapter 11). In the
decentralized case, each cell (which could be a small cell) performs its own resource
scheduling decisions. In the centralized case, the channel quality information from the
users is further forwarded by their respective small cells to a centralized entity in the
network, e.g. a macro cell, which performs coordinated scheduling decisions.
The optimization metric is delay-weighted sum rate maximization for each
resource block in a cell (in the decentralized case) or group of cells (in the centralized
case), considering all cellular (UL and DL) and D2D links in that cell or group of
cells, respectively. The scheduling potential of each link, either cellular or D2D,
depends on the achievable data rate on that link (based on SINR estimation from
interference conditions in the previous scheduling slot) and the packet buffer delay (to
provide user fairness in terms of delay) [23].
The scheduler decides for each available resource block which link should make use
of it, either an UL, DL or D2D link(s) (possibly with resource reuse between cellular and
D2D communications), based on a brute force search of the configuration that provides
the highest delay-weighted sum rate out of all the possible combinations. In the
decentralized case, the scheduling decisions are made for each cell independently,
whereas in the centralized case, they are made jointly for a group of cells including all
the possible schedulable links in the cluster. It should be noted that inter-cell D2D is also
supported in the decentralized case by means of a simple scheduling conflict resolution
mechanism that ensures the fulfillment of the half-duplex constraint in the system [22].
It is worth mentioning that the performance of the brute force scheme is to be seen as
an upper bound on the performance of any practical scheduling algorithm, and that the
scheduler assumes instantaneous knowledge of all channel gains between cellular and
D2D users.

5.2.3.3 Mode selection

Mode selection is especially relevant when the separation distance between users (with
traffic to be exchanged) increases. In that case, the routing of the D2D traffic through the
infrastructure may be more efficient than making use of a direct link between the
devices. Hence, it is important to investigate the adequate time scale to perform MoS
between D2D and DID communication. Here the choice is between fast (i.e. based on
instantaneous SINR information) and slow time scale (i.e. based on large-scale channel
conditions). Clearly, conducting fast MoS would imply that the decision is executed at
the MAC layer, whereas in the slow MoS case the decision would be performed at the
PDCP or RRC layer. In fact, the following forms of mode selection are considered:

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 118 Zexian Li et al.

• Direct D2D only: All D2D traffic is served through direct links between devices.
Reuse of resource blocks is allowed between cellular and D2D users.
• Indirect D2D only (Device-Infrastructure-Device, DID): All D2D traffic is routed
through the infrastructure. A D2D communication involves two hops, i.e. a UL
transmission and a subsequent DL transmission. No direct D2D is allowed.
• Path loss-based, slow mode selection: D2D traffic is routed through the infrastruc-
ture when the path loss toward the serving base station and a bias is lower than the path
loss of the corresponding direct D2D link. The bias favors direct D2D communication
over DID due to the inherent advantages of direct D2D. MoS is done before resource
allocation.
• Fast mode selection: D2D traffic is routed through the infrastructure or through the
corresponding direct D2D link depending on the comparison of estimated SINR
conditions between the link that connects the D2D UE to the infrastructure and the
direct D2D link. This calculation is done per scheduling slot based on the interference
conditions in the previous slot. The SINR of direct links is increased by a certain bias
in dBs to favor direct D2D decisions. MoS is made jointly with resource allocation.
More details can be found in [24], which is an extension and more rigorous imple-
mentation of the scheme introduced in [7].

5.2.3.4 Performance analysis

Results are shown for an ultra-dense multi-cell indoor scenario (25 cells, 10 m × 10 m
cells, cell-center BSs), with D2D link range up to 4 m. A scheduling slot, e.g. 2 ms,
consists of several time slots, with each time slot being 0.25 ms long. The system
bandwidth is 200 MHz composed of 100 resource blocks. A bursty traffic model is
assumed, with file size ratios of 4:1:1 for DL/UL/D2D traffic, respectively. A file is
transmitted as multiple packet segments during the course of a simulation, with packet
segment sizes related to the link data rate in a scheduling slot [23].
The cumulative distribution function (CDF) of packet segment serving delay,
defined as the difference between the arrival time and the serving time of a packet
segment, is depicted in Figure 5.4. The packet segment delay shall not be confused
with the MAC latency defined in Chapters 1 and 2. The figure shows the system
performance improvement from flexible TDD and centralized scheduling in terms of
overall packet segment delay, focusing on the worst-case performance between D2D
and cellular links, which is captured by the 99th percentile delay value. No mode
selection is carried out in this case, forcing all D2D traffic to be served through direct
D2D links. In the decentralized fixed TDD scheme, the first four out of the five
scheduling slots are assigned to DL, whereas one slot is used for both UL and D2D.
In the flexible TDD case, there is full flexibility to schedule UL, DL or D2D (with or
without resource reuse) in every scheduling slot depending on short-term traffic
requirements. Decentralized flexible TDD reduces the worst-case delay by 36% in
comparison with decentralized fixed TDD. With centralized flexible TDD, the overall
delay is further reduced by 24% from 245 ms to 185 ms. In fact, the centralized
scheduler allows to balance the delays of the different users and traffic types by means

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 119

0.99

0.98

0.97

0.96

0.95
CDF

0.94

0.93
Decentralized flexible UL/DL/D2D
0.92
Centralized flexible UL/DL/D2D
0.91 Decentralized fixed UL/DL/D2D

0.9
0 0.2 0.4 0.6 0.8 1
Packet segment serving delay [s]
Figure 5.4 Overall (including UL, DL and D2D traffic) packet delays for decentralized fixed and flexible
TDD and centralized flexible TDD.

of its global knowledge and coordinated decisions, improving fairness and worst-case
user experience.
The maximum allowed range of D2D links is now extended from 4 m to 8 m (with
10 m × 10 m cells) and mode selection is enabled. The results are shown in Figure 5.5,
that presents an overview of the compromise reached between cellular and D2D delay
performance for the different MoS variants described in Section 5.2.3.3. The vertical
axis averages the values of UL and DL packet segment serving delay at both 95th and
50th percentile. Proximity to the origin of coordinates means overall improved latency
experience, with the possibility to balance out cellular and D2D delays or to give priority
to one specific kind of traffic by applying different biases. The decentralized variants (in
grey) perform better for median delay values whereas the centralized ones (in black)
improve the delay experience at the 95th percentile. In general, fast MoS is able to
reduce the D2D delay (by around 20%), while keeping similar cellular delay values as
for the path loss-based MoS. The results in Figure 5.5 and in [24] show that fast MoS can
indeed bring gains in the form of a reduced 95th percentile packet delay for D2D
transmissions, without sacrificing cellular performance, but it should ideally be done
in conjunction with coordinated RRM across cells. Furthermore, the aforementioned
gains are on an order that requires careful consideration whether performing D2D MoS
on MAC layer is justified, with the associated likely larger burden in terms of signaling
overhead and complexity.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 120 Zexian Li et al.

95th
percentile
0.3
Cellular packet segment delay [s]

0.25

0.2
Decentralized path loss-based MoS
0.15 Centralized path loss-based MoS
50th Decentralized fast MoS
percentile
Centralized fast MoS
0.1 Decentralized direct D2D
Centralized direct D2D
0.05 Decentralized indirect D2D
Centralized indirect D2D

0
0 0.1 0.2 0.3 0.4 0.5
D2D packet segment delay [s]
Figure 5.5 Cellular vs. D2D packet segment delay for different MoS variants.

5.3 Multi-hop D2D communications for proximity and emergency

services

While cellular-network-assisted D2D communications can capitalize on proximity,

reuse and hop gains [13], so far the main driver for standardizing D2D protocols was
initially the requirement to support Public Protection and Disaster Relief (PPDR) and
National Security and Public Safety (NSPS) services [25]. More precisely, from a
PPDR and NSPS perspective, it is important that, as long as a cellular infrastructure is
operable, communicating devices should have access to broadband services and local
communication should be maintained when cellular coverage becomes unavailable
due to a disaster or emergency situation [26]. Along a related line of technology
development, the use of fixed and mobile relays provides a cost-efficient way of
extending the coverage of cellular networks and can help maintain access to cellular
services when some of the infrastructure nodes become dysfunctional, for example,
in a PDPR or NSPS scenario. In the remainder of this section, some of the key
requirements for NSPS services are highlighted. Afterward, two technology compo-
nents that play a key role in meeting these requirements are discussed. Both D2D
discovery and radio resource management for multi-hop connections should benefit
from network assistance when the cellular infrastructure is intact and should remain
operational, through a graceful degradation, when parts of the network become
dysfunctional.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 121

5.3.1 National security and public safety requirements in 3GPP and METIS
NSPS and PPDR scenarios pose a number of specific requirements that are not
typically found in traditional cellular communications. One of the key requirements
is robustness and ability to communicate irrespective of the presence or absence of a
fixed infrastructure. In many cases, there is at least partial cellular coverage in a
geographical area affected by a disaster or emergency situation, which can be
exploited for communication. Although some of these scenarios can be addressed
by temporary truck-mounted BSs moved into the disaster area, support for proximal
or direct D2D communication – to maintain connectivity among rescue personnel or
between officers and people in need – remains a critical requirement for NSPS
systems [25][26]. Broadband group communication is an example of a requirement
typically not supported or deployed in practice in traditional cellular systems; for
example, when a dispatcher needs to address multiple officers working in an emer-
gency situation, possibly outside network coverage. Figure 5.6 illustrates some of the
use cases that must be supported by the combination and integration of cellular and
D2D technologies.
As illustrated in Figure 5.6, in NSPS and PPDR situations, the rescue personnel,
including officers with public safety UEs, must be able to communicate in situations
in which the cellular BS may provide only partial network coverage. According to
the 3GPP requirements [25], such scenarios include proximity services discovery,
proximity services traffic initiation, UE with multiple traffic sessions, and proximity
services relay. Proximity service discovery is the scenario where a given UE

Proximity Service Discovery Proximity Service Traffic Initiation

Officer B
Officer A Officer A
Officer C

Public Safety
Officer A moves into the discovery Licensed
proximity of Officer C beacons Spectrum
Officer B
and Officer B

UE with Multiple Traffic Sessions Proximity Service Relay

Officer A
communicates with
Officer B and Officer
C concurrently.
Officer A
Officer A
Officer C
Officer B Officer C
Officer C’s UE is not within transmission
range of Officer A’s UE

Officer B

Figure 5.6 Examples of NSPS and PPDR scenarios.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 122 Zexian Li et al.

discovers one or more other UEs and the given UE can be with or without network
coverage. Proximity services traffic initiation refers to the use case where a public
safety UE initiates one-to-one direct user traffic with another UE. UE with multiple
traffic sessions means that a given public safety UE can concurrently maintain
one-to-one user traffic sessions with several other UEs. With proximity services
relay, it is meant that a given UE can act as proximity communication relay for one or
more UEs.
An important aspect of these scenarios and requirements is the inherent support for in-
coverage, out-of-coverage and partially in-coverage services and specifically the
requirement that local (proximal) communication services must be maintained in the
absence or partial availability of the cellular infrastructure.

5.3.2 Device discovery without and with network assistance

Peer and service discovery is a key design issue, both in mobile ad-hoc networks
operating in unlicensed spectra and in cellular network-assisted D2D communications.
The issue stems from the fact that, before the two devices can directly communicate with
one another, the devices or a network entity (such as a cellular BS or a core network
node) must recognize (discover) that they are near each other. In NSPS and PPDR
scenarios, peer discovery is an important service in itself, even without a subsequent
communication session. In fact, discovering devices may help rescue personnel take
appropriate measures without launching further cellular or D2D communication
sessions.
Peer discovery without network support is typically time- and energy-consuming, as it
involves beacon signals and sophisticated scanning and security procedures that often
include higher layers and/or end users. Therefore, when a cellular network is available, it
should assist peer discovery to reduce the discovery time and to increase the energy
efficiency of the discovery process. As shown in [12][27][28], peer discovery resources
in network-assisted mode can be made available and managed efficiently by the net-
work, which can make such peer discovery and pairing procedures faster, more efficient
in terms of energy consumption and more user-friendly. For a deeper analysis of the
achievable gains due to various levels of network assistance, see [26].

5.3.3 Network-assisted multi-hop D2D communications

Although multi-hop D2D communication requirements have been primarily defined
with NSPS scenarios in mind, it is clear that commercial and traditional broadband
Internet services can also benefit from range extension or multi-hop proximity com-
munications, as illustrated in Figure 5.7. As shown in the figure, for a UE positioned
outside the coverage area, it needs another UE that is willing to provide relaying
assistance hence extending the range of a cellular BS. The example in the figure has
two single-hop and two two-hop routes (Route 1, Route 2 and Route 3, Route 4,
respectively). Resources R-1 and R-3 are reused, while R-2 and R-4 are dedicated.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 123

Route 1
R-1

Cellular Multihop Proximity

Communications

R-2 R-3 Route 4

Single-Hop D2D

D2D Tx
Route 2 R-1
Route 3
R-4
D2D Tx
D2D Rx Range D2D Rx
Extension

R-3

D2D Tx

Figure 5.7 Single and multi-hop routes partially under network coverage [29], reproduced with permission
(Lic. no. 3664040827123).

Between each Source-Destination (S-D) pair, a route must be defined and

resources need to be allocated to each link along the route. In Figure 5.7, different
line types indicate different time and frequency resources (Resource Blocks, RBs),
while the same line type for different links indicates resource block reuse. Further it
is assumed that in the multi-hop case, the incoming and outgoing links of a relay
node must use orthogonal resources. A given S-D pair may have the possibility to
communicate in cellular mode through the BS or using single- or Multi-Hop (MH)
D2D communications.
Recall that for D2D communications in cellular spectrum, MoS and resource alloca-
tion (scheduling) and power control are essential. However, extending these key RRM
algorithms to MH D2D communication is non-trivial, since
1. Existing single-hop MoS algorithms must be extended to select between the single-
hop D2D link, MH D2D paths and cellular communications.
2. Existing single-hop resource allocation algorithms must be further developed in
order to not only manage spectrum resources between cellular and D2D layers, but
also to comply with resource constraints along MH paths.
3. Available D2D PC algorithms must be made capable of taking into account the rate
constraints of MH paths. Specifically, it must be taken into account that, along the
multiple links of a given path, only a single rate can be sustained without requiring
large buffers or facing buffer underflow situations at intermediate nodes.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 124 Zexian Li et al.

5.3.4 Radio resource management for multi-hop D2D

A system model that is appropriate for modeling RRM algorithms in MH D2D networks
consists of two parts. The first one is a routing matrix that describes the network
topology and associates links with resources. The second one is a utility function
associated with a S-D pair that characterizes the utility of supporting some communica-
tion rate between the end nodes of the S-D pair.
Recall from Figure 5.7 that MH D2D communications can be used advantageously in two
distinct scenarios. In the proximity communication scenario, a D2D relay node helps a D2D
pair to communicate. In the coverage or range extension scenario, a D2D relay node assists a
coverage-limited D2D Tx node to boost its link budget to a BS, or in NSPS scenarios to a so-
called Cluster Head (CH) node that is capable of taking over the core functionalities of a
cellular base station [26][28]. In the proximity communication scenario, the mode selection
problem consists of deciding whether the D2D Tx node should communicate with the D2D
Rx node (1) via a direct D2D (single-hop) link, (2) via a 2-hop path through the D2D relay
node, or (3) through a cellular BS or ad-hoc CH node. In the range extension scenario, by
contrast, the mode selection problem consists of deciding whether the D2D Tx node should
communicate via a direct transmission with its serving BS or via the D2D relay node. In the
next sub-section, mode selection algorithms are considered for the proximity communica-
tions and range extension scenarios (see Figure 5.7).

5.3.4.1 Mode selection for proximity communications

For the proximity communication scenario, the notion of the equivalent channel from a
D2D transmit (Tx) device to a D2D receive (Rx) device through a D2D relay based on
the harmonic mean of the composite channels from D2D Tx to D2D relay ðGTxRe Þ and
from D2D relay to D2D Rx ðGReRx Þ has been proposed [29]:

1 1 1
¼ þ ð5:1Þ
Geq GTxRe GReRx

The intuition of defining the equivalent channel according to the above is that the
equivalent channel gain tends to be high only when both composite channels are high;
this makes it an appropriate single measure for mode selection purposes. A pseudo-code
of a heuristic mode selection algorithm based on the equivalent channel is given by
Algorithm 1 below, where the channels are needed from the D2D Tx to the BS ðGTxBS Þ
and to the D2D Rx ðGTxRx Þ:
algorithm 1 Harmonic Mode Selection (HMS) for Proximity Communication
1: if Geq ≥ maxfGTxRx; GTxBS g then
2: Choose D2D two-hop communications
3: else if GTxRx ≥ GTxBS then
4: Choose D2D single-hop communications
5: else
6: Choose cellular mode, that is D2D Tx and Rx communication through the BS.
7: end if

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 125

5.3.4.2 Mode selection for range extension

In the range extension scenario, there are only two possible communication
modes (direct or relay-assisted) between the D2D Tx device and the BS or CH
device. Therefore, in this scenario, the definition of the equivalent channel must be
modified such that it includes the path gain between the relay device and the BS
ðGReBS Þ:

1 1 1
¼ þ ð5:2Þ
Geq GTxRe GReBs

This makes it possible to use the following modified version of the Harmonic Mode
Selection (HMS) algorithm:
algorithm 2 Harmonic Mode Selection (HMS) for Range Extension
1: if Geq ≥ GTxBS then
2: Choose D2D relay-assisted communication
3: else
4: Choose cellular mode that is D2D Tx transmits directly to the BS.
5: end if

5.3.5 Performance of D2D communications in the proximity

communications scenario
An effective way to control the fundamental trade-off between power consumption and
system throughput is to employ D2D power control that is not necessarily based on fixed
transmit power levels or the well-known LTE path loss compensating Open Loop (OL)
method. To this end, several power control algorithms have been proposed, the objective
of which is not only to ensure high throughput and energy efficient operation, but also to
protect the cellular layer from harmful interference caused by D2D traffic. Specifically,
the algorithm proposed in [20] and [29] can tune the power consumption and the
throughput of the cellular and D2D layers in single-hop D2D scenarios by setting a
parameter that can be seen as the cost of a unit power investment (i.e. a higher cost of unit
power implies a higher cost of increasing the system throughput by investing higher
transmit power levels). This basic idea has been extended for multi-hop D2D commu-
nication scenarios, including the range extension and proximity communication scenar-
ios of Figure 5.7.
The following figures compare the performance achieved by transmitting with some
fixed power level (“Fix”) or the legacy open loop (“OL”) power control algorithm
(employed by both the cellular and the D2D layers, using 12 dB of SNR target) with
the Utility Maximizing (UM) scheme with some parameter ω (“UM ω = 0.1” and “UM
ω = 100”). The parameter ω represents a trade-off between power consumption and
utility maximization [10][20][29]. Specifically, Figures 5.8 and 5.9 show the invested
power and achieved throughput trade-off in the range extension and proximity commu-
nication scenarios, respectively. These results were obtained in a seven-cell system of a

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 126 Zexian Li et al.

50
(5.6, 45.27)
Average throughput [Mbps] 45

40
(0.3, 35.03)
35
(0.41, 33.84)
Fix
30

25 OL
90% energy saving
with UM ω = 100
20 UM ω = 0.1

15 UM ω = 100
(0.03, 13.65)
10
0 1 2 3 4 5 6
Average power consumption [W]
Figure 5.8 The impact of power control on the power consumption–throughput trade-off in the range
extension scenario.

200
(25.29, 191.17)
Average throughput [Mbps]

More than
180 50% throughput
(20, 165.21)
increase
with UM ω = 0.1
160
Fix
(20, 138.06)
140 OL

UM ω = 0.1
120
(20, 112.78) UM ω = 100

100
19 20 21 22 23 24 25 26
Average power consumption [W]
Figure 5.9 The impact of utility maximizing power control on the fundamental trade-off between power
consumption and throughput in the Proximity Communications scenario.

cell radius of 500 m. The D2D users are randomly dropped in the coverage area of a cell
such that their distance is between 75 m and 125 m. There are 18 uplink physical RBs in
each cell. The other parameters of this system are given in [29]. Further, in this system,
D2D communications are supported in UL physical resource blocks using the Harmonic
Mode Selection algorithm detailed in Section 5.3.4.
Figure 5.8 is a scatter plot for the range extension scenario. The fixed power level of
the “Fix” power control scheme is set such that its performance becomes similar to
that of the “OL” scheme. Note that (x, y) near each symbol shows the x-axis (power
consumption in W) and y-axis (throughput in Mbps) values. Compared with the
traditional OL power control, utility maximizing power control (UM with ω = 100)
reduces overall power consumption at the expense of reducing system throughput. For

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 127

UM ω = 0.1, the utility maximization power control algorithm reaches the highest
average throughput, with a gain of approximately 34% over LTE OL power control.
However, this gain comes at the expense of transmitting at much higher power levels.
In contrast, with ω = 100, utility maximizing power control minimizes power con-
sumption at the expense of reducing the achieved throughput. Clearly, utility max-
imizing PC can reach high throughput when using low values of ω and can transmit at
low power levels with high values of ω.
Figure 5.9 is a scatter plot for the proximity communication scenario. Similarly to
Figure 5.8, with UM ω = 0.1 the average throughput gain is large (approximately 69%)
over the LTE OL scheme, at the cost of using approximately 26% more power. Notice
that in Figure 5.9 the average power consumption includes the power consumption of the
BS. However, with UM ω = 100 the average throughput gain is approximately 20%
using similar transmit power levels as LTE OL. UM ω = 100 boosts the average
throughput at the expense of a small increase in the transmit power level. If the power
consumption must be kept at low values with reasonable throughput values, utility
maximization with higher ω values or using the LTE OL power control technique is a
good design choice.

5.4 Multi-operator D2D communication

The business potential of commercial D2D would be rather limited if direct commu-
nication between devices subscribed to different cellular operators is not supported.
Inter-operator D2D support is also needed to meet the requirements resulting from D2D-
relevant scenarios, e.g. vehicle-to-vehicle communications [7]. In general, D2D support
in inter-operator scenarios becomes more complex as compared to single-operator D2D.
For instance, operators may not be willing to share operator-specific information, e.g.
network loads, utility functions, between each other or with external parties to identify
how much spectrum to allocate for inter-operator D2D communication. In this section,
inter-operator D2D discovery, mode selection, and spectrum allocation schemes are
discussed. Further, single-hop unicast D2D is considered.

5.4.1 Multi-operator D2D discovery

In a multi-operator setting, the D2D discovery cannot be based, for instance, on the time
synchronization and distribution of common peer discovery resources unless the opera-
tors agree to do so. Further, the D2D discovery should rely on both ends of the D2D pair
and on the networks of both operators. In Figure 5.10, an example procedure that enables
multi-operator D2D discovery is shown.
In this example procedure, using LTE terminology, the D2D devices send discovery
messages only on their home operator’s spectrum and hence no change to spectrum
regulation or roaming rules is required. Taking the UE#A as an example, after registration
of D2D operation and authorization process between UE#A, MME#A and MODS
(Multi-operator D2D Server), UE#A can obtain information on discovery resources

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 128 Zexian Li et al.

UE#A eNB#A eNB#B MME#A MME#B MODS

D2D registration of UE #A

Necessary info exchange between operator A and B for

inter-operator D2D operation

UE#A inter-operator D2D discovery authorization process between UE #A, MME#A and
MODS

Information broadcasting to facilitate

inter-operator discovery

Synchronized to eNB#B and reading system

information

Discovering UEs from Operator B

Figure 5.10 Procedure to enable multi-operator D2D discovery.

(both from the home operator and the other operators) based on the broadcasted
information from its home operator. MODS is a new logical network entity which
could be co-located with certain network elements within an operator’s network or
running independently, e.g. as a network service provided by a 3rd party. Example
functionalities of MODS can include D2D subscription management, network access
control, centralized security and radio resource management functions and so on. The
broadcasted key parameters from the home operator include, for example, radio resource
information related to different operators such as operator identifiers and the correspond-
ing operating frequency bands to facilitate inter-operator discovery. UE#A will listen to
both the home and the other operator’s resources to detect the presence of discovery
messages.

5.4.2 Mode selection for multi-operator D2D

D2D mode selection algorithms developed for single-operator networks may not be
directly applicable in a multi-operator system setup. Operators may not want to share
information regarding the locations of users or path loss data as in [30], or CSI between
the D2D users and their home operator’s BS (as required in the mode selection algorithm
described in Section 5.2.3). Moreover, operators may not want to cooperate in order to
estimate the D2D pairwise distance, see for instance [31], and use it as a criterion for
mode selection.
In a single-operator network, either dedicated spectrum can be allocated to the
D2D users (also known as D2D overlay), or D2D and cellular users can be allocated
to the same resources (also known as D2D underlay). In a multi-operator D2D
underlay, the cellular users are exposed to inter-operator interference generated
from the D2D users involved in inter-operator communication sessions. The problem
of inter-operator interference between cellular users and D2D users needs to be
resolved without an excessive information exchange between the operators.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 129

Clearly, at a first stage, an overlay multi-operator D2D scheme would be easier to

implement. In the overlay D2D setting, the key design issue is the way to divide
spectral resources between cellular and D2D users and the communication mode
selection scheme. One method to select the communication mode without incurring
excessive communication signaling overhead may rely on the received signal level at
the D2D receiver. Such algorithm has been proposed in [32] and it is straightforward
to extend it in a multi-operator setting because it does not require proprietary
information exchange between the operators.

5.4.2.1 Mode selection algorithm

Given the spectral resources allocated for inter-operator D2D communication, the D2D
receiver measures the interference level and communicates a quantized version of the
interference to its home BS. The BS compares the measurement report with a decision
threshold and chooses the D2D communication mode only if the measured interference
is low. The D2D receiver should signal the selected communication mode back to the
D2D transmitter, i.e. the source UE subscribed with the other operator, and the session in
which it may start.
Note that the mode selection threshold impacts the overall network performance
because it determines the amount of inter-operator D2D sessions and also the portion
of users in cellular communication mode. The mode selection threshold should be a
priori agreed i.e. optimized between the operators.
The mode selection algorithm described above could also be implemented in the
following manner: the interference measurements could be carried out at the D2D
transmitter instead of the receiver. In that case, the transmitter would be responsible
for reporting the measurements to its home BS. While discussing the spectrum allocation
algorithm for inter-operator D2D in Section 5.4.3, it is assumed that the mode selection
takes place at the transmitter, since the performance can be assessed using analytical
means (as long as the D2D pair distance is short).

5.4.3 Spectrum allocation for multi-operator D2D

D2D communication can be enabled either over licensed or unlicensed spectrum. D2D
communication in unlicensed bands would suffer from unpredictable interference.
Licensed spectrum seems to be the way forward to enable LTE D2D communication,
especially considering safety related scenarios such as vehicle-to-vehicle communica-
tion (see Chapters 4 and 7 for more information about V2V).
Overlay inter-operator D2D communication takes place over dedicated spectral
resources possibly originating from both operators. For FDD operators, the spectral
resources may refer to Orthogonal Frequency-Division Multiplexing (OFDM) sub-
carriers, while for the TDD operators they may refer to time-frequency resource
blocks. In the TDD case, inter-operator D2D support would pose a requirement for
time synchronization between the operators, which is more challenging. In Figure
5.11, two FDD operators contribute a part of their cellular spectrum, β1 and β2
respectively, for inter-operator D2D communication. Also, each operator i ¼ ð1; 2Þ

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 130 Zexian Li et al.

Cellular band Operator 1 Cellular band Operator 2

Cellular mode D2D mode Cellular mode

Shared Band

Figure 5.11 Spectrum divisions for two operators supporting inter-operator D2D communication.

allocates fractions βic and βid for cellular and intra-operator D2D communication
respectively.
When more than two operators are involved in spectrum sharing, it is possible to
realize inter-operator D2D communication based on bilateral agreements between
operators, or alternatively, all operators may commit some of their spectral resources
in a common spectrum pool, see also Chapter 12 for a detailed description of inter-
operator spectrum sharing based on mutual renting and spectrum pooling. The operators
should negotiate the amount of resources they want to commit, but they should not be
forced to take action. However, once the operators agree to share spectrum for some time
and commit certain resources for multi-operator D2D, they are not allowed to break the
agreement. The duration of the agreement should be set in advance and may depend on
the expected network traffic dynamics.
In general, operators are competitors and they may not want to reveal proprietary
information, e.g. utility functions, and network load. Ideally, the negotiations about
spectrum allocation for multi-operator D2D should be completed without exchanging
proprietary information. One possible way to do that is to model the operators as
selfish players, and use a non-cooperative game theoretical approach. For instance, an
operator can make a proposal about the amount of spectral resources it is willing to
contribute, taking into consideration its own reward and the proposals made by the
competitors. All operators can update their proposals based on the proposals submitted
by the competitors until consensus is reached. This kind of updating procedure is also
known as best response iteration and it is a common method to identify the Nash
equilibrium of a one-shot non-cooperative game [33].
In a non-cooperative game, one of the most important aspects is the existence and
uniqueness of a Nash equilibrium. A situation where there are multiple equilibrium
points may be undesirable because the realized equilibrium will depend on the selection
order and the initial proposals of the operators. As a result, it is important to note that
operators may be interested to share spectrum only if a unique Nash equilibrium exists.
For the time being, the spectrum allocation algorithm does not support coupled con-
straints between the operators. In that case, there may exist infinite normalized equili-
brium points [34]. Hence, some sort of extensive information exchange between the
operators might be needed to obtain an efficient equilibrium.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 131

Operator checks
concavity and
diagonal
dominance

Concave and
diagonal dominant

NO
YES

Optimize Optimize
considering considering Convergence

YES

Figure 5.12 Best response iteration algorithm for spectrum allocation in multi-operator D2D communication.

5.4.3.1 Spectrum allocation algorithm

A sequential updating procedure is considered until consensus is reached where each
operator strategy consists of responding to the others with the amount of spectrum each
operator is willing to contribute for multi-operator D2D support. This strategy is one-
dimensional. Further, each operator considers only its individual network utility and
performance constraints. It is well known that, for concave utilities and constraints, an
equilibrium exists. However, to establish uniqueness, the best response operator should
also be a contraction [35]. For one-dimensional strategies, the contraction principle can
be degenerated to the dominance solvability condition, which essentially means that an
operator can control its own utility more than all other operators can do. Fortunately,
each operator can check independently whether its optimization criteria are concave or
not and whether the dominance solvability condition holds true. The operators can
exchange binary messages regarding these conditions and provided that all indications
are positive, the operators become automatically aware about the uniqueness of the
equilibrium. Therefore, in that case, the best response iteration can start. Any operator
can be ranked first. In case an operator experiences a performance loss as compared to
no sharing, it should immediately break the agreement. The best response updating
procedure is also summarized in Figure 5.12.

5.4.3.2 Numerical example

Assuming that each operator wants to maximize its average D2D user rate including
own operator and inter-operator D2D users subject to transmission rate constraints
for cellular communication mode and intra-operator D2D users. With the MoS

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 132 Zexian Li et al.

2
Operator 2
1.9 Operator 1

1.8

1.7

1.6
Gain

1.5

1.4

1.3

1.2

1.1
0 20 40 60 80 100
Inter-operator D2D user density (#users/km2)
Figure 5.13 Performance gain in terms of average user rate for two operators as compared to the case without
multi-operator D2D support.

scheme described in Section 5.4.3.1, it can be shown that the utilities as well as the
constraint are concave [36]. Besides, in a spectrum-sharing scenario between two
operators, the dominance solvability condition holds always true irrespective of the
user densities [36].
The network is modeled for each operator using a Voronoi tesselation2 with an
average inter-site distance of 100 m. Full-buffer traffic model is assumed, with the user
density directly related to the network load. The densities of cellular and inter-operator
D2D users is 30 users/km2 (per operator) to model a scenario where the densities of the
users are comparable to the densities of BSs. The density of intra-operator D2D users is
30 users/km2 for Operator 1 and it varies for Operator 2 to model asymmetric network
loads between the operators. A 3GPP propagation environment is used with Rayleigh
fading [37]. The average D2D link distance is 30 m. The MoS threshold is fixed to -72
dBm both for inter-operator and intra-operator D2D users. The decision threshold
impacts the density of users selecting a D2D communication mode. A performance
evaluation with other threshold values is available in [36]. The baseline scheme for
comparison is not supporting multi-operator D2D communication. In that scheme, all
inter-operator D2D traffic is routed toward the cellular infrastructure.
In Figure 5.13, the performance gain is shown in terms of average user rate for both
operators. When both operators have an equal network load, they both experience

2
The base stations are distributed uniformly and each point of the plane is associated with the nearest base
station.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 133

1
No sharing
0.9 Spectrum sharing

0.8

0.7

0.6
CDF

0.5

0.4

0.3

0.2

0.1

0
0 0.5 1 1.5 2 2.5 3
Multi-operator D2D user rate [bps]
Figure 5.14 Rate distribution for multi-operator D2D users with and without multi-operator D2D support.

around 50% performance gain. The gains for both operators become high when the
network load of Operator 2 becomes low. In that case, Operator 2 is able to contribute
a high amount of spectral resources for multi-operator D2D support, and both
operators can enjoy performance gains close to 100% due to spectrum sharing and
D2D proximity.
Figure 5.14 depicts the rate distribution for multi-operator D2D users, where sym-
metric operators are assumed. Without spectrum sharing, all inter-operator D2D traffic is
routed to the cellular infrastructure and the achievable D2D user rate is low. One can see
that multi-operator D2D support can boost the median D2D user data rate by up to a
factor of 4. Hence, multi-operator D2D support is required in order to harvest the
business potential of D2D communications, e.g. in the context of vehicular commu-
nication for traffic efficiency and safety.

5.5 Conclusions

It is envisioned that integrated D2D communication will play a more important role in
the future 5G system thanks to the promising benefits on both network and end-user
sides, contributing to traffic offloading, very high throughput, significantly reduced
latency and low power consumption. In addition, D2D has the potential to increase
communication availability and reliability, and provide additional diversity. From a
service and application perspective, D2D possesses the capability of enabling a number
of new applications such as V2V and machine-type communications. Finally, in order to

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 134 Zexian Li et al.

capitalize on the full benefits of a system with native support of D2D operation, it is
needed to address in the coming years additional challenges related to for example
mobility management (exemplary solutions can be found in Chapter 11) and security.

References

[1] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M. Maternia, O. Queseth,

M. Schellmann, H. Schotten, H. Taoka, H. Tullberg, M. A. Uusitalo, B. Timus, and
M. Fallgren, “Scenarios for 5G mobile and wireless communications: The vision of
the METIS Project,” IEEE Communications Magazine, vol. 52, no. 5, pp. 26–35,
May 2014.
[2] NGMN Alliance, 5G White Paper, February 2015, www.ngmn.org/uploads/medi
a/NGMN_5G_White_Paper_V1_0.pdf
[3] Qualcomm, “LTE Device to Device Proximity Services,” Work Item RP-140518,
3GPP TSG RAN Meeting #63, March 2014.
[4] 3GPP TS 36.211, “Evolved Universal Terrestrial Radio Access (E-UTRA); Physical
channels and modulation,” Technical Specification TS 36.211 V11.6.0, Technical
Specification Group Radio Access Network, September 2014.
[5] Qualcomm, “Enhanced LTE Device to Device Proximity Services,” Work Item
RP-150441, 3GPP TSG RAN Meeting #67, March 2015.
[6] ICT-317669 METIS project, “Initial report on horizontal topics, first results and
5G system concept,” Deliverable D6.2, March 2014, www.metis2020.com/wp-c
ontent/uploads/deliverables/
[7] Z. Li, M. Moisio, M. A. Uusitalo, P. Lundén, C. Wijting, F. S. Moya, A. Yaver, and
V. Venkatasubramanian, “Overview on initial METIS D2D concept,” in
International Conference on 5G for Ubiquitous Connectivity, Levi, November
2014, pp. 203–208.
[8] ICT-317669 METIS project, “Intermediate system evaluation results,” Deliverable
D6.3, August 2014, www.metis2020.com/wp-content/uploads/deliverables/
[9] T. Peng, Q. Lu, H. Wang, S. Xu, and W. Wang, “Interference avoidance mechan-
isms in the hybrid cellular and device-to-device systems,” in IEEE International
Symposium on Personal, Indoor and Mobile Radio Communications, Tokyo,
September 2009, pp. 617–621.
[10] G. Fodor, M. Belleschi, D. D. Penda, A. Pradini, M. Johansson, and A. Abrardo,
“Benchmarking practical RRM algorithms for D2D communications in LTE
advanced,” Wireless Personal Communications, vol. 82, pp. 883–910, December
2014.
[11] A. Asadi, Q. Wang, and V. Mancuso, “A Survey on device-to-device communica-
tion in cellular networks,” IEEE Communications Surveys & Tutorials, vol. 16,
no.4, pp. 1801–1819.
[12] S. Mumtaz and J. Rodriguez (Eds.), Smart Device to Smart Device
Communication, New York: Springer-Verlag, 2014.
[13] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklós, and Z.
Turányi, “Design aspects of network assisted device-to-device communications,”
IEEE Communications Magazine, vol. 50, no. 3, pp. 170–177, March 2012.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 D2D communications 135

[14] N. Reider and G. Fodor, “A distributed power control and mode selection algo-
rithm for D2D communications,” EURASIP Journal on Wireless Communications
and Networking, vol. 2012, no. 1, December 2012.
[15] S. Hakola, Tao Chen, J. Lehtomaki, and T. Koskela, “Device-To-Device (D2D)
communication in cellular network: Performance analysis of optimum and prac-
tical communication mode selection,” in IEEE Wireless Communications and
Networking Conference, Sydney, April 2010.
[16] K. Doppler, C.H. Yu, C. Ribeiro, and P. Janis, “Mode selection for device-to-
device communication underlaying an LTE-Advanced network,” in IEEE Wireless
Communications and Networking Conference, Sydney, April 2010.
[17] G. Fodor and N. Reider, “A distributed power control scheme for cellular network
assisted D2D communications,” in IEEE Global Telecommunications Conference,
Houston, December 2011.
[18] C.H. Yu, O. Tirkkonen, K. Doppler, and C. Ribeiro, “Power optimization of
device-to-device communication underlaying cellular communication,” in IEEE
International Conference on Communications, Dresden, June 2009.
[19] H. Xing and S. Hakola, “The investigation of power control schemes for a device-to-
device communication integrated into OFDMA cellular system,” in IEEE
International Symposium on Personal Indoor and Mobile Radio Communications,
Istanbul, September 2010, pp. 1775–1780.
[20] G. Fodor, M. Belleschi, D. D. Penda, A. Pradini, M. Johansson, and A. Abrardo,
“A comparative study of power control approaches for D2D communications,” in
IEEE International Conference on Communications, Budapest, June 2013.
[21] P. Mogensen et al., “5G small cell optimized radio design,” in IEEE Global
Telecommunications Conference Workshops, Atlanta, December 2013, pp. 111–116.
[22] E. Lahetkangas, K. Pajukoski, J. Vihriala, and E. Tiirola, “On the flexible 5G dense
deployment air interface for mobile broadband,” in International Conference on
5G for Ubiquitous Connectivity, Levi, November 2014, pp. 57–61.
[23] V. Venkatasubramanian, F. Sanchez Moya, and K. Pawlak, “Centralized and
decentralized multi-cell D2D resource allocation using flexible UL/DL TDD,” in
IEEE Wireless Communications and Networking Conference Workshops, New
Orleans, March 2015.
[24] F. Sanchez Moya, V. Venkatasubramanian, P. Marsch, and A. Yaver, “D2D mode
selection and resource allocation with flexible UL/DL TDD for 5G deployments,”
in IEEE International Conference on Communications Workshops, London, June
2015.
[25] 3GPP TR 22.803, “Feasibility study for Proximity Services (ProSe),” Technical
Report TR 22.803 V12.2.0, Technical Specification Group Radio Access Network,
June 2013.
[26] G. Fodor et al., “Device-to-sevice communications for national security and public
safety,” IEEE Access, vol. 2, pp. 1510–1520, January 2015.
[27] Z. Li, “Performance analysis of network assisted neighbor discovery algorithms,”
School Elect. Eng., Royal Inst. Technol., Stockholm, Sweden, Tech. Rep. XR–EE-
RT 2012:026, 2012.
[28] Y. Zhou, “Performance evaluation of a weighted clustering algorithm in NSPS
scenarios,” School Elect. Eng., Roy. Inst. Technol., Stockholm, Sweden, Tech.
Rep. XR-EE-RT 2013:011, January 2014.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006
 136 Zexian Li et al.

[29] J. M. B. da Silva Jr., G. Fodor, and T. Maciel, “Performance analysis of network

assisted two-hop device-to-device communications,” in IEEE Broadband Wireless
Access Workshop, Austin, December 2014, pp. 1–6.
[30] C.-H. Yu, K. Doppler, C. B. Ribeiro, and O. Tirkkonen, “Resource sharing
optimization for device-to-device communication underlaying cellular networks,”
IEEE Transactions on Wireless Communications, vol. 10, no. 8, pp. 2752–2763,
August 2011.
[31] X. Lin, J. G. Andrews, and A. Ghosh, “Spectrum sharing for device-to-device
communication in cellular networks,” IEEE Transactions on Wireless
Communications, vol. 13, no. 12, pp. 6727–6740, December 2014.
[32] B. Cho, K. Koufos, and R. Jäntti, “Spectrum allocation and mode selection for
overlay D2D using carrier sensing threshold,” in International Conference on
Cognitive Radio Oriented Wireless Networks, Oulu, June 2014, pp. 26–31.
[33] M. J. Osborne, An Introduction to Game Theory, Oxford: Oxford University Press,
2003.
[34] J. Rosen, “Existence and uniqueness of equilibrium points for concave n-person
games,” Econometrica, vol. 33, pp. 520–534, July 1965.
[35] D. Gabay and H. Moulin, “On the uniqueness and stability of Nash equilibrium in
non-cooperative games,” in Applied Stochastic Control in Econometrics and
Management Sciences, A. Bensoussan, P. Kleindorfer, C. S. Tapiero, eds.
Amsterdam: North-Holland, 1980.
[36] B. Cho, K. Koufos, R. Jäntti, Z. Li, and M.A. Uusitalo “Spectrum allocation for
multi-operator device-to-device communication,” in IEEE International
Conference on Communications, London, June 2015.
[37] 3GPP TR 30.03U, “Universal mobile telecommunications system (UMTS);
Selection procedures for the choice of radio transmission technologies of the
UMTS,” Technical Report TR 30.03U V3.2.0, ETSI, April 1998.

Downloaded from http:/www.cambridge.org/core. New York University Libraries, on 16 Dec 2016 at 15:17:03, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1017/CBO9781316417744.006