You are on page 1of 37

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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

Routing in Multi-hop Cellular Device-to-Device

(D2D) Networks: A Survey
Farrukh Salim Shaikh and Roland Wismüller

Abstract—In recent years, device-to-device (D2D) communica- extent. Two hop communication is approved by 3GPP in
tion has attained significant attention in the research community. Release 13-15, which further minimizes the network capac-
The advantages of D2D communication can be fully realized in ity/coverage problem [6]. In Release 13, 14 and 15, to enable
multi-hop communication scenario. The integration of cellular
and multi-hop networks not only provides guaranteed Quality of network services by a device which is out of network coverage,
Service (QoS) and reliability as a traditional cellular network, but it can use another nearby device as a relay which is within
also has the flexibility and adaptability as a multi-hop network. network coverage, to communicate with the network. It can be
Routing in such multi-hop cellular D2D networks is a critical easily deduced from the proliferation of D2D communication
issue, since the multi-hop network can perform worse than that multi-hop communication will be a part of standard in
a traditional cellular network if wrong routing decisions are
made. This is because routing in these multi-hop networks needs near future. The advantages of D2D communication can be
to take care of the node mobility, dynamic network topology, fully realized in the multi-hop communication scenario since
and network fragmentation, which did not exist in traditional the single-hop communication is usually limited to a specific
cellular networking. This paper provides a comprehensive survey geographic area. This restriction is imposed (for a while) only
of routing in multi-hop D2D networks. Some future research to ensure minimum interference to other users in the network.
directions for the routing in D2D networks are also discussed at
the end of this paper. D2D communication can be used both for public safety
and commercial purposes. Most of the works have focused
Index Terms—Routing, multi-hop cellular networks, device-to- on providing D2D communication in case of natural disasters
device, D2D, LTE.
when the cellular network services are either partially or totally
unavailable, while very few works considered human-made
I. I NTRODUCTION disasters like fires, accidents, terrorist attacks etc. Tata and
A cellular network establishes direct single-hop commu- Kadoch [7] highlight the fact that many firefighters die because
nication between the user nodes and a base station (BS). they cannot get out of the fire spot. This is because they are
This direct connection provides the best services for delay- unable to communicate due to network congestion caused by
sensitive communications, but suffers from high data traffic the peak usage of the network in the locality. Authors of [8]
load due to the limited amount of available spectrum resources point our attention towards the terrorist attacks. Most of the
[1]. The increase in the network capacity requirements due terrorist attacks are targeted in urban areas in order to have
to the high data traffic load results in a reduction of the higher impact and causalities. This leads to networks being
overall network coverage. One way to handle this issue is overloaded or damaged. D2D communication provides an
to increase the number of BSs which is very expensive [2]. alternative communication in case of such network congestion
By using the low transmission power, two devices can reuse or network unavailability. Due to the low costs and high
the spectrum band which might already be used by some data rates provided by LTE D2D, Federal Communications
other cellular device or D2D pair(s), by utilizing the concept Commission (FCC) has endorsed it as the next generation
of space division multiple access. Thus, the third Generation public safety network in the USA [9].
Partnership Project (3GPP) identifies device-to-device (D2D) In D2D communication, the users communicate with each
communication as a potential candidate technology to handle other in an ad hoc manner using single-hop or multi-hop.
the network capacity/coverage problem [3]. This communication can happen both with and without the
It is approved in 3GPP Long Term Evolution (LTE) Release support of the cellular network infrastructure. According to
12 that two devices in close proximity to each other can com- different studies, reduced delay and energy consumption in
municate directly. This direct communication can be applied the single-hop communication and improved network cover-
in every possible kind of network scenarios. For example, the age in the multi-hop communication have been promised by
devices can be both under the network coverage (in coverage D2D communications. Other performance improvements in
scenario), or one device can be outside the network coverage both single-hop and multi-hop D2D communications include
(partial coverage scenario), or even both devices can be outside higher data rate, Quality of Service (QoS), spectrum efficiency,
the network coverage (out of coverage scenario) [4], [5]. This network capacity and availability of service and more balanced
alleviates the problem of network capacity/coverage to some network load. Since multi-hop D2D communication requires
intermediate devices to relay the data, therefore to motivate
Please direct correspondence to Farrukh Salim Shaikh. devices to assist in relaying data, a number of incentive
Farrukh Salim Shaikh and Roland Wismüller are with Department of mechanisms are proposed. Such performance improvements
Electrical Engineering and Computer Science, University of Siegen (email:
engr, Phone: and incentives are discussed in Section II-E in detail.
+00 (49)271 740 4050. A number of works advocate the usefulness of multi-hop

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

D2D communications in various network conditions/scenarios are discussed in Section IV, V and VI respectively.
[10]–[13]. For example, the work in [10] proposed a Multi-hop The type of relay depends upon the type of routing scheme
D2D-based Broadcast (MDB) service to enable broadcast in being used. The relay used by multi-hop D2I/I2D routing
a target area without having limitation of any cell boundaries. scheme is called network relay, as this relay helps the UEs
Multi-hop communication supports the services to scale to a to communicate with the BS. The relay used in multi-hop
large geographical area while keeping the advantages of D2D D2D routing scheme is called device relay or D2D relay, as
communication. The authors argue that the existing broadcast this relay helps UEs to communicate with other UEs or helps
communication channel i.e. Multicast Broadcast Single Fre- BS to communicate with other UEs. Network operators also
quency Network (MBSFN) in LTE for delivering the services provide some fixed relays for extending the network coverage.
like mobile TV, is inflexible due to some circumstances. For Thus, we have two sub-categories of network relay: fixed
example, the broadcast message is sent to all users in a cell network relay and mobile network relay. Fixed network relay
and cannot be geofenced to some specific area. Moreover, the is the static relay installed by the network operator, and mobile
MBSFN consumes the capacity in downlink channel, even if network relay is the user node which provides services for
there are no transmission requirements at all. The simulation data forwarding between the BS and the other users. Table V
results showed that MDB in a multi-cell network, is fast and lists the type of relays used in state-of-the-art topology based
reliable with no extra resource consumption (due to the uplink routing algorithms along-with the topology structure used.
frequency-reuse by D2D nodes). In order to understand the importance of the need of having
The integration of cellular and multi-hop network provides new routing protocols to meet the unique characteristics of
not only guaranteed QoS and reliability like a traditional D2D communications, we provide an evolution of the routing
cellular network, but also has the flexibility and adaptability protocols from wired networks to multi-hop cellular D2D
like a multi-hop network [14]. Routing in such multi-hop networks as follows. The routing protocols have been first de-
cellular D2D networks is a critical issue, since the multi- vised for wired networks in order to establish communication
hop network can perform even worse than traditional cellular among nodes which are not directly connected to each other
network if wrong routing decisions are made [15]. This is with the same physical medium. The distance vector routing
because, routing in these multi-hop networks need to take (DVR) and link state routing (LSR) have been proposed as the
care of the node mobility, dynamic network topology, and main routing approaches in this context. In DVR, nodes share
network fragmentation, which did not exist in traditional their distance vector information with neighbors to maintain
cellular networking. An efficient routing scheme provides next hop-node for each destination. The DVR suffers from
an enhanced performance, such as higher network capacity slow convergence and routing-loops problem, which limits its
and reduces interference [16]. The detailed discussion on the applicability to a small area. In LSR, each node floods in the
performance improvements promised by routing protocols for whole network, the complete status of its directly connected
multi-hop D2D networks, is provided in next sections (see links. In this way, by knowing the link state information of
Section IV, V and VI). all nodes, each node can draw a complete network graph for
With the introduction of multi-hop communications, users routing decisions [17].
can communicate in one of the four modes: (i) single- The emergence of the Internet necessitates the optimization
hop D2D communication (ii) multi-hop D2D communication of routing, hierarchical routing is proposed to make routing
(iii) multi-hop Device-to-Infrastructure (D2I)/Infrastructure- decision based on network addresses instead of host addresses.
to-Device (I2D) communication, and (iv) traditional cellular The inter-domain routing is used to route packets between
communication. A pictorial view of multi-hop D2D, D2I and different networks and the intra-domain is used within a
I2D communication is shown in Fig. 1. network to route packet to the required destination node.
There are three broad categories of D2D routing: multi-hop The proliferation of wireless networks necessitate the need
D2I/I2D routing, multi-hop D2D routing and ad-hoc routing of new routing protocols due to the unreliable nature of
for D2D networks. In multi-hop D2I/I2D routing, the commu- the wireless links. The Destination-Sequenced Distance-Vector
nication is always governed by BS. In multi-hop D2D routing, Routing (DSDV), Optimized Link State Routing (OLSR),
communication could or could not be governed by BS. When Ad hoc On-Demand Distance Vector (AODV) and Dynamic
communication is not under the control/supervision of BS, Source Routing (DSR) are few of the famous routing protocols
the routing behaves exactly as an ad-hoc routing and any ad- proposed for wireless networks, which make variations in the
hoc routing protocol can be used with or without modification, basic approaches used in wired network, i.e., DVR and LSR.
depending upon the requirements. The only difference between The DSDV is a proactive routing protocol that periodically
ad-hoc routing and unsupervised D2D routing is in the usage maintains a route to every other node in the network, while
of spectrum frequency bands. D2D nodes can use both the avoiding routing loops at the same time. The AODV is
licensed and/or the unlicensed bands while ad-hoc nodes can a reactive routing protocol that finds the route only when
only use unlicensed bands. When communication is governed required. For some special types of networks (e.g., wireless
by BS, or when D2D communication takes into consideration sensor networks) where nodes gather information and send to
the presence of BS and other nodes in the network, such some centralized node called sink node, Hierarchical Routing
routing is considered purely as the multi-hop D2D routing. has been proposed for such data gathering. The nodes have
Taxonomy and state-of-the-art routing algorithms proposed for merged into groups, in order to forward consolidated data to
multi-hop D2D, D2I/I2D and ad-hoc routing for D2D networks the sink node [17].

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

All these routing protocols are distributed in nature, where TABLE I: List of acronyms and their definitions.
all nodes take the route decision autonomously using local Acronym Definition
3GPP Third Generation Partnership Project
and/or global information. There is no routing protocol that ACK ACKnowledgment
takes the benefit of the presence of centralized control entity AODV Adhoc On-demand Distance Vector
B-RREQ Broadcast Route REQuest
(e.g., BS or access point (AP)) for centralized route decisions. BCH Broadcast CHannel
The development of D2D technology, with its promising per- CBR Constant Bit Rate
CCH Control CHannel
formance improvements, necessitates the need of new routing CoMP Coordinated MultiPoint
protocols that can take advantage of BS for making route CR Cognitive Radio
D2D Device-to-Device
decisions. The BS can guarantee performance and security by D2I Device-to-Infrastructure
complete command over the control and data plane. DSR Dynamic Source Routing
DTN Delay/Disruption Tolerant Network
The routing in D2D networks helps in establishing an ESTOP Early STOP
efficient multi-hop communication route between the devices FDD Frequency Division Duplex
I2D Infrastructure-to-Device
or user equipment (UE) where a single-hop direct connection I2I Indoor-to-Indoor
is not possible. The classification of routing algorithms will IFOM IP Flow Mobility
I-WLAN Interworking Wireless Local Area Network
help to select a routing algorithm based on varying user and LOS Line Of Sight
network requirements. This paper comprehensively discusses LSP Link State Packet
MCN Multi-hop Cellular Network
each routing scheme in order to have a clear understanding of MIP Mobile IP
the ongoing works. For convenience, some general acronyms NACK Negative ACKnowledgment
NLOS Non-Line Of Sight
and their definitions along with the acronyms of routing O2I Outdoor-to-Indoor
algorithms (topology based) used in this survey are listed in O2O Outdoor-to-Outdoor
ODMA Opportunity Driven Multiple Access
Table I and Table II respectively. OLSR Optimized Link State Routing
PAPR Peak-to-Average Power Ratio
PMIP Proxy Mobile IP
A. Contribution of This Survey Article ProSe Proximity Services
Although the concept of routing in multi-hop cellular D2D SARC Separation ARChitecture
networks exists from several decades (with various other SIPTO Selective IP Traffic Offload
SPR Shortest Path Routing
names, e.g., ODMA, MCN, IMCDN and ICN. See Section TOF Traffic Offload Function
II-A.), however to the best of our knowledge, there is no exist- UE User Equipment
VC Virtual Connection
ing survey so far that covers routing in multi-hop cellular D2D
networks. Thus, this is the first comprehensive survey which
discusses the routing in multi-hop cellular D2D networks with II. OVERVIEW ON D2D COMMUNICATIONS
detailed taxonomy. Major contributions of this paper are as
follows: This section provides detailed overview on D2D commu-
nications in many perspectives. The first subsection discusses
1) A clear differentiation between multi-hop D2D, D2I/I2D
various technologies which resembles with D2D communica-
and ad-hoc routing for D2D networks is provided. Such
tions. The next subsection provides an overview of existing
categorization of routing in multi-hop D2D networks
surveys published in D2D communications. Implementation
does not exist in any earlier work.
scenarios and the unique challenges in D2D communications
2) A number of topics that were not well classified in D2D
are presented in the Subsections II-C and II-D. Subsections
communications like various technologies that resembles
II-E to II-J discuss single and/or multi-hop network perfor-
D2D communications, comparison of existing surveys,
mance improvements and the incentives to users by using D2D
and discussion of various incentive mechanisms to sup-
port D2D communications have been presented (Section
3) A detailed overview on D2D communications including A. History
classification, advantages, implementation scenarios, and The idea of D2D communication dates back to nearly
unique challenges is provided (Section II). four decades, when opportunity driven multiple access
4) An in-depth survey of all state-of-the-art D2D, D2I/I2D (ODMA) was first proposed for extending the network
and ad-hoc routing schemes for D2D networks along- coverage. Different techniques with their specific names have
with their taxonomy is provided (Section IV, V and VI been proposed so far, for the direct (single and multi-hop)
respectively). communication of cellular users under various network
5) A comparative analysis of routing schemes is made scenarios [18]. All the techniques have been seamlessly
using various metrics, such as the type of relay and integrated into D2D communications. A brief description
the type of spectrum used. This would help researchers about all such popular techniques is provided below:
to select appropriate routing scheme based on their
requirements (Table VI and V). 1) ODMA: The initial idea of extending the coverage area
6) A discussion on some future research directions for the of the network to more than one hop, named as opportunity
routing in D2D networks (Section VII). driven multiple access (ODMA) is first proposed by South

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

African scientists in 1978 [19]. The idea was then discussed TABLE II: Acronyms of topology based routing algorithms.
by the 3GPP standardization committee in 1999 [20]. Algorithm Algorithm Name
ACMR Adaptive Cellular and Multi-hop Routing
2) MCN: In multi-hop cellular network (MCN), cellular ARCE Ad hoc Routing for Coverage Extension
users not only forward the data of BSs as in ODMA, but also BCR Base-Centric Routing
forward data of other nodes for inter-device communications CAR Centralized Adaptive Routing
CBSR Cellular Based Source Routing
[21]. ContouR Contour Routing
CRL CRoss Layer routing
CRSP Combined Routing, channel Scheduling and
3) IMCDN: In integrated multi-hop cellular data network Power control
(IMCDN), large number of fixed relay nodes are intelligently DST Distributed Spanning Tree
placed to efficiently deliver data from cellular users to the ERP Efficient Routing Protocol
FPS Fixed Power Scheme
BS, regardless of the cellular user positions. The relay node FRS Fixed Rate Scheme
is named as routing node (RN) and cellular user node is HyLoMM Hybrid Location-based Multiple Metric
named as mobile data terminal (MDT) in the original text. IAC Inhibit Access Control
IAR Interference-Aware Routing
The objective is to efficiently balance the load among the ICR Interference-Conscious Routing
BSs, instead of power saving or high mobility support (as JRC Joint Routing and Channel selection
are the objectives in mobile ad hoc networks (MANETs)). JRS Joint Routing and Scheduling
LBS-AOMDV Load Balancing-based Selective - Ad-hoc On-
Mobile switching centre (MSC) acts as central controller. demand Multipath Distance Vector routing
All link layer and network layer decisions are made at the LDR Light Dark Routing
MSC. This reduces the topology management tasks from the LoMM Location based Multiple Metric
LSSON Large Scale Self Organizing Network routing
BSs and RNs, which helps in making the network simpler, MCN Multi-hop Cellular Network Routing
power efficient and cost effective. The upstream packets are MCPR Minimum Consumed Power Routing
always destined at BS and the downstream packets are always MMRP Multiple Metric Routing Protocol
MRC Multiple Route Coding scheme
originated from BS [22], [23]. MR-D Maximum-Rate towards Destination
MR-DA MR-D Advanced
4) ICN: In integrated cellular networks (ICNs), operator PAC Priority Access Control
PIR Power and Interference-aware Routing
installed fixed relay nodes and/or cellular users are used to PSF Power-strength-based Selective Forwarding
forward the data of heavy loaded BS to another less loaded BS. RAR Resource-Aware Routing
When there is a high traffic load and BSs have no additional RCIL Routing with Combined Interference and Link
cellular resources to fulfill new resource requests, BS finds a RRR Reception Range-aware Routing
multi-hop route to a less loaded BS in case of a new resource xLoMM x-hop location permissive Location based Multi-
request from a node. The fixed relay node is named as traffic ple Metric
diversion station (TDS) and cellular user node is named as
mobile host (MH) in the original text.
Throughout this paper, we use the same naming notations as 1) Natural disaster: Different types of natural disasters
used in the original research papers, however, it is important have been listed in order of the magnitude of fatality in [45]. A
to remember that all the proposed schemes under different smartphone disaster recovery system (SDRS) protocol [45] is
naming conventions can be directly used with D2D commu- proposed to communicate with the BS, or otherwise establish
nications. D2D communication (using LTE-Direct) in case of natural
disasters. An emergency communication protocol for D2D
communications has been proposed in [46] which establishes
B. Comparison with Existing Survey Articles a cluster like network topology of D2D users. A node called
as chief D2D node (usually the first node in the network) is
A number of survey/review papers have been published to responsible for creating and maintaining the D2D network.
discuss the basics of D2D communications, and to discuss However, the high energy consumption of the chief node in
various other D2D research domains. Table III demonstrates maintaining the topology has not been considered. Multi-hop
the efforts of most of the existing survey papers related to D2D communication for extending the network coverage has
D2D. However, to the best of our knowledge, there is no been proposed in [47]. Study has shown that the network-level
survey paper on routing in D2D networks and this is the success probability increases with the increase in the number
first comprehensive survey paper which discuss the routing of relays against the varying values of the damage ratio caused
in multi-hop cellular D2D networks. by the natural disaster.
A survey on disaster response network for cognitive
radio (CR) [48] provides many new ideas that the D2D
C. Implementation Scenarios for D2D communications
communication networks can use to get benefit from. A few
There are a number of use case scenarios where D2D of them are discussed below:
communication is the best choice for implementation, however
we discuss three major scenarios, which are provided below: a) Immediate services: A number of disaster response
network (DRN) solutions exist, provided by various dis-

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

TABLE III: Comparison of existing surveys/reviews in D2D communications. All these survey articles generally discuss
different aspects of D2D in general, none of them discuss routing in D2D networks.
Citation Publication Discussed Topics Description
[24] 2014 Generally discuss D2D This is the first survey paper on D2D communications that provides detailed literature review based on
communications inband (underlay and overlay) and outband spectrum band. Use-case, platform, performance metrics
with achieved results, evaluating and analytical tools under each category is tabled and discussed.
[25] 2014 Generally discuss D2D A very short overview on challenges for interference mitigation using mode selection, resource
communications allocation and power control for cooperative D2D communications.
[26] 2015 Generally discuss D2D The survey provides deep insight in various D2D research domains, which include but not limited
communications to mode selection, interference management, power consumption and energy efficiency, relay
functionality and 3GPP standardization activities.
[27] 2015 Generally discuss D2D This paper surveys centralized and distributed interference management schemes for in-band and out-
communications of-band D2D communications, and also tabled and discussed the existing radio resource management
schemes. Prototypes and experiments for the implementation of D2D communications, 3GPP
standardization and candidate radio technologies for D2D communications are also discussed.
[28] 2015 Channel measurements The paper provides the current state of the research on D2D channel measurements in various different
and models in D2D scenarios like macro-cell/micro-cell outdoor-to-outdoor (O2O), outdoor-to-indoor (O2I), indoor-to-
communications indoor (I2I) for urban, suburban, rural, office, shopping mall, and urban-road/highway vehicle-to-
vehicle (V2V), and line-of-sight (LOS)/non-line-of-sight (NLOS) scenarios. The paper also surveys
state-of-the-art research work on D2D channel models for various scenarios.
[29] 2015 Offload techniques in In this paper, D2D is compared with RAN technologies like WiFi and small (femto) cell, in terms of
wireless access and core various criteria like QoS, coverage and data rate etc. D2D is also compared in terms of strength and
networks weakness with mobility and interworking protocols (like Mobile IP (MIP)/Proxy MIP (PMIP) used
by Interworking WLAN (I-WLAN), and IP Flow Mobility (IFOM)) and with core network offload
approaches (like Selective IP Traffic Offload (SIPTO) by implementing a Traffic Offload Function
[30] 2016 Separation framework The control and data plane separation architecture (SARC) have been proposed as an enabler for
for cooperative and D2D coordinated multipoint (CoMP) and D2D communications in the context of 5G. Various SARC
Communications proposals have been discussed to alleviate the problems related to energy efficiency, mobility
management, interference management and capacity maximization.
[31] 2016 Interference management The paper provides detailed discussion about interference mitigation strategies in D2D communica-
in D2D communications tions based on two broad categories, co-tier (same tier i.e. between two D2D users) and cross-tier
(different tier i.e. between D2D user and cellular user/BS) interference. Based on the quantitative
comparison, and on the strength and weakness/limitations of the literature work, future research
directions are also discussed in order to include D2D communications in 5G cellular networks.
[32] 2016 Generally discuss D2D The paper discusses in detail various features of 5G networks that can be integrated in D2D
communications communications. Open challenges in D2D communications are also discussed at the end.
[33] 2017 Security in D2D commu- The paper classifies and provides detailed discussion on the security and privacy preserving
nications approaches in D2D communications, along with their addressed attacks. Relationship in terms of
contradicting and supporting requirements between security and privacy are also identified.
[34] 2017 Network coding in D2D A new paradigm of network coding, instantly decodable network coding (IDNC) for D2D commu-
communications nications is surveyed in this paper. The paper first compares IDNC with random and opportunistic
coding and then compares the types of IDNC schemes like strict, generalized and order-2 IDNC using
various criteria. In the second part of this paper, the use of IDNC for centralized and distributed
D2D communications is discussed in detail.
[35] 2017 Security in D2D commu- The paper provides detailed overview on various challenges for D2D communications along-with
nications the overview on security algorithms for D2D communications.
[36] 2017 Generally discuss D2D A detailed literature review on centralized and distributed schemes for mode selection, interference
communications and resource management, and power control for inband D2D communications has been provided.
[37] 2017 Generally discuss D2D A very short overview on the classification of D2D communications based on control and frequency
communications perspective. A brief view of D2D hypothesis and analysis techniques is also presented.
[38] 2017 Security in D2D commu- The paper compares the advantages and disadvantages of state-of-the-art security solutions proposed
nications in literature, according to the security architecture and requirements.
[39] 2017 Green communication in In addition to other aspects discussed in the perspective of green networking, D2D communication
cellular networks has also been considered a competent technology of 5G. A short review of some example works to
enhance energy efficiency in mechanisms like mode selection, resource allocation and interference
mitigation in D2D communications have been discussed.
[40] 2017 Resource management in Resource management in D2D communications has been overviewed in terms of system performance
D2D communications objectives like throughput maximization, battery life etc. The different kinds of optimization problems
(like mixed integer linear programming, stochastic programming etc.) and their solution types (like
heuristic, evolutionary etc.) to optimize the D2D communication performance are also discussed.
[41] 2017 Security in D2D commu- This paper identifies and categorizes security solutions using layer-based approach, in order to provide
nications better understanding and design of D2D communications security.
[42] 2018 Energy consumption mod- This paper provides a brief overview and analysis on the power consumption models of the UEs
els in D2D communica- and of the various types of BSs (maco, pico, femto etc.) in D2D networks. The analysis has been
tions carried out using LTE and WiFi radio access technologies (RATs).
[43] 2018 Socially-aware D2D com- The paper first provides an overview on technical challenges and applications of D2D communications
munications without taking into account any social connections. The paper then exploit the online and offline
(neighbors, colleagues etc.) social connections while considering both the selfish and the selfless
users behaviors, and discusses in detail the technical issues and application with social-awareness.
[44] 2018 Generally discuss D2D A comprehensive survey which provides state-of-the-art and open research issues of challenges
communications which include interference management, mode selection, power control, mobility management, device
discovery, economic aspects, security/privacy, and the integration of D2D with 5G technologies.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

aster response organizations. Such organizations are gener- c) Cost-effectiveness: The services provided rapidly fol-
ally telecommunication operators, telecommunication equip- lowing the disaster event are generally very expensive. Almost
ment vendors, government and non-government organizations all solutions use cost-effective satellite backhaul system to
(NGOs). These organizations aim to enhance resilience against provide connectivity. These services are usually replaced by
natural and man-made disasters, usually provide communica- less-expensive solutions in the later weeks once the situation
tion services within 24 hours to affected areas in the immediate is under control. Thus the overall deployment of all such
aftermath of a disaster. Any delay in the response to the solutions incurs high costs.
disaster may result in further loss of life, injury and damage. D2D network can be deployed without any extra cost. The
AT&T’s national disaster recovery (NDR) program provides normal smartphones of users are used as D2D end-devices to
LAN, WiFi and voice (VoIP) connectivity within minutes of communicate directly with other D2D devices and to provide
arrival with dedicated generators. This connectivity is possible the services of a relay.
via solution called emergency communication vehicles (ECVs)
that uses satellite links. However the arrival of relief personnel d) Self-Organization: It is worth noting that most DRN
/ disaster response organizations at the affected areas always solutions lacks the capability of self-organization. The oper-
incur delay, and thus no connectivity is possible before that ating environment in the disaster area is unpredictable due to
time. Furthermore, the deployment might incur additional dynamicity in user demands, network topology and spectrum
delays due to the lack of network damages information, availability.
prior government and mobile operator agreement for spectrum D2D communication provides flexible reconfigurations
utilization and custom clearance. Thus these external solutions by observing the operating environments. Such operating
fail to provide an immediate first-aid recovery to the affected information can be accumulated through local sensing,
areas. individually and by cooperation among neighbor devices.
Local personnel who are less affected by disaster can Thus, local sensing is expected to play a vital role.
immediately start the first-aid relief operations for the D2D devices can then select appropriate strategies to
affected persons using D2D communication. These persons achieve specific goals. These self-organizing capabilities are
benefiting from the auto-configuring capability of D2D especially valued in the context of DRN, where immediate
devices can search the affected persons by discovering communication is required with little or no information
the proximate D2D devices. They can also check and about the operating environment and which may change
communicate with their close relatives and other personnel unpredictably. Auto-configuration capability of D2D devices
by utilizing the multi-hop communicating capability of D2D helps in quick adaption to the changes in operating
devices. D2D devices can also communicate multi-hop to any environment. This also helps in minimizing time consumption
of the nearby base stations or relief-centers. in the initial configurations.

2) Public safety: D2D communication has mainly emerged

b) Wide area coverage: National disasters usually spans to offload the high data traffic from the cellular network, and
wide geographical regions. For example Hurricane Katrina to provide pubic safety communication services in case of
affects 230,000 area in the US in 2005. The relief natural or man-made disaster. Most of the works have focused
personnel require wide area connectivity to coordinate relief on providing D2D communications in case of natural disasters
services. The network coverage provided by all the available when the cellular network services are either partially or
DRN solutions is limited. Due to the damaged civil infrastruc- totally unavailable [11], [49]–[63]. Very few works considered
ture, it might take several days to provide relief in all affected human-made disasters like fires [7], [64], accidents, terrorist
areas. D2D network provides the best connectivity solution attacks [8] etc. Tata and Kadoch [7] highlight the fact that
in disaster response network when the cellular coverage is many firefighters die because they cannot get out of the fire
partially unavailable or when the cell coverage does not spot. This is because they are unable to communicate due to
provide connectivity over the complete emergency area. We network congestion caused by the peak usage of the network
are considering the case when only a limited amount of cellular in the locality. Castel et al. [64] analyze the LTE-D2D as
sites are functional following the disaster event. A D2D a potential candidate standard for communication between
network acts as a glue to repair a damaged cellular network by the indoor body-to-body link (by considering firefighters as
providing extended coverage. Such network could also replace an example). Yuan et al. [8] point our attention towards the
damaged cellular sites in terms of proximity communications. terrorist attacks. Most of the terrorist attacks are targeted in
Highly-flexible D2D devices can provide multiple services urban areas in order to have higher impact and causalities.
reusing the same cellular spectrum. This leads the networks to be overloaded or damaged. D2D
As DRNs often provide limited connectivity in some target communication provides an alternative communication in
areas, forming sparse networks. Thus, the incorporation case of such network congestion or network unavailability.
of D2D devices with multi-hop support provides extended
coverage with high bandwidth efficiency. This involves 3) Expected and unexpected events and crowds: D2D tech-
challenges like next hop selection, channel selection, end-to- nology is able to fulfill the communication needs in case of
end delay, mobility of users and interference avoidance. both the ‘expected’ (planned) and the ‘unexpected’ events
and large crowds. The planned events could be pre-organized

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

seminars, conferences, social or personal parties, parades, road both shadowing and fast fading (increased Doppler effect).
skating or other sport events in stadiums, where the time and Thus mobility is more difficult to manage in small coverage
the crowd volume could be pre-estimated. Public speech or scenarios than with respect to their respective large coverage
huge sales event are also planed events, however it is hard areas of the BS [65].
to estimate the exact number of people in such events. The
unexpected events could be traffic jam or rally, where both 5) Extent of transfer of control from Network operators
the duration and the number of people involved are hard to to Users: Transferring the network control to users is a
predict. Thus, D2D is a quite flexible technology that can big topic of discussion by the telecommunication operators.
survive and provide communication services for all types of Since losing control over the users is never the desire
planned (normal) and the unplanned (emergency) events and of telecommunication operators. Thus further studies are
crowds. required to investigate optimal solutions for splitting the
The above adaptability of D2D in such diverse scenarios control functionality between the network and the UE [65].
is one of the very big advantage to the telecommunication
operators. For providing better services, operators need to 6) Impact on WAN as a whole: Mostly, all cellular
investigate the network requirements like number of users technologies are designed to optimized eNB (evolved Node
and user mobility of each location independently. For example B) - UE links. However finding the impact of these techniques
Train/Bus stations have higher number of users and high user on WAN also need considerations [65].
mobility, while academic institutions or offices have also high
number of users but comparatively low mobility. Shopping 7) Impact in Complex Heterogeneous Networks: In the
markets have high user mobility but varying number of users presence of macro cell, micro cells, home cells and fixed
in various times, and residential areas have low number of relays, the impact of the wide deployment of D2D communi-
users and low mobility. Thus, by using D2D communication cations also needs further considerations [24].
technique, network operators may alleviate some of the burden
of detailed network planning due to the D2D ability to provide E. Performance improvements and incentives in D2D commu-
communication services in any kind of network requirements. nications
D2D communication is advantageous for both the operators
D. Unique Challenges as well as the users. It gives incentives to users for relaying
D2D communications is still in its infancy. A number the data of other users and is beneficial for network operators
of open issues and problems are there which need serious in terms of network performance enhancements.
attentions. This section discusses such challenges. In traditional cellular communication systems, it is hard
to fulfill all the inter-dependent network performance metrics
1) Inter-operator or Multi-operator D2D operation: requirements. For example, in interference-limited systems,
Without inter-operator support, the applicability of D2D the coverage of the cell decreases with the increase in the
communications is very limited. In multi-operator D2D number of users. Thus, there is always a trade-off between
operation; which spectrum to use and how to coordinate and the capacity and cell coverage. With the introduction of D2D
control UEs operations are key issues to solve. communication, drastic performance gains are achieved in
various network metric and there is no longer a strict trade-
off between most of the network requirements. Three types of
2) Co-existence and interference management: Interference
gains are promised by D2D communications [66] which are
management between the coexisting D2D networks and
proximity gain, hop gain and reuse gain. The proximity gain
between D2D networks and cellular networks needs detailed
assures high data rates, low energy consumption and delay
investigations. Furthermore, resource allocations for D2D
by taking advantage of the proximity of the other device.
network within a single cell or across multi cells also needs
The reuse gain indicates the reuse of cellular spectrum for
serious attention. Although massive amount of research on
D2D communications. The hop gain is achieved by using
this issue has already been performed, however this issue
a single-hop for D2D communications instead of two-hops,
needs further considerations.
an uplink and a downlink resource in cellular mode. This
allows the flexibility to use any available channel (i.e., uplink
3) Mobility management: In case of mobility, new D2D
or downlink) for D2D communication. We now discuss the
handover criterion in addition to traditional handover is
performance improvements that could be achieved in both the
required. D2D-aware handover and D2D-triggered handover
single-hop and multi-hop D2D communications.
are some initial solutions in this domain which also needs
F. Single-hop Performance Improvements
4) Dual mobility: In traditional cellular communications, When two devices directly communicate with each other
one communicating entity that is the BS is always fixed and (i.e., single-hop communication), the two performance
the other communicating entity that is the UE is mobile. improvements that can be achieved are reduced delay and
However, in D2D communications, both communicating increased energy efficiency.
entities are the UEs and thus both are mobile. This affects

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

1) Reduced delay: Direct single-hop link between the two

devices in D2D communication ensures reduced delay as
compared to the minimum two-hop link in traditional cellular

2) Increased energy efficiency: Instead of using higher

power by each device to communicate with BS in traditional
cellular communication, D2D devices use lower power due to
smaller distance between each other in D2D communications
which increases the energy efficiency of D2D devices.

G. Multi-hop Performance Improvements

When two D2D devices want to communicate are not in
the vicinity of each other, other D2D devices are used as
relays, which makes a multi-hop communication between the
two D2D devices. The performance improvement achieved in
multi-hop communication is increased coverage. Fig. 1: Modes of multi-hop D2D communication.

1) Increased coverage: In multi-hop D2D communication,

D2D devices which are out of the coverage or in the dead 4) Network Capacity: Network capacity refers to the
spot regions could use other D2D devices as relays to com- number of users (or specifically the amount of traffic) the
municate with the infrastructure network or to communicate network can handle simultaneously. As we know that Capacity
with other end-users, resulting in overall network expansion = Bandwidth × Spectral Efficiency; and we also know that
and increased network coverage. maximum bandwidth is usually given/fixed for a network,
therefore the increase in the spectral efficiency results in the
overall increase in the network capacity. It also increases the
H. Single and Multi-hop Performance Improvements
overall system capacity by bypassing blocked base stations
In addition to the above performance improvements which [68].
are applicable either only to single-hop or multi-hop. There
are many other benefits which are applicable to any kind 5) Network Load: D2D communication always reduces
of network either established using single-hop or multi-hop the load on the network. This not only offload the traffic
D2D communications, which are discussed below: from the BSs but also offload the backhaul and the wide area
network, which further results in cost efficiency [69].
1) Quality of Service: Devices in the proximity of each
other, notice high signal to interference and noise ratio (SINR) 6) Availability of Service: In case of no/partial network
as compared to the traditional links with the cellular network. service, D2D communication serves as a fall back solution
However, the devices must use an efficient interference which enable devices to communicate directly or via relay(s).
mitigation scheme in order to be aware of and mitigate
the interference [12]. Field tests [67] also demonstrated
the improved end user QoS in the cell-overlaid and in the I. Incentives provided by BS to network relays
handover areas. To motivate the users to assist in relaying data of the BS
and other users, various kinds of incentives are provided by
2) Data rate: Devices in close proximity can use higher the BS and the users. The network relays are those users which
order modulation and coding schemes which leads to relay data for the BS, and D2D relays are those users which
sufficiently high data rates [9], [13]. relay data for other users. A number of incentive mechanisms
have been proposed in literature [70]–[73].
3) Spectrum Efficiency: The hop gain immediately doubles In compensation of the consumed resources of the user
the spectrum usage because of saving one spectrum resource offering the services as relay, following benefits could be
in each D2D communications. The reuse gain further improves provided by the BS to users involved in relaying data.
the spectrum efficiency by the effective use of scarce cellular
spectrum. Multiple D2D links can again reuse the same 1) Monetary incentives: Financial benefits like extra
spectrum by exploiting the spatial diversity. For multi-hop balance, discount on monthly bills, or some cash prizes could
D2D communication, Babub et. al [11] have verified through be rewarded to the users involved in relaying the data of BS.
results that the spectral efficiency increases with the increase
in number of hops (upto four hops considered in simulation). 2) Service incentives: Package promotions, increased band-
width and/or bonus minutes or SMSs can be rewarded.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials

Fig. 2: Classification of D2D communications.

J. Incentives provided by other D2D users for the user which has to provide relay service, and if the
If a user is outside the coverage of the BS or is at some meetings are quite often, then provide the relay services. This
dead spot location, it can utilize the services of other users, is to ensure that the user also gets the same reward of service
either to connect to the infrastructure network (i.e., BS), or from the user currently providing the relay services. Such
to connect with other users in a multi-hop manner. A number behavior creates uncertainty in relay services and therefore
of research works focusing on incentivizing the users to such devices are better to be off. Complete selfishness in
act as relay are present in literature. Some of the attractive which users demand some returns at the same instant does not
incentives are: create uncertain relay situations and therefore such devices
are better than those discussed before.
1) Direct bandwidth exchange: The user who uses the
service of a relay node transfers the bandwidth (or could 4) Token-based: In this approach, tokens are used as virtual
be amount) to the relay node immediately. The transferred currency [74]. Tokens are required in order to use the relaying
bandwidth will be equal to the bandwidth that is consumed service of other users and tokens are transferred on getting
by relay node in servicing as relay plus extra bandwidth. service from other users. The tokens received by a user can
This extra bandwidth is the reward of the resources (battery, be used later when a user requires relaying service from other
processing, time and storage) consumed during relaying users. Tokens are used only for D2D communications, they
operations. However, if the relay node already has sufficient cannot be used as an alternate of the money. This necessitates
resources such that it does not need any reward immediately, either to buy tokens from the operator (operator benefit) or
then the immediate bandwidth exchange will not be useful to earn tokens by servicing as a relay (motivates users to
here. relay data). The price of token needs to be set such that it
does not make burden to buyers, but it is still attractive for
users to provide service for relaying data for getting the same
2) Social-tie based: A user can relay the data of other
tokens. Thus, the decision of using this approach depends on
users within his/her social circles (e.g, friends, relatives etc.)
(a) service gains comparative to traditional communication
to benefit them. This is no financial or service benefit for
by receiving data through a relay (from the perspective of
the relay node in this approach, rather it is based on social
service node) and (b) if the future benefits outweigh current
relations with each other.
resource costs it incurs for relaying data for others users
(from the perspective of relay node). Results suggest that
3) Reputation based: In this approach, reputation level is
relaying cooperation is maximized when (a) the users have
maintained according to the participation in relaying service.
high relay energy budgets, (b) the users are highly mobile (e.g.
Thus, for users having very low reputation of relaying,
on vehicles) and (c) there are moderate number of tokens.
others might avoid providing relaying service. However,
false reputation and selfish behavior of users are the main
downsides of this approach. In false reputation, the users III. C LASSIFICATION OF D2D COMMUNICATIONS
misrepresent their reputation level. This problem could This section classifies D2D communications into three main
be avoided in supervised communication where BS can characteristics, i.e., type of communication, number of hops
track of the reputation of each user, while in unsupervised and type of spectrum. These characteristics are further clas-
communication, the problem is not easy to be dealt with. In sified into sub-categories which are illustrated in Fig. 2 and
selfish behavior, a users calculates meetings with deviations discussed below. It is important to note that the nesting level

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


of each sub-category has been mentioned in braces with the are discussed in detail below. The type of spectrum used in
name of the category, for quick referral from the figure. state-of-the-art topology based routing algorithms is listed in
Table VI.
A. Type of Communication (1)
There can be two types of D2D communications, i.e., either E. Inband (2.1)
supervised under the control of the BS or unsupervised when In inband communication, devices use available cellular
node is out of the coverage of the cellular network. Node channels opportunistically. The use of same spectrum en-
under the coverage of the BS can also be tuned to work in force easier cooperation among D2D and cellular networks.
unsupervised mode by not utilizing the services of the BS. BSs have higher control over the interference between the
coexisting D2D and cellular networks, thus QoS can be
B. Supervised communication (1.1) guaranteed. The same cellular interface is utilized for D2D
In supervised communication, the BS controls the commu- communications, so every device can use inband spectrum.
nication, and guarantees performance and security by complete There are two techniques to use the inband communications:
control over the control plane and the data plane [75], [76]. In overlay and underlay.
control plane, BS authenticates the devices and set appropriate
encryption for privacy. Connection establishment and mainte- F. Overlay (2.1.1)
nance is also the responsibility of BS. Due to global network
knowledge, issues in unsupervised (ad hoc) communication In this technique, spectrum resources are dedicated for D2D
like synchronization, collision avoidance and free spectrum communications and separate spectrum are scheduled for both
detection can be easily handled by the BS at the control plane. the cellular and the D2D networks. This ensures no inter-
In data plane, resource allocation or spectrum management by ference with cellular devices. If many D2D communications
the BS ensures that UEs do not have the burden of handover opportunities are possible, the same spectrum can be reused
in case of information brokerage. Mobility management is spatially within D2D communications.
also the responsibility of BS at the data plane. All these
services are not possible in infrastructure-less (or purely ad- G. Underlay (2.1.2)
hoc) networks like WiFi Direct or Bluetooth. Additionally, In this technique, spectrum resources are reused for D2D
large scale network communications is also not suitable for communications opportunistically. This increases spectral ef-
the infrastructure-less networks. ficiency by exploiting the spatial diversity. However, interfer-
There are two types of supervised communication: network- ence mitigation involves detailed investigations both at the BS
based communication and network-assisted communication. In and the D2D UEs. Underlay has two types of spectrum: uplink
network-based communication, all the devices are under the and downlink. In frequency division duplex (FDD) mode,
full control of network. In network-assisted communication, the reuse of downlink and uplink spectrum needs different
devices can take decisions autonomously, however their deci- considerations.
sions are based on the measurements provided by the network.

C. Unsupervised communication (1.2) H. Downlink (

Currently unsupervised communication is allowed to D2D The reuse of downlink spectrum has very stringent re-
devices only when infrastructure network is unavailable. In quirements because the D2D UEs might severely interfere
unsupervised mode, the network is stand-alone and works the cellular UEs which is not tolerable. On the other hand,
exactly like ad-hoc networks. Such network does not need D2D UEs also receive severe interference from the BS due
any expensive infrastructure for communications. Moreover, to the high transmission power of BS. Therefore, this D2D
the network is also quite flexible, and does not have any re- communication is usually possible near the cell edges. D2D
strictions due to unavailability of any centralized management UEs are also affected by the interference from other D2D UEs
and control. For example, in ad-hoc network, where all nodes of the same cell, and from the BS and D2D UEs of other cells.
works autonomously, failure of any node have negligible effect
on the overall network performance while in infrastructure I. Uplink (
based network, failure of a central processing node, results
In this type of communication, D2D UEs get most harmful
in total network failure. Similarly, in case of overload on
interference from the cellular UE (CUE). D2D UEs are also
the nodes in ad-hoc network, the network performance is
affected by the interference from other D2D UEs of the
slightly degraded, however overload on the central node in
same cell, and from the CUEs and D2D UEs of other cells.
the infrastructure-based network results in significant service
Conversely BS is interfered by all D2D UEs of same and
degradation and limited service availability.
other cells. However BSs are very powerful and can handle
interference efficiently. The use of uplink in D2D communi-
D. Type of spectrum (2) cation is the most widely used technique in the literature due
D2D devices can communicate in either of the two types to its numerous advantages. This is preferred to use for D2D
of spectrum i.e. inband or outband. These types of spectrum communications because:

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


• D2D UEs have reduce impact on BS and thus interference L. Unused (2.2.2)
tolerance is acceptable to some extent.
It is permitted by the FCC to use the licensed channels
• The uplink band is underutilized and thus have more free
(usually the TV white spaces) opportunistically provided that
it does not interfere with the licensed users (called primary
• UEs radio interface is by-default designed to send data
users) holding the license for certain time and frequency.
in Uplink [65]. Thus it would be complex if UEs had
Normally in BS-assisted D2D communication for the open
to send data in downlink band which requires OFDMA
access of TV white space, BS senses the available unused
transmitter. Moreover, SC-FDMA transmitter enjoy low
licensed spectrum in the cell using efficient spectrum sensing
peak-to-average power ratio (PAPR).
techniques. Subsequently, the unused available channels can be
• 3GPP also recommends to use uplink band for D2D
advised to the D2D UEs to be used for D2D communications
communications [4].
However, there are few constraints in SC-FDMA (uplink
band). The resource blocks (RBs) in frequency domain should
be contiguous and must use the same modulation scheme [77]. M. Number of hops (3)
Morever, the SC-FDMA receiver is also complex because of
A communication can either be single-hop or multi-hop. In
single carrier transmission.
single-hop communication, two devices are in the proximity
of each other and directly communicate without needing any
J. Outband (2.2) relay. In multi-hop communication, a device which want to
communicate with another device or BS is not in the proximity
In outband communication, devices use unlicensed spec- of destination node, therefore it needs atleast one intermediate
trum instead of cellular spectrum. Outband communication node to assist it in relaying its data to another device or BS.
has several advantages in D2D communications which are As discussed in Subsections II-F to II-H, single-hop and multi-
summarized below: hop have their own pros and cons. The advantages of single-
• Precious cellular resources are preserved. hop communication is reduced delay and increased energy
• No interference with the cellular resources. efficient. However, it is not also possible to have single-hop
• Simultaneous D2D and cellular communication is possi- communication when the devices which want to communicate
ble. are not in the proximity of each other. The main advantage
• No considerations of frequency, time and power are of multi-hop communication is increased network coverage.
required while scheduling outband communication. However, it incurs additional delay due to higher number of
transmissions by relay nodes.
However, on the contrary, outband communication has some
limitations as well which are summarized below.
• It consumes double power consumption due to using IV. T HE STATE - OF - THE -A RT IN M ULTI - HOP D2D
two interfaces. Thus, special techniques are required to ROUTING
minimize the power consumption.
The single-hop D2D communication limits the users to
• The second radio interface needs to be coordinated which
communicate with only nodes in the proximity. Therefore,
needs investigation of inter-technology architectural de-
if a user wants to communicate with another user out of
its communication range, multi-hop D2D communication is
• It causes some delay due to increased processing because
the best choice. In multi-hop D2D routing, a route has been
protocols are not same and thus packets are decoded and
established between two nodes which are not in the proximity
encoded again. However, we assume processing delay
of each other. The route decision could be (a) centralized,
is very small as compared to communication delay in
where the route decision would be solely taken by a centralized
inband communication.
entity (like BS, if both the end-nodes are under the coverage
Outband communication would be carried out either on of the same BS), or (b) distributed, where the route decision
license free band that is Industrial, Scientific and Medical could be taken autonomously by each node while taking
(ISM) band or an opportunistic reuse of licensed bands, as into account the presence of other nodes (including the BS).
permitted by FCC. Thus, the distributed routing does not face the problem of
network failure due to a single processing element (i.e. BS)
failure. Lastly, the route decision could also be the hybrid of
K. Unlicensed (2.2.1) centralized and distributed routing.
Due to the free license spectrum, a number of works use The state-of-the-art multi-hop D2D routing schemes have
wireless LAN spectrum for direct communication among the been classified into incentive-based, security-based, content-
users. The unlicensed spectrum is usually used for D2D based, location-based and flat topology-based routing. The
communications in networks having low number of nodes. flat topology-based routing schemes have been further sub-
This is because in case of high node density, the cross- classified based on the route discovery mechanism used, i.e.,
interference with other outband users is uncontrollable and (i) reactive routing, and (ii) proactive routing, and a variant
hence QoS provisioning could not be guaranteed. of these, which are (iii) hybrid routing (both the reactive and

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


Fig. 3: Taxonomy of state-of-the-art multi-hop D2D and D2I/I2D routing schemes.

proactive routing operate at the same time), and (iv) adap- only D2D routing schemes or D2I/I2D routing schemes.
tive routing (routing switches between reactive and proactive
routing depending upon the network dynamics).
A. Incentive-based Routing (1)
Reactive routing is used in networks having low number
of data transmissions or low mobility or in networks having The objectives of incentive based routing is to establish end-
high node density and vice versa for proactive routing. The to-end communication among the devices by motivating the
adaptive routing takes benefit from both proactive and reactive intermediate users to relay the data using different incentives.
routing, and switches the routing mechanism between reactive Social connection with other nodes is also used as a type of
and proactive routing depending upon network dynamics. The incentive to motivate other nodes to relay their data, and the
hybrid routing is used in large-scale networks where proactive routing is called social connection based routing.
routing is used to maintain neighborhood table for nodes in a
zone within few hops, while reactive routing is used to find B. Social connection-based routing (1.1)
route outside that zone. In social connection based routing scheme, the relay node
When confidentiality, authenticity and integrity of data are forwards the data based on strength of the social bond with
of utmost concern, security-based routing is used. If the the other nodes on the communication path. A relay node has
objective is to share the frequently used data among users, the given the independence to decide to relay the data or not. Thus,
routing scheme is called content-based routing. The routing the probability of establishing an end-to-end communication
scheme in which the users are motivated to participate in path depends upon the social ties of intermediate relay nodes
relaying the data of other nodes by using some incentive, is with each other. The state-of-the art schemes that uses social
called incentive-based routing. If the incentive is the social connection based routing are discussed below.
bond between the users, the routing is called social connection- A centralized social-tie based routing scheme in which
based routing, and is considered as a sub-class of incentive- the route is determined by BS is proposed in [79]. It is
based routing. assumed that mobile users are connected with social ties and
The complete taxonomy of all state-of-the-art multi-hop the network operators can extract the social ties between the
D2D and D2I/I2D routing schemes is shown in Fig. 3. Each users (by using some mobile apps, to find the contacts and
category has been numbered in order to have a quick referral activities of the user’s online social network (OSN) accounts,
from the text. In the figure, the name of category is followed by e.g., Facebook, Twitter etc.). Thus, network operators can
a two-column row, where references in the first (left) column build social network graph which is used in conjunction with
shows the D2D routing schemes while the second column network connectivity graph to find the probability of data
(i.e. column on the right) shows references for the schemes packet delivery under different percentages of possible D2D
proposed for D2I/I2D routing. It is very important to note that paths. Although the route decision is under the control of
the category numbers assignment does not taken differently network operators, however, the unexpected user behavior, due
the D2D and the D2I/I2D routing schemes, therefore readers to for example low remaining energy budget, makes the end-
might notice gaps in category numbers while going through to-end delivery uncertain. The objective is to minimize the

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


total communication cost (using the communication graph) and shared a secret key with neighbor mobile node. A session
while guaranteeing the delivery probability (using the social key is generated between the source and destination nodes so
connection graph). Results showed that for the increasing that intermediate nodes can not intercept the data. The shared
probability of delivery, the communication cost also increases, key shared between the neighbors nodes requires private key
while the percentage of D2D paths decreases. assigned from the BS. Thus, only authenticated nodes from BS
A greedy algorithm is proposed to decrease the delay and can be the next hop node. For node anonymity, i.e. to prevent
the average-transmission path length (APL) in terms of number revealing the the real identity of the node, every node generates
of hops by introducing long-range links (LLs) for multi-hop temporary identity (ID) by themselves using the private key
D2D communication networks [80]. To reduce the computa- of the BS. For integrity, a hash value is appended with the
tional complexity, the problem has been modeled as a coalition message, which is generated using the shared neighbor key
graph game. The LLs have been established by exploiting the and also the private key of the sender node.
social information. In social networks, community represents The route discovery processed as follows. The source node
the group of persons having common interests. Thus, LLs have sends the RREQ packet to each neighbor node using its
been established between the same community, as information temporary ID and the shared secret key with the neighbor
sharing in intra-community is frequent. Results showed that node. The RREQ packet also contains the calculated hash
by using LLs, the number of hops have been significantly value. Intermediate node on receiving the RREQ, decrypts the
decreased between source and destination nodes. packet and check the hash value. On successful verification, a
A group meeting aware routing (Groups-Net) protocol for new hash value is generated and the RREQ is enrypted using
delay tolerant data like video advertisements or non-critical the shared key with the next hp node. If the contents of the
application updates for multi-hop D2D networks is proposed message within RREQ could be decrypted by the receiving
in [81]. The routing protocol is aware of the social group node, then the node is the destination node. In case of receiving
meetings by using proximity traces, like classrooms, office multiple RREQs, RREQ with the minimum hops is selected
buildings etc. Since normally the persons are involved in and the route reply is sent back to the source node. The
some specific schedules and routines, so it is expected to same process of encryption using shard neighbor key and
have group meetings with regularity or predictability. The adding hash value is used for sending the RREP packet. The
forwarding algorithm forwards the data, if the next hop node simulations have been performed on network simulator NS-
had some meetings with the destination node in the near past. 2, and the results promise improve throughput and reduced
As opposed to the normal trend, the algorithm does not require routing overhead.
social community group detection, and is argued to be an
expensive and complex task. Results showed that Groups- D. Topology-based Routing (4)
Net is a promising strategy in large scale scenarios, where
it reduced the overall network (or data) overhead. The state-of-the-art topology-based routing schemes for
A brief summary/comparison between the schemes pro- multi-hop D2D communications have been categorized into
posed for social connection-based routing is as follows. Kout- location-based routing and flat topology-based routing, as
sopoulos et al. [79] exploit the social ties between the mobile discussed below.
users to draw the social network graph. The social network
graph is used in conjunction with network connectivity graph E. Flat topology routing (4.1)
to find the probability of data packet delivery under different When nodes in the network do not have any specific
percentages of possible D2D paths. Zhao et al. [80] groups structure, nor have any location awareness mechanism then
the social ties having common interests into the community such arbitrary distribution of nodes is referred as flat topology
because the information sharing between intra-community is network structure, and the routing is called flat topology
frequent. The long-range links have been established between routing. Flat topology routing has been divided into reactive
the same community, which significantly reduce the number and proactive routing, each with further sub-categories as listed
of hops and delay. Nunes et al. [81] exploit the schedules in Table IV, and discussed below. The Table IV also lists
and routines of people to predict their group meetings, and the performance parameters and simulation tool used in each
the routing using this information is named as group meet- scheme.
ing aware routing protocol (Group-Net). The Groups-Net is
a promising strategy for delay tolerant data in large scale
scenarios, and it reduces the overall network overhead. F. Reactive distributed routing (
In this type of routing, autonomous nodes find route only
when required. The routing protocols proposed under reactive
C. Security-based Routing (2) distributed routing are discussed as follows.
Security of data in terms of confidentiality, authenticity and Kaufman et al. [83] proposed a routing scheme which we
integrity of data are of critical concerns in this age of data named as power and interference-aware routing (PIR). In this
communications. scheme, a node sends the transmit power and interference
A shortest hop based routing protocol ensuring confiden- information in the route request (RREQ) packet. The next hop
tiality, authenticity and integrity is proposed in [82]. For con- node forwards the RREQ if (a) the SINR of an active cellular
fidentiality, each mobile station uses symmetric cryptography link do not fall below a certain allowed threshold value k, (b)

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


TABLE IV: Comparison of topology based multi-hop D2D routing algorithms

Topology Proactive / Centralized / D2D Routing Performance parameters Simulation Tool
structure Reactive BS-assisted / Algorithms
PIR [83] Failure probability of discovering D2D MATLAB
Reactive Distributed
link, Outage probabilty of D2D users and
cellular users, Average number of hops,
Power savings with D2D
ICR [84] No performance evaluation -
MCN [85] Throughput GloMoSim
Centralized BCR [86] Throughput under BCR/UDP, FTP/TCP GloMoSim
with/without mobility
CBSR [87] Delay, Packet delivery ratio, Routing over- NS-2
Hybrid LDR [88] No performance evaluation -
Hybrid Distributed ACMR [14] Delivery rate, Average delay, Overheads NS-2
Adaptive Centralized CAR [89] No performance evaluation -
CRL [90] Total power spent per packet transmission, -
Dynamic call dropping probability, Num-
ber of nodes bathed with the interference,
End-to-end delay, End-to-end throughput,
Location based Reactive Centralized Total incentives paid
IAR [8], [91] D2D outage probability, Success probabil- -
ity, End-to-end delay, Throughput
FRS and FPS Throughput -
MR-D [93] Outage probability, Average Hop Count -
MR-DA [94] Outage probability, Average Hop Count -
JRC [95] Energy efficiency, Average hop count, NS-3
Outage probability
LoMM & Hy- Application and routing packet delivery NS-2
LoMM [96] ratio, Application throughput, Total en-
ergy consumed
xLoMM [97] Relative performance under various x-hop NS-2

the minimum SINR at the previous hop node could be achieved SINR above a minimum required SINR threshold participate.
with a given power while taking into account the interference. The proposed routing metric then selects route having high-
This ensures that bidirectional communication is possible with est SINR and lowest interference together, and thus assigns
the previous hop node. Destination node on receiving the weights according to their priority. The value of MIIS from
route request packet, replies following the reverse route order node n to m is calculated as: M IISn,m = (SIN Rn,m )α ×
to confirm the existence of bidirectional route between the ( I1m )(1−α) . The SIN Rn,m is the signal to interference and
source and destination nodes. Dynamic Source Routing (DSR) noise ratio from node n to m, as is calculated as: SIN Rn,m =
[98] is used for route discovery in which nodes which are σ 2 +Im . While ignoring noise α (σ 2 ) as constant, M IISn,m
near to the route also have the route information. This helps becomes: M IISn,m =
(RSSn,m )
. Where Im is the observed
in easily repairing the route with the help of these nearby interference for a given channel at current node m from all
routes. In case of small pathloss environment, nodes lower other cellular and D2D nodes. α is a weighting parameter. If
their transmit power to minimize interference to BS which the value of α is zero (i.e., α = 0), the routing metric behaves
results in route with multiple hops. However there are some simply as an interference metric. While if the value of α is
limitations in this work. The work is limited to single source one (i.e., α = 1), the routing metric is only an SINR metric.
of interference, multiple interfering nodes in case of multiple Thus the proposed routing metric is actually a compromise
flows are not taking into consideration. To ensure that the between SINR and interference.
interference is not above the threshold value, D2D nodes must Interference-conscious routing (ICR) scheme used DSR as
need to coordinate either with the BS or with all other D2D a basis for route discovery mechanism, while incorporating
nodes. This coordination is not explicitly mentioned in this MIIS metric as follows. The node adds observed signal
work. Since such synchronization is usually the responsibility strength of all the channels having interference below a given
of BS, thus the proposed scheme would not be applicable for threshold in the RREQ packet. The node broadcasts the RREQ
nodes outside the cellular network coverage. packet using the common control channel (CCH) which is
Shaikh and Wismüller [84] proposed a weighted sum rout- received by all the nodes within the communication range.
ing metric named as MIIS (Metric for Interference impact The next hop node on receiving the RREQ packet, matches its
and SINR). In route discovery, only those nodes having the available channels having interference below a threshold with
observed interference below a given interference threshold and the previous node’s channels. The channel having the lowest

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


interference at both nodes will be selected for communica- However, there are some serious limitations in this ap-
tion. It might be possible that there is no matching channel proach. The route updates are sent to the BS only when
available, in which case the node cannot act as relay node there is a considerable change in the received power (Rx)
and thus will not participate any more in route discovery. The of the neighbor node. So, if there is a bad link which
signal strength of the selected minimum interference channel fluctuates then the immediate route updates will result in a high
along-with the received signal strength (RSS) of the RREQ routing overhead which may severely degrade the network
packet is used to derive the SINR value. The SINR of the performance. On the other hand, when the network is stable,
previous hop node is also calculated using the interference the same route will be used for a long time which puts burden
information given in the RREQ packet. The lowest SINR from on the specific set of nodes.
the two SINRs will be selected as the SINR of the link between Base-centric routing (BCR) protocol which is a hybrid of
the two nodes. In this way, provision of minimum SINR in demand-driven and table-driven routing is proposed in [86].
bidirectional communication can be ensured. This minimum Neighbor nodes periodically exchange beacon messages with
SINR together with its respective interference value is further each other. The current link state information of neighbor
used to calculate the MIIS value. nodes in the form of link state packets (LSPs) is also peri-
The MIIS value is also appended to the RREQ packet odically unicast via multi-hop to the BS. In this way, BS uses
and the process continues until the RREQ packet reaches the table-driven method to maintain topology information of the
destination node. The destination node selects the route having whole network. Route to BS is known by the HELLO message
highest MIIS values while also taking care of hop count using periodically flooded by the BS. This route is used by LSPs,
the given formula: BROKEN LINK error packets, and unicast path request (U-
If ( hops R2 β M IISminR2
hopsR1 ) > M IISminR1 then R1 Else R2 PREQ) messages.
Where hopsR1 and hopsR2 are the number of hops in In BCR, the nodes use the demand-driven approach to find
route R1 and R2, and M IISminR1 and M IISminR2 are the the route to the required destinations. These nodes send U-
minimum values of MIIS in route R1 and R2 respectively. β PREQ to the BS, and if the route is not found, then broadcast
is a weighting factor having default value one (i.e., β = 1), route request (B-RREQ) like ad hoc on-demand distance
however it can be tuned based on varying network dynamics. vector (AODV) routing protocol. BS replies the complete route
After selecting a route, the RREP is sent back following the to the destination in the path reply (PREP) message. Since the
primary selected route. PREP is unicast to the source node, therefore intermediate
The two schemes proposed for reactive distributed routing nodes in the path from the source node to the required
are summarize as follows. Kaufman et al. [83] ascertain that destination node do not know this path. Thus the source node
the value of SINR for an active cellular link does not fall below send pointer (PTR) message, which contains the information
a threshold value, in both directions. That is, from the current of path and hop count as received from the BS in the PREP
node to the next-hop node and from the next-hop node to the message. This PTR message is not used in case of B-PREQ
current node; in order to ensure bidirectional communication. message. If an intermediate node finds a broken next hop link,
Shaikh and Wismüller [84] maintain the threshold values both it sends BROKEN LINK and path error (PERR) messages to
for the minimum SINR and maximum interference, and then the base station and source node respectively.
select a multi-hop D2D route having the highest SINR as well The results show that BCR outperforms AODV, because
as lowest interference together. path request messages are mostly unicast to the BS only, which
significantly reduces the routing overhead. Without node mo-
G. Proactive centralized routing ( bility in the network, UDP performs better than TCP. However
In proactive centralized routing, a centralized entity (for when mobility is enabled, paths are frequently broken which
example BS) regularly gathers neighbor nodes information results in data loss. Thus throughput of UDP is lower than
from all the network nodes, in order to always have a complete TCP because of not having any flow control and retransmission
updated network. mechanisms. When the node mobility is enabled and the load
One of the pioneer routing scheme, which efficiently utilize is light, the throughput of AODV is higher than BCR. This is
the services of BS for route decision is the multi-hop cellular because, BS might provide disconnected path to the nodes.
network (MCN) routing [85]. It is worth mentioning here that One other variant of proactive centralized routing is cellular
MCN is proposed by Lin and Hsu [21], however the routing based source routing (CBSR) [87]. In this proposal, each node
for MCN is first proposed by Ananthapadmanabha et al. [85]. periodically reports neighborhood table to the BS. The report
In MCN routing, all the nodes notify the received power contains each neighbor’s load, link quality and HELLO packet
(Rx) of their neighbor nodes to the base station (BS). After receive time. The BS keeps N ∗ N matrix in which each
collecting the neighbor information, the BS draws the network node’s link to every other node is maintained and includes an
topology graph. When a node wants to send information to additional attribute of hop count in the table. The nodes send
another node, it sends route request packet (RREQ) to the BS RREQ to find the routes and the BS replies with all available
which contains the information of the destination node. BS routes via RREP, or otherwise RERR in case of having no
sends either a reply packet (RREP) containing suitable route route. RERR is also sent by a node to inform about an error
information to the requesting node or an error packet (RERR) in the route suggested by the BS.
if a route to the destination does not exist. Route information In a nutshell, the three schemes for centralized proactive
is also cached by nodes for reducing the BS overhead. routing are MCN routing, BCR and CBSR. In all these

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


schemes, nodes periodically share HELLO packets with each sends route reply (RREP) packet back to the original sender,
other. In MCN, nodes notify their neighbors’ received power else will forward the RREQ. If the BS receive the RREQ,
(Rx) to BS and then update to BS only if there is a change in it will not further forward the RREQ and replies only if
the neighbors’ Rx. In BCR, all the nodes use their minimum it has route information to the destination node. All nodes
power and communicate only to their neighbor nodes. Thus, that forwards RREQ but not received RREP, will discard the
nodes communicate to the BS also via multiple hops. The RREQ information after timeout. If more than one RREPs are
results have shown that both MCN routing and BCR can received by the source node, it keeps only one RREP and
improve the network throughput in their considered network discard rest others.
scenarios. In CBSR, each node reports their neighbors’ load, Simulation results showed that ACMR minimizes the time
link quality with the neighbor nodes and receive time of delay for data packets, because (1) of restriction on the
HELLO packet to the BS. The results have shown that CBSR number of hops in ACMR (2) the load of base station is
can improve the network performance in terms of reduced reduced by using the nodes as relays. There are three modes
delay and routing overhead and increased packet delivery ratio. of communication in ACMR, multi-hop mode, cellular mode
and hybrid mode (accessing BS through multi-hop relays).
H. Proactive hybrid routing ( The impact of node density (to show connectivity of the
network) and the movement speed (to show the frequency
Proactive hybrid routing is an adaptive routing technique
of topology changes) is evaluated for the multi-hop network.
that adapts between proactive centralized routing and proactive
The cellular network packet delivery is affected by traffic load
distributed routing, based on the network constraints.
and channel bandwidth rather than the node density. Similarly
In this type of routing, a routing scheme called as light
cellular network has no impact of movement speed since all
dark routing (LDR) [88] is proposed that adapts to a changing
the movements are within the coverage of base station to
network topology. It works in both centralized and distributed
guarantee the reliability of connection. Therefore there is no
modes. LDR uses the same format and operations (for example
extra communication overhead in cellular network with the
RREQ packet format) as dynamic source routing (DSR) pro-
increase in the node density and movement speed.
tocol. So in the absence of the BS the process continues as a
The communication overhead increases both in ACMR and
DSR distributed routing protocol. In LDR, each node reports
OLSR with the increase in node density, movement speed
changes in the link state and traffic load to the BS which
and p-hop values. In all cases the communication overhead
computes all routes in the network using latency metric and
of ACMR is less than OLSR because of adopting zone based
informs sub-routes information to the respective relay nodes.
routing which limits the broadcast area. When p-hops are less
Nodes under the coverage of the infrastructure network report
than equal to 2, the routing table is maintained by just using
any changes directly to the BS in one hop. Nodes outside of
HELLO packet while the communication overhead increases
the coverage report any considerable change to the BS via
with the increase in p-hop values. In comparison to cellu-
multi-hop using the help of some proxy node already under
lar network, the extra communication overhead to maintain
the coverage. Thus, a proxy node communicates with the out-
proactively the topology of the network still not guarantee the
of-the-coverage node in distributed fashion and with the BS
up-to-date state of the network due to high network dynamics
in centralized fashion.
which lead to reduction in packet delivery. On the other hand,
the spatial frequency reuse results in increasing the packet
I. Hybrid distributed routing (4.1.3) delivery rate. Thus the overall packet delivery rate is higher
In hybrid routing, the nodes which are within the specific than the cellular network but fluctuating.
hop range always communicate proactively while farther nodes
always communicate re-actively, this type of routing is also
termed as zone based routing. J. Adaptive centralized routing (4.1.4)
An adaptive cellular and multi-hop routing (ACMR) which In adaptive routing, the routing algorithm switches between
uses zone routing technique and maintains the zone routing reactive and proactive routing depending upon varying net-
information based on optimized link state routing (OLSR) is work dynamics. The adaptive routing is different from hybrid
proposed in [14]. In ACMR, each node proactively maintains routing. In hybrid routing, both the reactive and proactive
the routing table for p-hop neighbors using OLSR. Higher p- routing work at the same time. The proactive routing maintains
hop values improves network connectivity and coverage, and the neighborhood table for nodes in a zone within few hops,
lower p-hop values reduce the complexity of route mainte- while the reactive routing is used for finding route outside that
nance, communication overhead, bandwidth consumption and zone. While in adaptive routing, either proactive or reactive
energy. The value of p is not constant and is variable with routing works at a time. Moreover, the routing algorithm works
dynamic network topology. The ACMR protocol is proactive for the whole network; so there are no zones in adaptive
(table driven based) maintaining the topology of the network routing. In adaptive centralized routing, a central controller
within p-hops, while reactive (on-demand) for more than p- usually BS is responsible for all kinds of such switching and
hops. If the information about the destination is not available in routing decisions.
the routing table, the source node sends route request (RREQ) Shaikh and Wismüller proposed a route discovery mecha-
packet. The RREQ packet is forwarded uptil 2 ∗ p hops. nism named as centralized adaptive routing (CAR). For a given
If any node has information about the destination node, it cell, the BS is responsible for route decision. The centralized

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


routing algorithm switches between reactive and proactive (ii) Interference metric: Interference management is done
routing depending upon network dynamics. For example, if using the power control mechanism. By knowing the distances
the node density is high, or average node mobility or traffic between the source, destination and the nodes which get
load is low in the network, reactive routing performs better in interfered by the communication, it is ensured to maintain
terms of low routing overhead [99], [100]. minimal interference in the network.
In centralized proactive routing, each node periodically (iii) Dynamic call dropping and connectivity metric: Dy-
exchanges HELLO packet with each other, and subsequently, namic call dropping due to mobility of nodes or emergency
the updated neighbor list is sent to the BS. When a node call situation is incorporated, which selects a neighbor node
requires a route to another node, it sends RREQ packet to around the particular depleted node as relay node. The avail-
the BS. The BS will either provide a route if possible, or ability of an alternate relay node is ensured by having sufficient
will establish a traditional cellular connection. In centralized number of neighbor nodes around the depleted intermediate
reactive routing, generally BS does not have updated topology relay node. Thus intermediate relay node was selected if it
graph, so upon receiving an RREQ packet from a node, BS has k number of neighbor nodes where k>θ(logn), where
broadcasts the same RREQ packet in the whole network. All n is total number of neighbor nodes in the network. If
the nodes then exchange HELLO packets and then send the k<0.074(logn), it means network is asymptotically discon-
updated neighbor list. BS draws the network topology graph nected and k>5.1774(logn) means connectivity probability is
based on the neighbor list tables and finds the optimal route. approaching one.
The route is then communicated to all the nodes participating (iv) End-to-end throughput: End-to-end throughput is en-
in that route. In order to further reduce the routing overhead, sured by maintaining minimum threshold SINR at each inter-
each node decides about whether to participate in the route mediate relay. Interference at the node due to other communi-
discovery or not, based on their remaining battery energy and cations assumes Rayleigh fading, because of prevalent fading
current traffic load [89]. factor in cellular networks instead of impractical commonly
used distance-decay law.
K. Location-based routing (4.3) (v) End-to-end delay: The transmission delay at each hop
Location based routing protocols depends on location and the propagation delay constitutes the end-to-end delay
servers (or some other central entity) to know the location which is minimized by selecting route having minimum num-
of the destination node. The route decision is taken by nodes ber of hops.
autonomously using the location information (i.e. distributed The routing algorithm filters nodes sequentially from n
routing strategy), except in [90], where the decision depends nodes. Firstly set of cooperative nodes is constructed from n
upon the BS (i.e. centralized routing strategy). It is important nodes. The nodes satisfying interference constraint are further
to note that all state of the art location based routing proto- shortlisted from the set of cooperative nodes. The resulting set
cols are reactive in nature, apart from the fact that in real, of nodes are further shortlisted in terms of satisfying sufficient
every location based routing need to proactively maintain the neighborhood connectivity, minimum required SINR and then
location update. Due to this regular location update, location the delay.
based routing protocols do not need entire network broadcast. Interference aware routing (IAR) which constructs route
It is also interesting to note that all location based routing from source to destination, using nodes on the cell edges is
protocols inherently assume that a given channel is used by proposed in [8], [91]. Firstly, the route sends data towards
only one D2D pair. This would probably be managed by nodes in the edge, then the data travels along the cell edge
location servers; however, it is not explicitly mentioned in towards the direction of destination; and then finally data is
any state-of-the-art work. A number of location based routing forwarded from cell edge to the destination. Each stage of the
protocols are proposed, which are discussed below. IAR utilizes Shortest Path Routing (SPR). Therefore routes in
The work in [90] proposed a cross layer routing (CLR) IAR are longer than direct shortest path, but this IAR path has
algorithm with multiple constraints. The algorithm incorporate overall lesser interference. The algorithm uses the downlink
MAC layer’s interference levels and physical layer’s propa- band for D2D communications and the outage probability of
gation channel conditions into the routing layer’s decisions. cellular communication due to D2D communication is also
The constraints are categorized as node constraints and path calculated. Some limitations in IAR scheme are as follows.
constraints. The constraints imposed on the selection of in- The work assumes that it knows the location of the cellular
termediate relay node are the sufficient neighborhood connec- UEs whose spectrum are being reused by the D2D nodes and
tivity, dynamic call dropping, interference to other nodes and thus avoids interference among the cellular UEs and D2D UEs.
cooperation. The path constraints are the throughput and delay. Moreover, it can be easily seen that when the distance between
All these are detailed as follows. the source and destination nodes is small, the shortest path
(i) Cooperation metric: Full cooperation is not assumed and routing is better in terms of both the interference and delay.
the cooperation is incentivized by a charging and rewarding The work in [92] shortlists the feasible D2D links which
policy. When a call initiation request is received by BS, it uses are reliable (by maintaining minimum required SINR) and
broadcast control channel (CCH) to forward the cooperation avoids harmful interference to the cellular users. The routing
request which contains the incentive amount, source and algorithm considered two different scenarios, fixed rate scheme
destination address to the nodes. Only interested nodes reply (FRS) and fixed power scheme (FPS). In FPS, each link
via CCH. achieves a different rate based on the SINR at the receiver.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


Thus the maximum throughput path is not necessarily the and previous load (as opposed to delay in packet reception
shortest path. In FRS, a threshold SINR is maintained at each as in CBR). The energy efficiency is maximized based on
link, and the end-to-end throughput is calculated by dividing fair distribution of energy among all nodes (instead of power
the rate achieved at an individual link with the number of control mechanisms). LoMM by exploiting location infor-
hops. Thus with the given SINR, the optimal path is usually mation avoids intermediate nodes that are farther from the
the shortest path in terms of number of hops in the fixed rate destination (BS) which reduces the candidate relaying nodes.
scheme. The FRS-Improved (FRS-I) is proposed in [101]. In This reduces the signaling overhead and energy consumption
FRS-I, the link outage is calculated by taking into account the while increasing the application throughput. However results
channel fading effect between D2D pairs; and between D2D showed that it is 95% less probable (comparing to ad hoc MM
pair and cellular network. The actual (calculated) interference routing) to find a next hop node in the direction of destination.
could be higher when taking channel fading into consideration. Hybrid LoMM (HyLoMM) is proposed in [96] which is
Thus, the overall throughput increases after accounting the link combination of LoMM and adhoc MM routing. HyLoMM first
outages. The fixed rate schemes are preferred because same uses LoMM routing and if the route is not found then use the
modulation and coding scheme can be used on all links. ad hoc MM routing. However HyLoMM has less application
The results show that the throughput is multiple times throughput.
higher than the one achieved with one hop communications. The x-hop location permissive LoMM (xLoMM) allows x-
However, it is assumed that only one D2D link could be hops that are not in the direction towards the destination.
active under a particular frequency channel in a given area. Optimal value of x-permissive hops that reduces network
The algorithm uses downlink band for D2D communications flooding and energy cost while maintaining end-to-end per-
and interference to cellular users is avoided by assuming that formance is investigated in [97]. It is found via simulations
each D2D node has complete knowledge about the location under a certain given scenario that with the increase in cell
and channel information of cellular nodes. radius, higher x-hop values are required. However values of
In Maximum-Rate towards Destination (MR-D) routing x-hop above four does not improve performance but only
algorithm [93], D2D UEs use uplink band for communication. increase the signaling load. The xLoMM has also been tested
The interesting work in this paper is to avoid interference in three different location areas: line of sight (LOS), street
without calculating the total interference impact at any node. (Non LOS) and intersection . Interesting results are found in
This is because due to underlay approach, nodes farther from the intersection, where LOS propagation is used with four road
primary cellular nodes and Base Station provides higher data segments, among which two road segments progress towards
Rate (SIR ratio) than those which are nearer. Thus just by tak- the destination. These two road segments that progress towards
ing into account the Rate requirement, it avoids interference. the destination node, increase the route diversity and the end-
In downlink band, the route formed by the MR-D algorithm to-end performance while also increasing signaling load and
goes towards the cell edge which increase hop count. MR-D energy consumption.
advanced (MR-DA) is proposed [94] to reduce the number of In the following text, a summary and/or comparison be-
hops by directing the traffic towards the next hop which is tween the various strategies which have been proposed under
in the direction of destination. Two set of nodes Φ1 and Φ2 location-based routing is provided, in order to have clear
are defined. Φ1 is the set of nodes towards the direction of understandings of the importance of each strategy. In all
destination and Φ2 towards the opposite. If there is no next location-based routing, nodes depend on location servers to
hop neighbor node possible in the direction of destination i.e. find the location of the destination node, thus proactively
if Φ1 = null then Φ2 is used. maintains the location update. In CLR, the BS decides the
A heuristic routing algorithm and a channel selection strat- route based on multiple node and path constraints. The results
egy is proposed [95]. The algorithm uses the downlink band have shown that CLR can improve network performance in
and assumes that every node knows the channel and location terms of various performance metrics. In all the location based
information of every other cellular node, D2D node and BS. routing (except CLR), each node takes the route decision
The algorithm selects the optimal channel that provides the autonomously. The IAR constructs route using nodes near the
best channel gain and the minimum interference. With the cell edges, thus have longer routes than direct shortest path, but
given channel, the next hop node is selected based on the have lower D2D outage probability when distance between the
combination of two weighted factors, the first is the energy end-nodes is greater than cell radius. When distance between
efficiency of the current hop (up to the next hop) and the nodes is lower than cell radius, SPR (shortest path routing)
second is the direct transmission rate from next hop to final is preferred. The FRS and FPS schemes can improve the
destination. network throughput under fixed power or fixed rate (in terms
The work in [96] proposed location based multiple metric of maintaining threshold SINR at each link) scenario.
(LoMM) routing protocol that guarantees packet delivery by The MR-D scheme avoids the interference by taking into
discovering complete route from source to destination. The account the rate requirements, and thus enhanced the outage
multiple metric weighted cost function includes the number probability. The MR-DA scheme is an advanced version of
of hops, traffic load and energy consumption. Congestion is MR-D, which reduces the number hops by directing the traffic
calculated by using delay in the reception of periodic beacon towards the next hop node which is in the direction of destina-
messages, and the traffic load for bursty traffic is finally tion. In joint routing and channel selection (JRC) algorithm,
calculated by the cumulative weight of current congestion after selecting a channel having minimum interference and

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


best channel gain, the next hop is selected based on the energy M. Summary and Insights
efficiency and data transmission rate. In LoMM, location As we know that the well-known DSR (Dynamic Source
awareness is used by nodes to avoid intermediate nodes that Routing) and AODV (Ad-hoc On-demand Distance Vector
are farther away from the destination node. Thus, LoMM routing), both are distributed reactive routing protocols, how-
reduces the number of RREQ packets (signaling overhead) ever their route discovery process are different. Similarly,
which improves the application throughput. In hybrid LoMM under some categories discussed above, more than one routing
(HyLoMM), if an LoMM routing fails to find a route, MM protocols are found in literature which differs based on their (i)
routing is used, which improves the application packet delivery route discovery process, and/or (ii) routing metric, and/or (iii)
ratio (PDR). Another technique to improve the PDR is x- objective or network performance criteria. Therefore, a sum-
LoMM. In xLoMM, x-hops are allowed that are not in the mary and/or comparison between the various strategies which
direction of destination node. Although this increases the have been proposed under the same category is provided (at
signaling overhead (in comparison to LoMM) but it reaches the end of each category), in order to have clear understandings
almost 95% of the PDR of MM. of the importance of each strategy.
There are some schemes proposed for communication
among users in vehicles. As an example, consider a cluster-
based routing scheme proposed for multi-hop D2D commu-
L. Content-based routing (5)
nication in vehicular network in [116]. With BS assistance,
the clustering based on communication range and mobility
Content-based routing is normally used to establish routes
(CCRM) is used to simplify the network topology, by dividing
for sharing the frequently used data among users, like viral
the large scale network into clusters of different connecting
videos etc. Thus the route does not depend upon the desti-
probabilities, movement direction and speed. A node that can
nation node, but is established based on the contents of the
maintain for long duration; the highest number of connection
information being shared.
with other nodes is selected as cluster head. The routing is
A group-based approach for content-based routing is pro- then used to enable multi-hop communication within clusters,
posed in [102]. In this approach, peer-to-peer (P2P) nodes in between inter-connected clusters and among the isolated clus-
the network forms multiple groups. In each group, one peer is ters. It has been noticed that the network topology for vehicular
selected as group owner (GO) after negotiation with other peer routing is quite different from normal D2D network topology,
nodes in the group. One peer is selected as a relay client, that therefore we neither cite nor discuss any other routing protocol
acts a bridge between two (or more) groups using a different for vehicle-to-vehicle (V2V) communication.
MAC identity at each group. Other nodes are normal P2P It is very interesting to note that all the muti-hop D2D
clients. After the formation of groups, nodes can share their routing schemes use only D2D relays for communications as
contents with other nodes in the network. It is assumed that shown in Table V. For further readings, it would be interesting
each content is provided by exactly one node, and once that to go through the diverse types of spectrum used by the multi-
node is disconnected, the content also becomes unavailable. hop D2D routing schemes as shown in Table VI.
The routing table that contains information about route
to each content in the network is called as content routing V. T HE STATE - OF - THE -A RT IN M ULTI - HOP D2I/I2D
table (CRT). There are two steps to build the CRT, content ROUTING
registration and content advertisement. In content registration, In multi-hop D2D routing, the route is established between
a client advertises a new content by sending message to the the two user end-nodes, while in multi-hop D2I/I2D routing,
GO. The GO registers the content and replies acknowledgment multi-hop route is established between the node and the
(ACK) to the client. In content advertisement, GO advertises network service entity, i.e. BS. The BS serves as a common
the content internally by sending a broadcast message to all source in the downlink and common destination in the uplink
the nodes in the group so that all the nodes update their communications. The BS periodically broadcasts the pilot
CRT. The relay client replies with an ACK message to the signals which is received by all the nodes in the network. The
GO. In order to advertise the content to other groups, the received signal strength (RSS) of the pilot signal is used in
relay client forwards the advertisement message to the GO most of the schemes as link cost with the BS, which is used for
of other group. The GO of the other group also replied with routing decision. Thus due to the assistance from BS in form
ACK for confirming the reception of the advertised message. of pilot signals for link cost calculation, this type of routing
It then broadcast the message within its group and the process is more information-rich which is major difference from the
continues till all the groups and their nodes update their CRT. multi-hop D2D routing [103].
Thus, based on the ACKs, all the GO and relay clients have In multi-hop D2I/I2D routing, a node can not only commu-
confirmed information about the available contents. So when a nicate with its own BS, but can also communicate with other
node needs some contents, it first checks its CRT and in case of BSs using relay nodes. This results in further load balancing,
no entry, send content request to its GO. Results showed that as nodes are no longer restricted to single cell only [2]. This
the application layer throughput increases with the increase type of routing has been further extended by relay nodes to
in offered load. However with the increase in the number of forward the data of other nodes beyond the service area [117].
hops, the channel contention also increases resulting in overall The state-of-the-art multi-hop D2I/I2D routing schemes
decreased throughput. have been broadly classified (based on their objectives) into

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


TABLE V: Type of relays used in state-of-the-art topology based routing algorithms

Topology Relay type
Network Algorithm Network Relay
Flat Location based Tree based D2D Relay
Mobile Fixed
PIR [83] 3 3
ICR [84] 3 3
MCN [85] 3 3
BCR [86] 3 3
CBSR [87] 3 3
LDR [88] 3 3
ACMR [14] 3 3
CAR [89] 3 3
CRL [90] 3 3
IAR [8], [91] 3 3
FRS [92] 3 3
FPS [92] 3 3
MR-D [93] 3 3
MR-DA [94] 3 3
JRC [95] 3 3
LoMM [96] 3 3
HyLoMM [96] 3 3
xLoMM [97] 3 3
PAC [103] 3 3
IAC [103] 3 3
PSF [104], [105] 3 3
RCIL [106] 3 3
ERP [107] 3 3
ContouR [1] 3 3
RRR [108] 3 3
MRC [2] 3 3
LSSON [109] 3 3
JRS [23] 3 3
RAR [110] 3 3 3
MCPR [111] 3 3 3
MMRP [112] 3 3
Ad-hoc LBS-AOMDV [7] 3 3
CRSP [113] 3 3
ARCE [68] 3 3
DST [114], [115] 3 3

source selection based, content-based and topology based (re- neighbors) and global (pilot signal from BS) information. The
active, proactive and tree-based) routing schemes. The source tree-based routing is used when wireless network scalability
selection-based routing scheme is used when it is required to is of major concern. All the schemes under each category use
select a source node, from which the cellular resources can be BS-assistance for their route decision.
snatched in order to allocate them to the new user request. The The complete taxonomy of state-of-the-art multi-hop
source node can then communicate via multi-hop to another D2I/I2D routing schemes is shown in Fig. 3 and explained
BS as suggested by the current BS. This is to balance the load in the subsequent subsections. In Fig. 3, each category of
among multiple BSs and is done only if a multi-hop route is D2I/I2D routing has been numbered in order to have a quick
possible. Content-based routing is used to establish routes for referral from the following text explanations. The category
sharing the frequently used data among users. numbers assignment does not uniquely distinguish between
Depending upon the network structure, topology based D2I/I2D and D2D, and therefore the readers would notice gaps
routing has been categorized into hierarchical routing and in the category numbers for only D2I/I2D routing.
flat topology routing. The hierarchical (or tree-based) routing
scheme can be used only in the presence of a central control
entity. The flat topology based routing schemes have been A. Source Selection-based Routing (3)
further classified into two sub-categories: reactive routing In source selection based routing, fixed relay nodes are used
and proactive routing. In proactive BS-assisted routing, nodes by the network operators to divert the traffic from the high
periodically update routing table, usually for delay sensitive or loaded BS to a low loaded BS. When there is a high traffic
real-time applications. The objective is to instantly provide a load and BSs have no additional cellular resources to fulfill
multi-hop route with better BS channel capacity from node to new resource requests, then in case of a new resource request
the BS, when BS is overloaded or when the channel quality (or from a node, BS updates its table listing the relay stations and
capacity) with BS is not good. In reactive BS-assisted routing, their neighborhood mobile nodes with their bandwidth status.
nodes find the route on-demand, by using their local (from The source selection procedure (SSP) is then applied to find

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


TABLE VI: Spectrum type used in state-of-the-art topology is referred as the pseudo-source node or simply the source
based routing algorithms node and it could also be the same original source node which
Type of Spectrum request the new cellular resources. Conclusively BS can use
Inband any proposed SSP algorithm depending on the number and
Routing Algorithm
Outband Overlay Underlay
Downlink Uplink bandwidth of TDSs to efficiently discover the relaying routes.
PIR [83] 3 The SSPs are proposed for the voice calling traffic which has
ICR [84] 3 fixed bandwidth requirements, however the algorithm can also
MCN [85] 3
be used for data traffic by considering the required bandwidth
BCR [86] 3
CBSR [87] 3 for the data transmission with the available free bandwidth.
LDR [88] 3
ACMR [14] 3
CAR [89] 3 B. Topology-based Routing (4)
CRL [90] 3 3 Most of the routing protocols have been proposed based on
IAR [8], [91] 3
FRS [92] 3 the particular topology of the network. In general, there are
FPS [92] 3 three major categories of topology based routing protocols,
MR-D [93] 3 which are hierarchical routing, location based routing and
MR-DA [94] 3
JRC [95] 3
flat topology based routing protocols. However, for multi-hop
LoMM [96] 3 3 D2I/I2D communications, state-of-the-art schemes have been
HyLoMM [96] 3 3 categorized into hierarchical routing and flat topology based
xLoMM [97] 3 3 routing.
PAC & IAC [103] 3 3
PSF [104], [105] 3 The categorization of all the state-of-the-art topology-based
RCIL [106] 3 multi-hop D2I/I2D routing schemes with their performance
ERP [107] 3 parameters and simulation tool used are listed in Table VII,
ContouR [1] 3 3
and discussed below.
RRR [108] 3
MRC [2] 3
LSSON [109] 3
JRS [23] 3
C. Flat topology routing (4.1)
RAR [110] 3 When nodes in the network do not have any specific
MCPR [111] 3 structure like tree or cluster, nor have any location awareness
MMRP [112] 3
LBS-AOMDV [7] 3 mechanism then such arbitrary distribution of nodes is referred
CRSP [113] 3 as flat topology network structure, and the routing is called flat
ARCE [68] 3 3 topology routing. As like normal wireless network, routing
DST [114], [115] 3
protocols in multi-hop cellular network are categorized as
reactive routing and proactive routing. All the reactive and
proactive routing schemes for multihop D2I/I2D communica-
a node which can release the available cellular resources and tion are BS-assisted in nature.
switch to ad hoc mode using the relay stations (or called the
traffic diversion stations or TDSs).
Three algorithms have been proposed for source selection D. Reactive BS-assisted routing (
procedures (SSPs) for the routing protocols in ICNs [118]. The When the route is updated only when required, such routing
first SSP (SSP1) selects the node which has highest number is called reactive routing. In reactive BS-assisted routing, when
of neighbor TDSs. However in such case, a node with highest required, a node finds the route towards the destination while
number of neighbor TDSs is always selected which results also using the global information provided by the BS. A
in congestion. The second SSP (SSP2) selects node with number of schemes are proposed under this category, which
the highest available average bandwidth of the neighborhood are discussed in detail below.
TDSs. Thus SSP2 can avoid blocking some specific nodes. A fast and reliable route discovery solution considering
However the request rejection rate (RRR) of SSP2 would be mobility of nodes is proposed in [103]. The work finds the
higher than SSP1 since some specific TDS could be favored next hop node among the neighbor nodes while minimizing
due to its positioning among the available TDSs of the selected route setup delay and flooding of control messages required for
node. the route discovery. The pilot signal is periodically broadcast
The third SSP (SSP3) combines SSP1 and SSP2 and thus by the BS, and the received power of this signal (P LBS ) is
a lower RRR could be achieved than SSP1 and SSP2. To used as a link cost by all the nodes in the network. Therefore
simplify the working of SSP3, the BS sorts the table according the neighbor node having the minimum total path loss (from
to the number of TDSs and the available free bandwidth. Thus mobile node to neighbor node (P L1 ) and from neighbor node
node with highest number of TDSs and with highest available to BS (P L2 )) is selected as the next hop node.
free bandwidth is simply selected. Location aware SSP (SSP- The three schemes, inhibit access control (IAC), priority ac-
LA) further extends SSP3 and keeps track of the mobility of cess control (PAC) and hybrid of IAC and PAC are proposed in
the mobile nodes and select the node which has the highest [103]. In IAC, the node sends RREQ with its measured P LBS .
possibility of moving out of the home cell. The selected node Only those neighbor nodes having their measured power path

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


loss from BS (P L2 ) less than P LBS reply with RREP after calculates the distance (d) with the BS using Friis free space
random backoff. This reduction in number of RREPs reduces formula by using the average received power and the transmit
collision of RREP messages and the route discovery time. power of BS.
This reduction is quite significant when the user density is For establishing multi-hop route, the signaling overhead by
high. The routing overhead in terms of number of RREPs the mobile nodes induces interference among the nodes. The
is also reduced. In case of collision, RREPs are retransmitted effect is negligible in mobile adhoc networks (MANETs) but
according to classical exponential backoff algorithm. The work is nontrivial in the multi-hop cellular D2D networks due to
argues that the conventional schemes have more number of high transmission power of the cellular nodes. This consumes
RREPs and thus more collisions with the increase in number network resources and drains the battery of the standby nodes.
of users which results in longer route discovery time. Thus design of a routing algorithm with low signal overhead
In PAC, the priority is decided only on the base of the is crucial. Therefore, value for the maximum number of
link cost between mobile node and neighbor node. There- hops (Nopt ) is specified to limit the RREQ flooding. The
fore the lesser the P L1 , the faster is the RREP sent by value of Nopt depends on the maximum allowed interference,
the neighbor node. When enough high priority RREPs are permissible end-to-end delay requirement or in other words
received, source node prevents further RREPs by sending an the maximum route discovery time. Each hop distance (dopt )
early stop (ESTOP) message. The time of sending ESTOP based on the Nopt relay UEs would be dopt = d/(Nopt + 1).
message depends on the relative density of users and is The connectivity probability at each hop (Ps ) must be greater
assumed one fourth the average number of neighbor nodes than threshold probability (Ps , min) and is set by the network
for the simulations. Thus PAC reduces the route discovery operator and it depends on Nopt , the transmission radius (R)
time by sending an early ESTOP. This reduction is quite of RREQ and the UE density (m/A), where m is the number
significant when the user density is high, because more number of UEs in the cell under investigation and A is the cell area.
of neighbor nodes would have better link quality. However, in This hop limiting routing is known as power efficient routing
PAC the number of RREP collisions are higher due to absence (PER) in literature.
of inhibit access control. Since the link cost P LBS is not The signaling overhead is further reduced by the proposed
required in PAC, therefore PAC can be applicable to simple power-strength-based selective forwarding (PSF), by selecting
ad hoc networks. only those candidate UEs located between the distance r (dopt
In hybrid scheme, the two schemes work in conjunction and - ∆d) and R (dopt + ∆d). This is the overlapping shaded
the priority is calculated by the multiplication of two path loss area bounded by the two transmission circles of the UE and
costs P LBS (i.e. P L2 < P LBS ) and P L1 (lower path loss BS. The value of ∆d is calculated by BS from given values
has higher priority). In all schemes, neighbor nodes reply with (P s, Nopt , A, m) and the distance d reported by the sending
P L1 , and the node having minimum total path loss P L1 and UE, which determine the transmission power of each RREQ
P L2 is selected as next hop. (PRREQ ). The BS informs the value of PRREQ to all the
All the proposed schemes perform cross layer design taking UEs. The value of ∆d is highly dependent on UE-density and
into account physical layer, MAC layer and network layers. it affects the connectivity probability Ps . The higher values
In the physical layer, the received signal strength is used to for ∆d may increase the number of forwarding participants
compute the path loss costs. The processes of inhibit access (Backer UEs) as well as the radio interference, while smaller
control (IAC) and priority access control (PAC) to reduce the values result in lower connectivity probability which could
number of collisions and make the discovery process faster results in unsuccessful route discovery. An analytical model is
is performed at the MAC layer. The network layer selects proposed to find the optimal value of ∆d. Finally, only those
the appropriate next hop and the appropriate time to send UEs which are within the specified distances will forward the
ESTOP based on the information received from lower layers. RREQ packet.
Conclusively, IAC scheme finds the most reliable path and However exact location estimation by the UEs is not easy,
hybrid scheme finds the shortest discovery time. However, all hence two approaches are proposed to implement the concept.
the proposed schemes are limited to two hop paths only. In In the first approach, the UE will forward the RREQ if the
case of more than 2 hops, end-end route is not guaranteed. received power from the previous node is greater than Pr and
Furthermore in PAC, neighbor nodes nearer to the mobile less than PR . But in this approach, a high number of RREQs
nodes have lower path loss, and selecting such node results are possible, since it could not specify any overlapping area
in higher number of total hops. with the UE after next. In second approach, the UE calculates
A power-strength-based selective forwarding (PSF) mecha- its position based on the distance from the sending UE and the
nism to efficiently flood route requests (RREQs) during the BS. The distances are calculated from their respective received
route discovery process, is proposed in [104], [105]. Each powers. However this approach requires the sending UE to
node uses received power strength of BS, to decide to be act also broadcast its beacon signal for distance calculation by the
as relay or not. This helps in reducing the signal overhead intermediate UEs. The results show that the signaling overhead
and interference during the route discovery process. The in terms of number of RREQs is considerably reduced while
power threshold to ensure minimum connectivity probability is ensuring minimum connectivity probability. However mobility
solved analytically. A detailed description of the complete PSF of nodes is not considered in this work. The proposed solution
mechanism is as follows. The BS periodically sends cell access involves high number of computations which further limits its
parameters on the broadcast channel (BCH). The sending UE scalability.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


Contour routing protocol uses received signal strength indi- probability (Ps ), cell area (A) and number of UEs in the cell
cator (RSSI) as a measure of location prediction to communi- under investigation (m). Both hybrid PAC/IAC and PSF are
cate with the BS. Due to fluctuations in the RSSI, the location proposed to reduce the signaling (or routing) overhead, which
of a static destination node (BS) could be predicted within a also reduce the average route discovery time.
donut-shaped region [1]. The algorithm neither uses GPS nor In Contour routing, a single copy of message is relayed hop
cellular triangulation for locating the node. GPS consumes a by hop until it reaches the vicinity of BS wherein multiple
lot of power and is inaccurate in indoor environment. Cellular copies of message is forwarded using limited flooding. It is
triangulation requires complicated synchronization among the assumed that due to the fluctuations in RSSI, the exact location
BSs and is usually less accurate. The source node knows its of the BS could not be identified and could be predicted within
own RSSI and the RSSI of destination node and thus the a donut-shaped (contour) region. By controlling the contour
data packet proceeds towards the destination node in multi-hop region, the time instant to change single copy to multiple
using WiFi, by selecting next hop node having higher RSSI copies could be adjusted. Thus contour routing reduces the
than the current node. Thus, the routing algorithm relieves burden over the BS by using multi-hop communication for
the BS from the high traffic load. However the multi-hop delay tolerant traffic. In RCIL, the routing metric selects the
communication is not suitable for delay sensitive data and for route having the best link gain and having the minimum
such application cellular communication is the only choice. A interference with the objective to get the best throughput under
single copy of message is relayed hop by hop until it reaches varying number of users in the cell.
the vicinity of destination node wherein multiple copies of a
message is forwarded using limited flooding. This flooding is E. Proactive BS-assisted routing (
because of the fact that we cannot pinpoint the location of When the route decision is made by each node individually
destination node using the RSSI, however a circular region however the central entity helps in providing some global
with possible location could be identified. By controlling the information, the routing is said to be the BS-assisted routing.
contour region, the time instant to change single copy to In proactive BS-assisted routing, nodes periodically update
multiple copy relaying could be adjusted. Results showed routes based on the their local observations as well as the
similar performance to flooding with reduced traffic overhead global information provided by the centralized entity.
with the proposed routing algorithm. An Efficient routing protocol (ERP) for reliable communica-
The work in [106] proposed routing metrics and route tion with the BS is proposed in [107]. In ERP, if the channel
discovery mechanism for communications with the BS using quality (measured in terms of channel capacity) of a node
uplink band. Four routing metrics are compared to determine with the BS is not good or the BS is overloaded, then the
the one with the best throughput under varying number node chooses a relay node with a better BS channel capacity
of users in a cell. Routing with shortest distance (RSD), to forward data to the BS. A node first selects the neighbor
routing with minimum interference (RMI), routing with best nodes having sufficient buffer capacity, it then selects node
link gain (RBL) and routing with combined interference and having the highest channel capacity. For every data packet,
link gain (RCIL). The cell is divided into three concentric the status of neighbor nodes is checked and the best next-hop
disks numbered from 0 to 2. Nodes within the disk 0 could node is selected, according to the adaptive routing algorithm.
communicate directly with BS. The nodes in disk 1 and 2 In the route/neighbor discovery process, BSs send beacon
could communicate with the BS with the help of nodes in message periodically, which are used by nodes to determine
inner disks. In the route discovery mechanism, only those the channel capacity with the BSs. The nodes also periodically
nodes having higher link gain reply to the route request. It broadcast HELLO message to neighbors to inform about their
is found via simulation that the routing metric RCIL performs presence and the information about the BS channel capacity
the best among all the four metrics. This is because, the RCIL is also piggybacked in it. Due to piggybacking, no extra
is the combination of two other routing metrics RMI and routing overhead incurred to inform about channel status in
RBL, wherein node having the best link and have minimum this discovery process.
interference over the other nodes is preferred. Results showed that ERP can achieve higher throughput
To sum up, the four main algorithms proposed under this than AODV. This is because in AODV, all nodes use shorter
category are Hybrid PAC/IAC, PSF, Contour and RCIL al- hops paths which results in congestion and thus have more
gorithms. In hybrid PAC/IAC, a node selects the next hop packet drops. While longer paths have more probability of
node having the minimum total path loss (i) from current link breakage due to nodes mobility, thus higher packet drops
node to next hop node and (ii) from next hop node to BS. and overall less network throughput. While in ERP, always
In PSF, two-hop communication is assumed, and only those the next higher capacity relay node is chosen or otherwise
nodes participate in route discovery which are in between the is sent directly to BS, therefore have overall good network
UE and BS. PSF further shortlists the candidate next hop throughput. With respect to routing overhead, AODV routing
nodes by selecting only those UEs which locate between the overhead is significant with the increase in number of nodes
distance (dopt - ∆d) and (dopt + ∆d). This is the overlapping which result in high congestion, while ERP does not have any
shaded area bounded by the two transmission circles of UE significant overhead because of its simple neighbor discovery
and BS. The values of the hop distance (dopt ) and ∆d are process. However ERP has been limited to two hops only in
calculated by BS from the given values of distance with the simulation.
the BS (d), maximum number of hops (Nopt ), connectivity

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


TABLE VII: Comparison of topology based multi-hop D2I/I2D routing algorithms

Topology Proactive / Distributed / D2D Routing Performance parameters Simulation Tool
structure Reactive Centralized / Algorithms
PAC & IAC Average route discovery time, Average total -
[103] received power
Reactive BS-assist
Flat PSF [104], [105] Successful connectivity probability, Total num- NS-2
ber of RREQs, Total transmission power re-
quired for RREQs
ContouR [1] Average delay time, Traffic overhead, Delivery ONE
RCIL [106] Throughput -
Proactive BS-assist ERP [107] Throughput, Routing overhead rate NS-2
RRR [108] Throughput Numerical results
MRC [2] Number of control packets, Average packet -
Tree based Proactive Centralized
error rate
LSSON [109] Average hop count -
JRS [23] Percentage of traffic assigned to each cell (for MATLAB
load balancing)

F. Hierarchical routing (4.2) reception range of the BS is comparatively larger and is also
affected by the traffic load however the effect is not such
Hierarchical routing is the mostly widely used technique in
adverse as on the mobile nodes. The routing algorithm grows
the communication networks. Nearly all the existing wired
a tree initiating from the BS. The nodes are added with the
networks use hierarchy in the form of ’network’, ’subnet’
increasing distance with the BS. The selection of relay requires
and ’sub-subnet’. Using this hierarchy, routing in the internet
that (in decreasing priority) (a) reception range distance is less
use inter-domain routing, to forward the traffic based on only
than the maximum threshold distance (b) load is balanced (c)
the ’network’ address. After reaching the destination network,
number of hops are minimized. The average throughput for
the intra-domain routing is used to forward the traffic to the
each node is calculated based on the successful probability of
destination node using the ’subnet’ and ’sub-subnet’ addresses.
transmission. The successful transmission requires that (a) a
Thus, routing tables in hierarchical routing are summarized,
node transmit its own data with probability p0, (b) the next hop
since routing table entry for the ’network’ is suffice for all the
node j is not transmitting with probability 1−pt(j) and (c) the
’subnets’ and ’sub-subnets’ in the network.
interference at the next hop node (receiver) is low enough to
However, in wireless networks, where most of the wireless
meet the required SINR threshold with the probability ps(j).
nodes have limited amount of battery, maintaining wireless
Assuming all events to be independent, the user i0 s throughput
network hierarchy is quite energy-consuming. Anyhow, due
is calculated as u(i) = p0(1−P t(j))ps(j). However the work
to the appealing benefits of hierarchical networks, the wireless
only considers a unidirectional traffic from mobile nodes to a
networks have two possible types of network hierarchy, which
single BS only, using uplink band with no fading or power
are ’clusters’ and ’trees’. Accordingly, the hierarchical routing
has also two types; the cluster based routing and the tree
based routing. It is interestingly found that all the proposed The work in [2] proposed routing scheme for multiple
hierarchical routing schemes for multi-hop cellular networks route coding. The routing scheme consists of two phases.
are the tree based routing. In proactive phase, routes between BS and relay nodes are
In a tree-based routing scheme, the BS or the mobile proactively established and updated periodically. BS sends
switching center (MSC) acts as a root of the tree and manages reference packets which is forwarded by the relay nodes until
the tree structure. In some works, BS acts as a simple the maximum number of hops is reached. The maximum hop
destination node and the hierarchy is constructed by the mobile limit is defined which reduces the number of control packets.
nodes in the network. The tree-based routing is preferred In this way, multiple trees rooted at the BSs are constructed
when network scalability is of major concern. The tree-based by the relay nodes. Multiple trees increase the possibility of
communication is worst if source and destination nodes are disjoint routes. In reactive phase, a user node establish route
close to each other. All state-of-the-art tree-based routing to the BSs on demand, when data transmission is required.
protocols are proactive, since BS needs to regularly update The RREQ packets are generated by the mobile nodes which
the tree structure. The detailed description of these existing are forwarded until the BSs. The BSs replies with routes
tree based routing schemes are as follows. which are mutually disjoint that is each route does not share
Andrew et al. [108] investigate the effects of the traffic the relay nodes. This reactive approach further reduces the
load or the intensity of nodes on the reception range of the number of control packets. Results have shown the decrease
nodes. It is found that with the increase in traffic intensity, the in number of control packets and also reduce average packet
interference also increases which results in reduced reception error rate. However some limitations in the proposed scheme
range of the nodes and thus finally increase in the number are as follows. The scheme assumes that dedicated relay nodes
of hops between the source and the destination nodes. The are used which forwards the data of only one source node at

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


a time. Moreover, the work is limited to uplink transmissions broadcast the depth of the layer to be infinite to behave like
only. a leaf node, in order to avoid any further relaying packets.
The Large Scale Self Organizing Network (LSSON) routing For an upstream data packet the source node sends the
scheme that scales to large areas keeping relatively constant data packet to any nearby RN and then the RN selects
user density is proposed in [109]. The routing scheme con- the layer according to the layer scheduling algorithm. In
structs trees which are rooted and maintained by the BSs. downstream transmission the MSC selects the serving RN
Each node maintains a parent node for each tree. Cluster is and corresponding layer (means a BS) according to the layer
finally formed by joining a cell with its 6 neighboring cells in scheduling algorithm. The intermediate RNs cannot change the
hexagonal structure. Such clusters are maintained by roots by serving RN and other header fields of the data packet. Each
periodic updates from the nodes. intermediate RN will forward the data packet only if it receives
The hierarchical proactive routing scheme is employed with from its parent and the destination is one of its descendants RN
two tiers, the intra cluster routing and the inter cluster routing. in the respective layer. The MSC collects the load information
The intra cluster or local routing is used when the message of each layer from the respected BS, and then informs about
reaches in the cluster/cell which contains the destination node. the updated load status to all the BSs. The BS forwards to the
The message is forwarded upstream to the BS via multi-hop, RNs not only its load information but also of other layers being
and then the BS finally broadcast the message directly to serviced by the MSC. This load information is the available
the destination node in one hop. The downstream message channel capacity for a layer (BS). This information is used
could be more than one hops if the destination node is out by the RNs to make layer scheduling decision. It is noted
of the BS communication range. The inter cluster routing is that if the available channel (overlay) capacity and packet
used to forward the message between the cells (clusters). The size is same for all BSs (layers) then the round robin is the
message is placed on the tree of the next hop cell towards best scheduling algorithm however such setup is not used in
the destination node. When the message reaches the cluster real practice. The simulation results showed that the proposed
which contains the destination node then the intra-cluster numerical scheduling algorithm is suitable for heterogeneous
routing is continued ahead. Farthest neighbor node is selected networks as the algorithm performs well and the performance
for message forwarding in case of more than one available variation effect is unnoticeable under drastic changing traffic
neighbor nodes towards the destination. load.
Nevertheless, this work uses strategic placement of fixed To summarize, the four algorithms proposed under this cat-
relay nodes, relieving mobile nodes from the routing overhead. egory are RRR, MRC, JRS and LSSON. The RRR algorithm
Geographic location of BSs and destination node is also uses mobile nodes as the relay nodes, while MRC, JRS and
required to be known for the routing process. Stationary nodes LSSON use operator installed fixed relay nodes for their route
are considered in the simulation. It saves cellular resources decisions. The RRR routing investigates the effects of the
only in the upstream communication, since downstream com- traffic load on the reception range of the nodes. By utilizing
munication is still carried out in classical one hop manner. the fixed relay nodes, cellular resources are saved in upstream
Furthermore, it can be seen that the communication is worst communications in LSSON, the number of control packets and
if the source and destination nodes are very close to each other. average packet error rate are reduced in MRC, and a stable
A joint routing and scheduling algorithm is proposed in algorithm that performs well, and the performance variation
[23]. The algorithm behaves differently both the for the up- effect is unnoticeable under drastic changing traffic load, is
link and the downlink transmissions. For uplink transmission, proposed in JRS.
strategically placed static relay nodes (or routing nodes or
RNs) runs the algorithm in a distributed manner. For downlink
transmission, the algorithm works in a centralized manner, G. Content-based Routing (5)
being used by the mobile switching center (MSC). The routing Content-based routing is normally used to establish routes
algorithm balances the traffic load among the bottleneck BSs for sharing the frequently used data among users, like viral
(similar to balancing the load among intermediate nodes in videos etc. The BSs act as the file servers and multi-hop route
the ad hoc network). Spanning tree rooted at the BS is used are established to share the frequently used files among users.
as routing topology based on the distance vector technique. The three different versions of content-based routing are
Spanning tree is referred as layer and the number of layers proposed in [119], D2D routing without peer-to-peer (P2P)
equal to the number of BSs. Each non-root node maintains share, D2D routing with direct P2P share (R-DPS) and D2D
information about all available layers for routing purpose. routing with coverage-based P2P share (R-CPS). In these rout-
Each layer table entry contains that layer ID, parent ID and its ing, D2D nodes which download data from the file servers are
depth. A level-d node in the layer hierarchy has a depth d and called as subscribers. In order to increase average download
is d nodes (hops) away from the root. Each RN periodically rate and average throughput, it is desired that the subscriber
broadcast control packet so that the neighbor nodes updates nodes that download similar contents, share their data during
its existence. When a new RN is added into the network, it transmissions. Such subscriber nodes are called as peer D2D
finds neighbor nodes with smallest depth for each available subscribers.
layer. The new RN sets the neighbor node as the parent node In the first routing scheme i.e. D2D routing without P2P
and sets its depth to be the one plus the depth of the parent share, shortest hop route from each server to subscriber is
node in its layer table. When the load on an RN is high, it established, regardless of sharing the same contents with

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


different subscribers. In the second routing scheme i.e. the R- All the state-of-the-art ad-hoc routing protocols for D2D
DPS scheme, the route formed by the first scheme is extended networks are distributed in nature, and can be classified as:
in order to share the contents of a subscriber with other incentive (or profit)-based, multipath coding-based, security-
subscribers. Then extended route is used only if the number based, QoS-based, tree-based, reactive and proactive routing.
of hops ’from current subscriber to another subscriber’ is less The routing scheme in which the users are motivated to
than the original number of hops ’from the other subscriber participate in relaying the data of other nodes by using
to its server’. This improves the overall network throughput some incentive (or profit), is called incentive (or profit)-
by sharing the data packet with its peer. In the third routing based routing. When quality of service (QoS) cannot be
scheme i.e. the R-CPS scheme, the neighbor nodes of a compromised, for example in live video streaming, the routing
subscriber are the candidate nodes to search for route to other scheme is called QoS-based routing. When data confidentiality
subscribers. If any such route is found, and the number of hops is of utmost concern, the routing scheme is called security-
’from the candidate node to the new subscriber’ is less than based routing. If each node decide to participate or not in
the original number of hops ’from the new subscriber to its relaying the data based on its capabilities, the routing is
server’, then this new route is used. It is important to note that called device-aware routing. For large size networks, tree-
(1) this new route do not pass through the subscriber node, as based routing is preferred. The reactive distributed routing
in R-DPS scheme, and (2) both subscribers receive data from is the most widely used routing type. In reactive distributed
the same server, which improve the average download rate. routing, routes are discovered only when required, which saves
The processes defined above for R-DPS and R-CPS schemes precious resources from unnecessary periodic route updates.
are repeated for every other subscriber to find the best optimum The proactive distributed routing is used for delay-sensitive
routes. In all these schemes, the cellular user spectrum that applications and updates routing table regularly in order to
keeps interference to other users under a tolerable level is instantly provide the route when required without any delay.
used. A number of ad-hoc routing protocols have been proposed
for D2D networks in literature. These routing protocols have
been classified as shown in Fig. 4 and discussed in detail
H. Summary and Insights below. The performance parameters and the simulation tools
In all state-of-the-art multi-hop D2I/I2D routing schemes, used in the respective routing schemes are listed in Table VIII.
network relays are used to establish multi-hop routes as shown
in Table V, and thus the routing is also regarded as network A. Incentive-based Routing (1)
relay routing. The fixed network relays are used only in tree-
The objectives of incentive based routing is to establish end-
based routing, while the mobile network relays are used in all
to-end communication among the devices by motivating the
reactive and proactive BS-assisted routing. As an exception,
intermediate users to relay the data using different incentives.
a tree based routing scheme named ’reception range aware
Profit is one of the incentive that motivate the nodes to act as
routing (RRR)’ also uses mobile network relays for route
relay, and the routing is called profit based routing.
decision. All the topology based (reactive, proactive and tree-
based) routing schemes work in underlay mode and utilize
uplink band for multihop D2I/I2D communication, except in B. Profit-based routing (1.1)
JRS (a tree-based routing scheme) which uses overlay mode Routing algorithm that provides some cost to the interme-
for communications. Such classification of routing algorithms diate nodes for providing their services as relay, with the aim
based on the type of relay and spectrum used, will help to find a low cost route from source to destination is known
to quickly select a routing algorithm as per the network as profit based routing.
model requirement. It is interesting to note that no scheme A profit-based routing algorithm that maintains a multi-hop
is proposed for finding an optimal route from BS to UE in route to the BS when the profit of the BS and the relays are
underlay mode. Therefore, schemes for efficient delivery of positive is proposed in [117]. The algorithm aims at gaining
data from BS to UE(s) in underlay mode should be explored. profit instead of maximizing profit, nevertheless route with
minimum number of hops are preferred because of high profit.
On the other hand, the cost demand of each intermediate node
depends on the respective power consumption, thus nodes
When the cellular network services are either partially or on the route with higher number of hops have to consume
totally unavailable in case of natural disasters (floods, earth- comparatively less power. Therefore, in some case, the overall
quakes etc.), or when the network is overloaded or damaged in cost of route having high number of hops could be smaller
case of human-made disasters (fires, accidents, terrorist attacks than the route having lower number of hops.
etc.), or when the user is out of coverage or in dead-spot area, Each node do not need to remember route to any other
then the ad-hoc routing would be the only choice for multi-hop node but BS. The cost of uplink connection with the BS
D2D communications. In case the user is inside the network increases with the increase in distance with the BS. Thus
coverage, and the routing decision does not take the advantage if a route request failed then the source regenerate RREQ
of the presence of centralized control entity, nor care for BS with higher revenue offer, as motivation. However there are
or any other on-going user transmissions, then such routing is some serious limitations in the proposed scheme. The energy
also purely ad-hoc routing. consumption of nodes during control packets forwarding is

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


Fig. 4: Taxonomy of state-of-the-art ad-hoc routing schemes for multi-hop D2D networks.

not considered. Such forwarding of control packets might the number of control packets must be kept small [2]. To
only waste the battery energy without any gain if the final increase the diversity gain; more and more number of routes
route is not established using these nodes. Moreover nodes are are discovered.
assumed to be static and having equal amount of power. Each
node waits for some time to gather enough RREQs before D. Topology-based routing (3)
forwarding further. This induces high delay, but reduce the
The state-of-the-art ad-hoc topology-based routing schemes
routing overhead. If a better RREQ is available after the wait
for D2D networks have been categorized into flat routing and
time, then such RREQ is wasted. Furthermore unlicensed band
tree-based routing. All the topology-based schemes with their
is used for communication between relays. All the revenue-
performance parameters and simulation tool used are listed in
cost calculations are performed at the BS, thus the algorithm
Table VIII and discussed next.
is BS dependent and cannot work autonomously.

E. Flat topology routing (3.1)

C. Multipath Coding-based Routing (2)
There are two broad categories of flat routing, i.e. reactive
In single path routing, when a path fails, the communication
routing and proactive. All the flat routing schemes are dis-
is interrupted until a new path is established between the
tributed in nature, as discussed below.
source node and the destination node. Therefore, to increase
reliability, multipath routing is used which establish multiple
paths (preferably disjoint paths) between source and destina- F. Reactive distributed routing (3.1.1)
tion. As the paths are disjoint, so if one or more paths fails, In reactive distributed routing, autonomous nodes find routes
but not necessarily all, the communication can still be carried on demand. Th ad-hoc on-demand distance vector (AODV)
out however at lower rate. Thus, multipath routing provides routing is one of the famous routing protocol in literature
better QoS and better resource utilization [120]. that uses reactive distributed routing. The AODV has been
Multipath routing is used as one way for cooperation modified in thousands of ways in different type of wireless net-
between the traditional cellular communication and the D2D works, in order to fulfill their respective network requirements.
communications by using multiple parallel paths between For example, in order to have multiple paths to a destination
source and destination nodes, where the main data stream node, AOMDV (ad-hoc on-demand multipath distance vector)
is split into streams of lower data rates and routed to the is proposed. The routing protocols proposed under reactive
destination through cellular and ad-hoc D2D connections. distributed routing are discussed as follows.
[121]. Load balancing-based selective - AOMDV (LBS-AOMDV)
In [122] and [123] a multipath route coding scheme is routing [7] is an extension of AOMDV. As like AOMDV, it
proposed that encode a packet and divide it into sub-packets. establishes multiple loop-free and link disjoint paths, and to
Each sub-packet traversed along its route and finally the further reduce the routing overhead it limits the flooding by
destination BS combine and decode the packet. This reduces selecting only those nodes having links with given bandwidth
the packet loss and the error rate by exploiting diversity gain. threshold. In this way the real-time traffic demands could be
To further reduce the influence of the route loss and thus also ensured.
the packet errors; disjoint routes are selected in multipath route Resource aware routing (RAR) protocol that uses the re-
coding scheme in which the forwarding nodes are not shared active distributed routing technique is proposed in [110].
between the source node and the BS (destination). However, Three performance metrics have been studied for the resource

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


TABLE VIII: Comparison of ad-hoc topology based routing algorithms for multi-hop D2D networks
Topology Proactive / Centralized / D2D Routing Performance parameters Simulation Tool
structure Reactive BS-assisted / Algorithms
LBS-AOMDV Bandwidth consumption, Hop Count, MATLAB
[7] Number of RREQs
Reactive Distributed
RAR [110] - -
MCPR [111] Average total consumed power, Average Numerical results
success rate, Average number of hops per
MMRP [112] Delivery ratio, End-to-end delay, Routing NS-2
overhead, Remaining energy capacity
CRSP [113] Transmit power, Throughput per BS Numerical evaluations
Proactive Distributed
ARCE [68] Route lifetime probability, Packet loss, NS-2
Packet overhead
Tree based Proactive Distributed DST [114], [115] Percentage of optimal total throughput -

aware routing (RAR) considering energy efficiency and relay nodes with high received power is smaller than the one having
capabilities in [110]. Two type of relay nodes have been single hop with low received power. This is made possible with
used, fixed relay nodes which are installed and operated by new timer-based MAC protocol, which modifies the Carrier
the network operators, and mobile relay nodes which are Sense Multiple Access/Collision Avoidance (CSMA/CA) for
normal mobile nodes that provide services of relaying. The rebroadcasting RREQs. This ensures that long distance nodes
first metric preferred routes having more number of fixed will not forward the data packets earlier than the multi-hop
(dedicated) relay nodes than mobile nodes to take advantage nodes with smaller distances. Thus the final route has more
of the unlimited energy of dedicated nodes. The BS waits number of hops compared to shortest path but has the high
for pre-defined duration after receiving first RREQ, for the energy efficiency because of lower total consumed power.
possibility of getting more RREQs. In case of receiving more If the destination is BS, the destination IP address is a
than one (shortest path) RREQ, BS chooses earliest arrived predefined value and thus BSs do not need to advertise their
RREQ. The second metric avoids routes having nodes with IP addresses. If the RREQ is received by a BS, then the BS
minimal residual energy. This prolongs the network lifetime selects an MCPR path and forward upward to the controller.
and circumvents network partitioning. Third metric considers The controller selects an MCPR path among paths received
the relay capabilities in terms of processing power and mem- from various BSs, and send RREP to the selected BS. If a
ory vacancy (buffer capacity). While using this metric, if BS mobile node receives RREPs from several BSs operating under
receives a better RREQ later, then it sends additional RREP. different controllers, then a mobile select the one received first
However there are no evaluations regarding the performance and send NACK to the remaining BSs.
of the studied metrics, nor any new metric to improve the per- The results have shown the superiority of proposed algo-
formance of cellular communications is proposed. Moreover rithms Timer-based MAC + MCPR over CSMA/CA + AODV
the routing algorithm does not count number of energy-limited and CSMA/CA + MCPR. Moreover the algorithm is scalable,
mobile relays compared to energy- unlimited fixed relays. This since the algorithm runs on each mobile node in a fully
might results in higher number of mobile relays in a path and distributed way.
thus high consumption of energy. Furthermore no minimum Multiple metric routing protocol (MMRP) that integrates
threshold value for energy is specified for mobile relays, which hop count, traffic load and energy consumption into the path
might results in route failures. This is because of using those cost calculations is proposed in [112]. The purpose is to ex-
nodes whose energy is reaching to zero. plore path that minimize energy consumption and the number
A cross layer design using time-based MAC and power of hops while avoiding congestion at most favorable nodes.
aware routing algorithm is proposed in [111], named as First extension of this protocol MMRP-I (MMRP-Improved)
minimum consumed power routing (MCPR). The algorithm enables discovery of more optimal paths after the discovery
considers the minimum total transmission power among all of first path. Second extension of this protocol MMRP-A
probable routes. Each RREQ packet contains information (MMRP-Adaptive) enables heterogeneous devices to set the
about not only transmission power, but also receiving power path cost according to their classifications e.g. battery and
and total consumed power along the route. All the control bandwidth. The disadvantage of a single path cost metric is
packets like RREQ and RREP are sent with maximum power the overloading or depletion of resources in the hotspot (along
and data packets are sent with adjustable minimum required the preferred paths). The proposed MMRP uses a simple linear
power. combination of cumulative weight metrics. It modifies the
To prefer the routes with minimum required power, RREQ traditional AODV protocol. Every node in the path updates
packet which is received with high power is forwarded with the path cost field in the RREQ using the proposed cumulative
a very small wait interval, while RREQ having low power weight metric while forwarding the RREQ. When destination
has to wait a large interval of time. The wait interval (Twait) node receives the first RREQ, it waits for some particular time
is set such that the cumulative wait time for at least two hop to receive other possible RREQs.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


In MMRP-I, if a node finds an RREQ message from to the destination). The information is shared between the
the source with path cost better above some threshold than neighbor nodes periodically and the then the optimal routes are
the previous one, then it rebroadcast the new RREQ. This calculated. The base stations are the terminal points of packets,
increase the performance of the route in terms of delivery and a zero cost is reported by the BSs to their neighbors.
ratio and delay (because new RREQ probably have better However there is one serious limitation in this paper. This
delay option) at the expense of higher routing overhead and paper assumes that a mobile station can participate in at most
energy consumption. In MMRP-A, the value of path cost one data flow for a given time which is unrealistic for packet
metrics could be set by the nodes themselves according to their data service.
classifications or real time system measurements/requirements.
For example, a device with fixed power supply does not Ad hoc routing for coverage extension (ARCE) protocol
consider the energy metric. The results showed improved is a proactive routing scheme proposed in [68]. The ARCE
performance and overall less system resource consumption increases the capacity by bypassing the blocked BSs by simply
when appropriate weights metrics are used for the calculation not specifying the destination BS address in RREQ packet.
of the path cost calculations. However, all the schemes assume BS will reply to RREQ only if having sufficient resources.
constant bit rate (CBR) traffic and thus the approach is not To ensure sufficient route lifetime and to avoid frequent route
directly feasible for bursty traffic. updates, ARCE limits the maximum number of hops to five.
To summarize, there are four schemes proposed for reactive Each node waits for maximum of 10ms to merge all the
distributed routing. The LBS-AOMDV scheme reduces the received RREQs into a single RREQ, which contain link
routing overhead by selecting only those nodes having links information of all previous routes till that node. Therefore the
with given bandwidth threshold. In this way, the real time size of RREQ and RREP is larger than packets of other source
traffic demands could also be ensured. The RAR scheme based routing. This increases time delay however reduces the
considers energy efficiency and relay capabilities in routing overhead significantly. In contrast, if multiple RREQs are
decision. The RAR prefers to use operator installed fixed forwarded by a node then it increases the routing overhead
relay nodes over mobile relay nodes, and avoids routes having while keeping delay to minimum. If the node is neighbor of
nodes with minimal residual energy. The MCPR selects route the BS it forwards the RREQ only to the BS, or otherwise
with minimum consumed power for improving the energy broadcast to all its neighbors.
efficiency. The MMRP uses cumulative weight routing metrics
like hop count, traffic load and energy consumption, for route BS assigns an associativity value to each link. The as-
decision. The value of path cost metric could be set by sociativity value shows the freshness information received
themselves according to their requirements. For example, a about the link and the link is deleted when the associativity
device with fixed power supply does not consider the energy value decreases to zero. BS receives multiple RREQs from
metric. many routes and update the link load and associativity value
accordingly. The link load is calculated by taking into account
all routes that use a particular link. Multiple disjoint routes
G. Proactive distributed routing (3.1.2)
can then be calculated based on link load and associativity
In proactive distributed routing scheme, autonomous nodes value by using Dijkstra algorithm multiple times. Every time,
periodically update their routing tables in order to have con- BS increases the weights and then the Dijkstra algorithm
stantly updated route information to every other node in the is repeated to find a new route. BS informs about all the
network. candidate routes to the source node via the primary route. The
A Combined routing, channel scheduling and power control data packets contain the source route information. In case of
(CRSP) scheme has been proposed in [113]. The scheme any failed link, the intermediate node generates a route error
jointly considers the physical, data link and network layers. (RERR). On reception of RERR, the source node deletes all
The aim is to maximize the total network throughput by find- routes using that particular link and then use alternate backup
ing optimal solution for joint routing, scheduling and power route. Simulation results showed the improved performance
control. The optimal solution depends on the information of ARCE with two additional backup routes. However further
available about the network, the system constraints and the additional backup routes result in more overhead with no
defined objective function. The paper assumes that complete considerable performance gains since the routes used to break
channel state information is available which include path loss often. ARCE is not suitable for delay sensitive network due
and shadowing information, known to both the transmitter and to RREQ delay at every hop. It is also not suitable for large
the receiver while fading information is available only to the scale network since the maximum number of hops are limited
receiver. to five.
The only system constraint is to transmit C ∗ T bits in
T seconds where Cbits/sec is the required capacity. The To summarize, the two schemes for proactive distributed
objective function is to minimize the total transmit power routing are CRSP and ARCE. The CRSP scheme is proposed
for a fixed amount of information transfer, where power is to maximize the total network throughput by finding optimal
considered to be the only system resource. In general, every solution for joint routing, scheduling and power control. The
node knows about the traffic rate, induced interference, queue ARCE scheme increases the network capacity by bypassing
size, transmit power and the total induced interference (up the blocked BSs.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


H. Tree-based routing (3.2) and shuffling is used with NC in [124] and with multipath
In tree-based routing scheme, a hierarchy is maintained in routing in [50].
the form of tree with BS usually acting as a root node. The
tree-based routing is preferred when network scalability is of J. Quality of Service-based Routing (5)
major concern.
A distributed spanning tree (DST) algorithm that avoids In communication networks, three mechanisms are gen-
RREQ flooding problem and maintains the routes to the BS erally used to ensure the quality of service (QoS). These
with minimum routing overhead has been proposed in [114], are reservation, differentiation and over provisioning [125].
[115]. DST maintains a close to optimal spanning tree by using Among all the three mechanisms, reservation is the most
distributed topology trees. Maximum spanning tree is used to commonly used strategy, in which the resources are allocated
maximize the throughput. Each link capacity is updated based for a particular connection. In QoS based routing, the QoS
on measuring the response time from the periodic updates guarantee not only cover the single hop, but instead cover the
of the neighbor nodes. The refresh rate depends upon the entire multi-hop route.
frequency of the network changes. An admission control scheme with an on-demand quality of
There are two strategies proposed to update the child parent service routing (OQR) protocol is proposed in [126] for the
relationship. In the first strategy, a node cut the link with the multi-hop cellular network, that establishes virtual connection
parent and update the parent based on the periodic neighbor (VC) between the source and destination. The process of
node probe packet. O (d log n) messages are generated where establishing a QoS route is described as follows. The source
d is number of neighbor nodes, and n is the total number node broadcasts the route request (RREQ) packet. All the
of nodes in the network. In the second strategy, parent node neighbor nodes that received the RREQ packet, check the
sends updates periodically and the link with the parent is cut source address within the source list of packet to avoid
if the rate falls below threshold. The first strategy has more routing loops and the sequence number to reduce redundancy.
overhead but is also more optimal. Long routes are difficult to maintain in dynamic environment
By using the tree structure, the route discovery requires thus time-to-live (TTL) field in the RREQ packet limits the
only O (log n) messages instead of O(n) as in flooding. number of hops and also controls the flooding traffic since
For calculating maximum capacity path for m-th concurrent unlimited packet flooding degrades the network performance.
flow, the resources already consumed in m -1 flows are first The admission control scheme guarantees the bandwidth for
subtracted. It is important to note that due to interference, not real-time applications by finding a feasible bandwidth route.
only adjacent links but also links adjacent to neighbor nodes Thus the required bandwidth for an application is mentioned
are also blocked. Results show high network throughput which in terms of number of time slots.
ensure the suitability of DST for large size networks. However Each node calculates the link bandwidth by the intersection
there are two questionable things in the DST algorithm. Firstly, of the free slots of itself and the previous node. If the
the reduction of routing overhead in comparison to RREQ number of slots are not equal to required bandwidth slots
flooding is not shown in results. Secondly, the periodic route then RREQ is dropped. The process of forwarding the RREQ
updates are used, which are not suitable for highly dynamic packet continues and if the path bandwidth (or called end-
networks since the outage probability of such networks is very to-end bandwidth) is met till the destination node then the
high. RREP is replied to the original source. Multiple paths could
be established to ensure more robust packet delivery. When
enough RREQs are received by the destination node, then
I. Security-based routing (4) CLEAN RREQ packet can be broadcasted by the destination
The security of data in terms of confidentiality, authenticity node to clean the RREQ packets that are still roaming in
and integrity of data are of critical concerns in this age of data the network in order to reduce the flooding overhead. RREP
communications. sent by the destination node will reserve the slots backward
In multiple path routing protocols, data splitting and data till the source node. In case of failure (for example slot is
shuffling strategies are used in [50], [124] to encode the data, already booked or route breaks) at any node, a NACK (i.e.
for making the data confidential. In conjunction with encoding, RESERVE FAIL) is sent back to the destination node to clear
network coding (NC) scheme is used in [124] and disjoint all the reservations in between. If no route is available till the
multi-path routing scheme is used in [50] in order to safe data destination, the source node can be notified by a NACK (i.e.
from eavesdropper attacks. The data is splitted in chunks of NO ROUTE). If the source node do not receive any kind of
six bits each, and then some random pattern synchronized by response packet, then it can re-initiate the route request after
the source and destination is used to shuffle the data between a timeout.
different chunks. Each data chunk then travel through different Once a VC is setup, the source begins to send data packets.
paths to reach to the destination,. Results showed that an When data transfer is complete, the slots are released in order
eavesdropper can never intercept the data completely in any of to be available for new connections. If this last packet is lost,
the given cases. However, interception of data is more higher then the slots may erroneously remain reserved, thus each slot
in internal attack (i.e. when the eavesdropper is nearer to the is reserved for a timeout. If the route breaks while sending the
source) than as compared to external attack. Moreover, the data packet, a NACK (i.e. ROUTE BROKEN) is sent by the
quantity of data interception is more lower, when data splitting nodes at the breakage point to both the source and destination

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


in order to release the all the reserved slots in between, and the communication range of CA which is unrealistic. It is
also drops the data packets of this connection waiting in the also assumed that the algorithm avoids collision from other
queue. Results showed that the scheme greatly reduces the D2D UEs however CA is not defined for any other D2D UEs.
failure of the call setup and the routes are mostly very close Furthermore, the strategy is proposed only for downlink band,
to shortest paths. since it considers interference from BS but not from mobile
K. Device-aware routing (6) Shabany et al. proposed joined downlink rate allocation and
routing algorithm to improve the throughput of the system
In device-aware routing, each node decides to participate
[16]. Based on traffic pattern, the transmit power of the relays
in relaying the data based on its capabilities. Thus, the
and BSs is adjusted which results in cell breathing. However,
routing can be easily deployed in network having devices with
the proposed heuristic algorithm is limited to use two operator
heterogeneous capabilities.
installed relays and two BSs while using unlicensed band for
A device-aware routing and scheduling algorithm (DARA)
their data forwarding. The two famous ad-hoc routing proto-
is proposed in which each device decides how much data it
cols AODV and DSDV have been simulated in the cellular
can handle based on its processing power, remaining energy,
network environment to check the suitability in terms of data
and incentives [127]. A network utility maximization (NUM)
latency in [51]. The NS3 simulator with LENA framework is
formulation has been developed in order to incorporate device
used to evaluate latency of the AODV and DSDV in varying
capabilities. A node after deciding to participate in forwarding
number of nodes. Bello and Zeadally identify the requirement
the data (routing decision); activates the link for the next hop
of adaptable and scalable routing algorithms that can work in a
node (scheduling decision) and thus the algorithm involves
complex and heterogeneous scenarios and in varying network
both the routing and the scheduling decisions. The DARA
sizes. The authors compare the advantages and limitations
has been compared to the traditional back-pressure algorithm
of three categories of routing, suitable for intelligent D2D
wherein each device has to forward and schedule all the data
communications in the Internet of things (IoT), which are
it receives, even if a device does not have sufficient resources.
stochastic, bio-inspired, and context-aware routing [129].
The DARA has been implemented between the transport layer
It is worth mentioning that due to the diverse types of possi-
and the application layer and thus can be easily adjusted to
ble distributed routing schemes, the ad-hoc routing protocols
any operating system of the mobile devices. The results from
can use any of the suitable relays for their data forwarding
the test-bed have shown significant throughput improvement
(unlike previous sections, where D2D relay is used for only
over the back-pressure algorithm.
D2D routing, and network relay for only D2I/I2D routing) as
shown in Table V. The same is true for the type of spectrum
L. Summary and Insights used as shown in Table VI.
In case of network congestion or network unavailability, an
ad-hoc routing protocol is the only choice for multi-hop D2D
communications. However, it is important to note that these
distributed routing protocols can not ensure security as any D2D communication has attained significant attention by
other infrastructure-based centralized control entity can do. researchers in current decade. Thousands of efficient schemes
There are many works in literature that do not propose any under various D2D research domains have been proposed.
new routing scheme but instead use or check the suitability of However it is considerably noted in some domains that perfor-
existing routing schemes. Few of the works are discussed in mance comes at the cost of complexity. For example, in most
detail as follows. Yuan et al. and Wang et al. use shortest path of the network coding schemes for D2D communications, it is
routing, and evaluate the collision and outage probability in found that performance enhancements is achieved at the cost
multi-hop and conventional cellular communication scenarios of high computational complexity [34]. Such complexity is
[3], [128]. As the location accuracy is complex and power usually multiplied in case of multi-hop communications. Thus,
consuming in cellular networks, therefore collision area (CA) routing in multi-hop D2D networks needs special attention
is defined to mitigate interference from the BS and other in order to avoid incorporating high computationally complex
D2D UEs [3]. The shortest path routing (SPR) is used to algorithms in its design.
explore the multi-hop path and find the collision probability Routing in D2D communication must be capable to incorpo-
for the intermediate nodes in the routing path. The probability rate and integrate unique characteristics of D2D communica-
for collision invokes the switching strategy between D2D tions, such as resource allocation, power control, interference
and conventional cellular (CC) communications to minimize management, mode selection, scheduling, device discovery, se-
the interference. A collision area is defined as a circular curity and medium access control (MAC) mechanisms. There-
area (CA) around an interfering node like BS. The radius fore, compatibility/applicability of promising routing tech-
of CA is variable and depends on the QoS requirements. niques, such as opportunistic routing, cooperative routing and
Each node can check whether it is in CA or not through the network coding-aware routing, need to be verified/validated in
pilot channel power from the BS. Thus the receiver node’s D2D network scenarios. Some important routing techniques
SINR is greater than required SINR only if it is outside the which can be considered in D2D communications are dis-
range of CA. However interference from neighbor BSs is not cussed below. In each subsection, a brief introduction of the
considered. It is assumed that destination node is not within technique is first discussed, followed by the advantage (in

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


terms of an opportunity or a method to solve a problem etc.) D. Cross Layer Routing

that D2D communication can take to improve the D2D routing. A cross layer routing protocol presented in [90] has shown
better performance in terms of throughput, delay and in-
terference. A survey on routing metric for cognitive radio
A. Opportunistic Routing networks (CRNs) suggests that based on the challenges in
CRN, cross layer routing in CRN should be a must, and not
Opportunistic routing provides reliable delivery of packet
an option [136]. The importance of cross layer-aided routing
from a source node to a destination node in sparse and
schemes aiming at energy efficiency is also presented in detail
disconnected networks. Although this routing technique is not
in [137]. Based on these studies, it can easily be derived
suitable for delay sensitive applications but is found flexible
that if the information about physical layer and data link
and easily adaptable to high network dynamics. This is because
layer mechanisms like D2D resource allocation and scheduling
opportunistic routing does not require any network topol-
decisions are available at the network layer, then better routing
ogy information which is very expensive in highly dynamic
decisions can be made accordingly which can improve the
networks. Instead, opportunistic routing uses the broadcast
overall performance of D2D networks.
nature of nodes, thus the packet is received by all neighbor
nodes, and only the opportunistically selected candidate relay
nodes among the next hops nodes will forward the packet. E. Multicast Routing
The process continues until the packet reaches the destination In highly dynamic networks, when high amount of data
node. Opportunistic routing has been classified into five broad transfer to multiple destinations is required, in applications,
categories [17]. Among them, the concept of Geographic such as video conferencing or mobile software updates; mul-
Opportunistic routing can be utilized in Location based D2D ticast routing shows improved QoS and energy efficiency in
routing specifically, and the Link State aware Opportunistic such scenarios [138]. In order to get benefit from the multicast
routing (for improving network reliability) and Probabilistic routing technique in D2D networks, few schemes have been
Opportunistic routing (for supporting nodes mobility) in gen- proposed [139]–[141]. However some efficient schemes that
eral, to leverage the D2D routing. Opportunistic routing also can perfectly integrate unique characteristics of D2D commu-
supports high amount of data transfers which along with other nication is still missing. Some general key-points to easily
benefits is very appealing for D2D communications. Marin et and efficiently design a multicast routing algorithm have been
al. [130] provide a good example (as basis) of integrating the discussed in [142].
D2D and opportunistic networks.
F. Impact of Mobility Models on Routing Decision
Mobility models provide a good estimate of current move-
B. Cooperative Routing ment of node in terms of position, direction, velocity and
acceleration. By inspecting the movement pattern of the node,
Cooperative communication is used to alleviate the mul- respective mobility model can be applied in order to have
tipath (channel) fading and to improve the communication better routing decisions. A number of human and vehicle
reliability in wireless networks. It can be used in D2D com- mobility models have been proposed which would be helpful
munications to reduce the interference caused by D2D UEs. in designing a better routing algorithm in D2D networks. A
In cooperative communication, a node helps other nodes by survey on the impact of mobility on opportunistic routing
exploiting the broadcast nature. It is called cooperative routing algorithms [143] is a good example to derive ideas for routing
because it takes advantage of such cooperative transmission at in D2D networks.
the physical layer [131], [132]. Since interference mitigation
is a very critical problem in D2D communication, therefore
G. White space-aware routing
this routing technique is a very good candidate to be used in
D2D communication. The unused TV white spaces frequencies spectrum have
been opportunistically exploited for cellular network in [144].
The results have shown improved network throughput. A
number of routing protocols in CRNs also take advantage
C. Network Coding-aware Routing
of the presence of free (unused) TV white spaces and make
Interestingly, like opportunistic routing and cooperative spatial use for certain time and frequency to improve various
routing, the Network coding-aware (NC-aware) routing tech- network performance parameters [145], [146]. Since, the D2D
nique also uses omni-directional antenna to exploit broadcast communication is flexible to use a variety of frequency spec-
nature of the wireless medium. NC-aware routing technique trum for their communications, as discussed in section III,
shows many performance improvements over traditional rout- therefore it would be interesting to exploit TV white space
ing which includes improved throughput, higher reliability, spectrum in routing protocols for D2D networks. There are
lower data transmission rate, lower delay and minimum energy four white space paradigms: underlay, overlay, interweave, and
consumption [133]–[135]. Network Coding also proved its hybrid [147]. One future research direction is also to consider
effectiveness in terms of achieving maximum information these four white space paradigms and propose routing schemes
flow in D2D networks [34]. Thus, D2D routing can use this in future to exploit such white space paradigms for TV white
technique to get benefit from its performance improvements. spaces.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


H. Experimental evaluation of D2D routing establish route from node to BS or vice versa i.e. for D2I/I2D
To the best of our knowledge, there are only two practical routing.
implementations for multi-hop D2D communications, so far.
In [13], a set of ten nodes had been setup in an indoor envi- VIII. C ONCLUSION
ronment for validating the network performance improvement
(higher data rate, in this experiment) as promised by multi- This survey provides a comprehensive and in-depth analysis
hop D2D communications. In another field experiment [148], of the routing schemes for multi-hop cellular D2D networks.
a set of twenty nodes had been deployed in Sendai City, Japan. The aim of this survey paper is to study the characteristics
Both studies use WiFi, thus neither use the cellular licensed of each routing protocol. The classification of the routing
spectrum, nor take advantage of the presence of infrastructure schemes, presented in this paper, will help to select a routing
network. Therefore, there is a need of routing prototype that scheme based on varying user and network requirements. This
provides experimental evaluation of large number of nodes in paper also discussed various unexplored areas of D2D com-
an urban scenario while offloading real-time cellular traffic. munications, such as D2D history (as there are many routing
schemes for D2D networks in the past with different names),
and performance improvements including the operators/users
I. Routing for mm-wave D2D Communication in 5G Networks benefits. An explanation for the suitability of using uplink band
Millimeter-wave (mm-wave) communications have shown for D2D communications is also presented in this paper. We
great potential for providing efficient high speed data transmis- also discussed the open research problems and highlighted the
sion for small distance communications. This is because mm- future research directions that require further investigations.
wave uses channels from 30 GHz to 300 GHz, which although Thus, we believe that our survey will serve as the foundation
have (i) limited diffraction capability, (ii) difficulty in com- and benchmark for routing in D2D networks.
munication signals passing through walls and (iii) difficulty
in NLOS transmissions, but provides very high bandwidth. In
order to cope with mentioned challenges associated with the R EFERENCES
extremely high frequency range and to provide high data rate [1] S.-H. Kim and S.-J. Han, “Contour routing for peer-to-peer DTN
using the highly available bandwidth, mm-wave D2D multi- delivery in cellular networks,” in 4th International Conference on
hop routing protocols for 5G networks have been proposed Communication Systems and Networks (COMSNETS), 2012, pp. 1–9.
[2] H. Imai, H. Okada, T. Yamazato, and M. Katayama, “The effect of
[149], [150]. The proposed schemes promise higher QoS multipath hybrid routing protocol in multihop cellular networks,” in
for multimedia applications, however require mathematical IEEE 17th International Symposium on Personal, Indoor and Mobile
modeling and are distributed in nature. Thus some simple and Radio Communications (PIMRC), 2006, pp. 1–5.
[3] H. Yuan, W. Guo, and S. Wang, “D2D multi-hop routing: collision
efficient routing protocols for mm-wave D2D networks which probability and routing strategy with limited location information,” in
also take benefit from the presence of infrastructure network IEEE International Conference on Communication Workshop (ICCW),
(e.g. [151]) are highly desired. 2015, pp. 670–674.
[4] Third Generation Partnership Project (3GPP), “Study on LTE Device
to Device Proximity Services; Radio Aspects,” in 3GPP TR 36.843
version 12.0.1, 2014.
J. New research gaps identified by our proposed classification [5] Third Generation Partnership Project (3GPP), “LTE Device to Device
of D2D routing (D2D) Proximity Services (ProSe) User Equipment (UE) radio trans-
mission and reception,” in 3GPP TR 36.877 version 12.0.0, 2015.
Our proposed taxonomy of state-of-the-art research works [6] Third Generation Partnership Project (3GPP), “Study on further en-
provide a very good basis and foundation to identify the hancements to LTE Device to Device (D2D), User Equipment (UE) to
research gaps in D2D and D2I/I2D routing, where either no network relays for Internet of Things (IoT) and wearables,” in 3GPP
TR 36.746 version 15.1.1, 2018.
work or very small amount of work has been done, and [7] C. Tata and M. Kadoch, “Multipath routing algorithm for device-
thus further research investigations can be carried out. For to-device communications for public safety over LTE heterogeneous
example, social networks have been well exploited in D2D networks,” in 1st International Conference on Information and Com-
munication Technologies for Disaster Management (ICT-DM), 2014,
communication [43], however socially-aware algorithms have pp. 1–7.
not been utilized for D2I/I2D routing except for few works [8] H. Yuan, W. Guo, and S. Wang, “Emergency route selection for D2D
in D2D routing [79]–[81]. Similarly, an extensive research cellular communications during an urban terrorist attack,” in IEEE
International Conference on Communications Workshops (ICC), 2014,
has been made for security solutions in D2D communications pp. 237–242.
[33], [35], [38], [41], however there is one only scheme [9] S.-Y. Lien, C.-C. Chien, G. S.-T. Liu, H.-L. Tsai, R. Li, and Y. J.
proposed that takes advantage of the presence of BS for Wang, “Enhanced LTE Device-to-Device Proximity Services,” IEEE
Communications Magazine, vol. 54, no. 12, pp. 174–182, 2016.
keeping confidentiality, authenticity and integrity of the data
[10] G. Nardini, G. Stea, and A. Virdis, “A Fast and Reliable Broad-
[82]. It is important to note that there is no work on D2I/I2D cast Service for LTE-Advanced Exploiting Multihop Device-to-Device
routing scheme that provides secure route. The idea of content- Transmissions,” Future Internet, vol. 9, no. 4, p. 89, 2017.
based routing is very appealing for D2D networks, where [11] L. Babun, A. I. Yürekli, and I. Güvenç, “Multi-hop and D2D communi-
cations for extending coverage in public safety scenarios,” in IEEE 40th
the route does not depend upon the destination node, but is Local Computer Networks Conference Workshops (LCN Workshops),
established based on the contents of the information being 2015, pp. 912–919.
shared. However, not enough work has been made neither [12] L. Melki, S. Najeh, and H. Besbes, “Interference management scheme
for network-assisted multi-hop D2D communications,” in IEEE 27th
for D2D nor for D2I/I2D routing. Last but not the least, no Annual International Symposium on Personal, Indoor, and Mobile
existing works take advantage of the ’location information’ to Radio Communications (PIMRC), 2016, pp. 1–5.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


[13] H. Qin, Z. Mi, C. Dong, F. Peng, and P. Sheng, “An Experimental [36] S. Ali and A. Ahmad, “Resource allocation, interference management,
Study on Multihop D2D Communications Based on Smartphones,” in and mode selection in device-to-device communication: A survey,”
IEEE 83rd Vehicular Technology Conference (VTC Spring), 2016, pp. Transactions on Emerging Telecommunications Technologies, 2017.
1–5. [37] S. Umrao, A. Roy, and N. Saxena, “Device-to-device communication
[14] P. Li, Y. Chen, J. Jin, and J. Mou, “Adaptive routing protocol based from control and frequency perspective: A composite review,” IETE
on OLSR for hybrid multi-hop and cellular network,” in International Technical Review, vol. 34, no. 3, pp. 286–297, 2017.
Conference of IEEE Region 10 (TENCON), 2013, pp. 1–5. [38] M. Wang and Z. Yan, “A survey on security in D2D communications,”
[15] B. Timu and P. Soldati, “A joint routing-MAC model for cellular- Mobile Networks and Applications, vol. 22, no. 2, pp. 195–208, 2017.
relaying networks,” in IEEE 19th International Symposium on Per- [39] P. Gandotra, R. K. Jha, and S. Jain, “Green communication in next
sonal, Indoor and Mobile Radio Communications (PIMRC), 2008, pp. generation cellular networks: A survey,” IEEE Access, vol. 5, pp.
1–6. 11 727–11 758, 2017.
[16] M. Shabany and E. S. Sousa, “Joint rate allocation and routing scheme [40] M. Ahmad, M. Azam, M. Naeem, M. Iqbal, A. Anpalagan, and M. Ha-
in multihop cellular CDMA networks,” in 9th International Symposium neef, “Resource management in D2D communication: An optimization
on Computers and Communications (ISCC), vol. 1, 2004, pp. 442–447. perspective,” Journal of Network and Computer Applications, vol. 93,
[17] N. Chakchouk, “A survey on opportunistic routing in wireless com- pp. 51–75, 2017.
munication networks,” IEEE Communications Surveys & Tutorials, [41] O. N. Hamoud, T. Kenaza, and Y. Challal, “Security in device-to-device
vol. 17, no. 4, pp. 2214–2241, 2015. communications: a survey,” IET Networks, vol. 7, no. 1, pp. 14–22,
[18] D. Cavalcanti, D. Agrawal, C. Cordeiro, B. Xie, and A. Kumar, “Issues
[42] M. Höyhtyä, O. Apilo, and M. Lasanen, “Review of Latest Advances
in integrating cellular networks WLANs, and MANETs: a futuristic
in 3GPP Standardization: D2D Communication in 5G Systems and Its
heterogeneous wireless network,” IEEE Wireless Communications,
Energy Consumption Models,” Future Internet, vol. 10, no. 1, p. 3,
vol. 12, no. 3, pp. 30–41, 2005.
[19] I. W. I. C. S. Inc., “ODMA History,” accessed: April 2018. [Online].
[43] M. Ahmed, Y. Li, M. Waqas, M. Sheraz, D. Jin, and Z. Han, “A
Survey on Socially-Aware Device-to-Device Communications,” IEEE
[20] Third Generation Partnership Project (3GPP), “Opportunity Driven Communications Surveys & Tutorials, 2018.
Multiple Access,” in 3GPP TR 25.924 version 1.0.0, 1999. [44] F. Jameel, Z. Hamid, F. Jabeen, S. Zeadally, and M. A. Javed, “A
[21] Y.-D. Lin and Y.-C. Hsu, “Multihop cellular: A new architecture for Survey of Device-to-Device Communications: Research Issues and
wireless communications,” in 19th Annual Joint Conference of the Challenges,” IEEE Communications Surveys & Tutorials, 2018.
IEEE Computer and Communications Societies (INFOCOM), vol. 3, [45] S. Adibi, “A mobile health network disaster management system,”
2000, pp. 1273–1282. in 7th International Conference on Ubiquitous and Future Networks,
[22] H. Lee and C.-C. Lee, “An integrated multihop cellular data network,” 2015, pp. 424–428.
in IEEE 58th Vehicular Technology Conference (VTC Fall), vol. 4, [46] K. Gomez, L. Goratti, T. Rasheed, and L. Reynaud, “Enabling disaster-
2003, pp. 2232–2236. resilient 4G mobile communication networks,” IEEE Communications
[23] H. Lee and Y. Li, “A hierarchical routing with load balanced scheduling Magazine, vol. 52, no. 12, pp. 66–73, 2014.
in the integrated multi-hop cellular data network,” in 1st International [47] A. Al-Hourani, S. Kandeepan, and A. Jamalipour, “Stochastic geometry
Symposium on Wireless Pervasive Computing, 2006, pp. 6–pp. study on device-to-device communication as a disaster relief solution,”
[24] A. Asadi, Q. Wang, and V. Mancuso, “A survey on device-to-device IEEE Transactions on Vehicular Technology, vol. 65, no. 5, pp. 3005–
communication in cellular networks,” IEEE Communications Surveys 3017, 2016.
& Tutorials, vol. 16, no. 4, pp. 1801–1819, 2014. [48] S. Ghafoor, P. Sutton, C. Sreenan, and K. Brown, “Cognitive radio for
[25] R. Alkurd, R. M. Shubair, and I. Abualhaol, “Survey on device-to- disaster response networks: survey, potential, and challenges,” IEEE
device communications: Challenges and design issues,” in IEEE 12th Wireless Communications, vol. 21, no. 5, pp. 70–80, 2014.
International New Circuits and Systems Conference (NEWCAS), 2014, [49] K. Muraoka, J. Shikida, and H. Sugahara, “Feasibility of capacity
pp. 361–364. enhancement of public safety LTE using device-to-device communica-
[26] P. Mach, Z. Becvar, and T. Vanek, “In-Band Device-to-Device Com- tion,” in International Conference on Information and Communication
munication in OFDMA Cellular Networks: A Survey and Challenges,” Technology Convergence (ICTC), 2015, pp. 350–355.
IEEE Communications Surveys & Tutorials, vol. 17, no. 4, pp. 1885– [50] C. Tata and M. Kadoch, “Secure Multipath Routing Algorithm for
1922, 2015. Device-to-Device Communications for Public Safety over LTE Hetero-
[27] J. Liu, N. Kato, J. Ma, and N. Kadowaki, “Device-to-Device Communi- geneous Networks,” in 3rd International Conference on Future Internet
cation in LTE-Advanced Networks: A Survey,” IEEE Communications of Things and Cloud (FiCloud), 2015, pp. 212–217.
Surveys Tutorials, vol. 17, no. 4, pp. 1923–1940, Fourthquarter 2015. [51] P. Masek, A. Muthanna, and J. Hosek, “Suitability of MANET Routing
[28] X. Cheng, Y. Li, B. Ai, X. Yin, and Q. Wang, “Device-to-device Protocols for the Next-Generation National Security and Public Safety
channel measurements and models: a survey,” IET Communications, Systems,” in Internet of Things, Smart Spaces, and Next Generation
vol. 9, no. 3, pp. 312–325, 2015. Networks and Systems. Springer, 2015, pp. 242–253.
[52] D. Munir, J. Gu, and M. Y. Chung, “Selection of UE relay considering
[29] R. Maallawi, N. Agoulmine, B. Radier, and T. Ben Meriem, “A com-
QoS class for public safety services in LTE-A network,” in 20th Asia-
prehensive survey on offload techniques and management in wireless
Pacific Conference on Communication (APCC), 2014, pp. 401–405.
access and core networks,” IEEE Communications Surveys & Tutorials,
[53] X. Li, D. Guo, H. Yin, and G. Wei, “Drone-assisted public safety
vol. 17, no. 3, pp. 1582–1604, thirdquarter 2015.
wireless broadband network,” in IEEE Wireless Communications and
[30] H. Mustafa, M. Imran, M. Shakir, A. Imran, and R. Tafazolli, “Separa-
Networking Conference Workshops (WCNCW), 2015, pp. 323–328.
tion Framework: An Enabler for Cooperative and D2D Communication
[54] G. Fodor, S. Parkvall, S. Sorrentino, P. Wallentin, Q. Lu, and
for Future 5G Networks,” IEEE Communications Surveys & Tutorials,
N. Brahmi, “Device-to-device communications for national security
vol. 18, no. 1, pp. 419–445, Firstquarter 2016.
and public safety,” IEEE Access, vol. 2, pp. 1510–1520, 2014.
[31] M. Noura and R. Nordin, “A survey on interference management [55] A. Al-Hourani and S. Kandeepan, “Temporary cognitive femtocell
for Device-to-Device (D2D) communication and its challenges in 5G network for public safety LTE,” in IEEE 18th International Workshop
networks,” Journal of Network and Computer Applications, 2016. on Computer Aided Modeling and Design of Communication Links and
[32] P. Gandotra and R. K. Jha, “Device-to-device communication in cellular Networks (CAMAD), 2013, pp. 190–195.
networks: A survey,” Journal of Network and Computer Applications, [56] L. Goratti, G. Steri, K. M. Gomez, and G. Baldini, “Connectivity and
vol. 71, pp. 99–117, 2016. security in a D2D communication protocol for public safety applica-
[33] M. Haus, M. Waqas, A. Y. Ding, Y. Li, S. Tarkoma, and J. Ott, “Security tions,” in 11th International Symposium on Wireless Communications
and Privacy in Device-to-Device (D2D) Communication: A Review,” Systems (ISWCS), 2014, pp. 548–552.
IEEE Communications Surveys & Tutorials, 2017. [57] E. Yaacoub and O. Kubbar, “Energy-efficient device-to-device com-
[34] A. Douik, S. Sorour, T. Y. Al-Naffouri, and M.-S. Alouinig, “Instantly munications in LTE public safety networks,” in IEEE Globecom
Decodable Network Coding: From Centralized to Device-to-Device Workshops (GC Wkshps, 2012, pp. 391–395.
Communications,” IEEE Communications Surveys & Tutorials, 2017. [58] E. Yaacoub, “On the use of device-to-device communications for
[35] P. Gandotra, R. K. Jha, and S. Jain, “A survey on device-to-device QoS and data rate enhancement in LTE public safety networks,” in
(D2D) communication: Architecture and security issues,” Journal of IEEE Wireless Communications and Networking Conference Work-
Network and Computer Applications, vol. 78, pp. 9–29, 2017. shops (WCNCW), 2014, pp. 236–241.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


[59] L. Goratti, K. M. Gomez, R. Fedrizzi, and T. Rasheed, “A novel device- in IEEE International Conference on Communications (ICC), 2015, pp.
to-device communication protocol for public safety applications,” in 2961–2966.
IEEE Globecom Workshops (GC Wkshps), 2013, pp. 629–634. [81] I. O. Nunes, P. O. V. de Melo, and A. A. Loureiro, “Leveraging
[60] S. B. Seo, J. Y. Kim, and W. S. Jeon, “Robust and fast device discovery D2D Multi-Hop Communication Through Social Group Meetings
in OFDMA-based cellular networks for disaster environment,” in 18th Awareness,” IEEE Wireless Communications, 2016.
International Conference on Advanced Communication Technology [82] S. Othmen, F. Zarai, A. Belghith, and L. Kamoun, “Anonymous and
(ICACT), 2016, pp. 498–502. secure on-demand routing protocol for multi-hop cellular networks,”
[61] E. Onem, S. Eryigit, T. Tugcu, and A. Akurgal, “QoS-Enabled in International Symposium on Networks, Computers and Communi-
Spectrum-Aware Routing for Disaster Relief and Tactical Operations cations (ISNCC), 2016, pp. 1–6.
over Cognitive Radio Ad Hoc Networks,” in IEEE Military Communi- [83] B. Kaufman, J. Lilleberg, and B. Aazhang, “Spectrum sharing scheme
cations Conference (MILCOM), 2013, pp. 1109–1115. between cellular users and ad-hoc device-to-device users,” IEEE Trans-
[62] M. Hunukumbure, T. Moulsley, A. Oyawoye, S. Vadgama, and actions on Wireless Communications, vol. 12, no. 3, pp. 1038–1049,
M. Wilson, “D2D for energy efficient communications in disaster and 2013.
emergency situations,” in 21st International Conference on Software, [84] F. S. Shaikh and R. Wismüller, “Interference-Conscious Routing in
Telecommunications and Computer Networks (SoftCOM), 2013, pp. 1– Multihop D2D Communications,” in 2nd International Conference on
5. Computer and Communication Systems (ICCCS), 2017, pp. 146–151.
[63] M. Macuha, D. Amgalan, and T. Derham, “Device-centric approach to [85] R. Ananthapadmanabha, B. Manoj, and C. Murthy, “Multi-hop cel-
improve resiliency of emergency communication,” in IEEE Region 10 lular networks: the architecture and routing protocols,” in 12th IEEE
Humanitarian Technology Conference (R10-HTC), 2013, pp. 124–129. International Symposium on Personal, Indoor and Mobile Radio Com-
[64] T. Castel, S. Lemey, S. Agneessens, P. Van Torre, H. Rogier, and munications (PIMRC), vol. 2, 2001, pp. G–78.
C. Oestges, “LTE as a potential standard for public safety indoor [86] Y.-C. Hsu and Y.-D. Lin, “Base-centric routing protocol for multihop
body-to-body networks,” in IEEE Symposium on Communications and cellular networks,” in IEEE Global Telecommunications Conference
Vehicular Technology in the Benelux (SCVT), 2015, pp. 1–6. (GLOBECOM), vol. 1, 2002, pp. 158–162.
[65] X. Lin, J. Andrews, A. Ghosh, and R. Ratasuk, “An overview of 3GPP [87] H. Li, D. Yu, and H. Chen, “New approach to multihop-cellular based
device-to-device proximity services,” IEEE Communications Magazine, multihop network,” in IEEE 14th Proceedings on Personal, Indoor and
vol. 52, no. 4, pp. 40–48, 2014. Mobile Radio Communications (PIMRC), vol. 2, 2003, pp. 1629–1633.
[66] G. Fodor, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklós, [88] X. Han, H. Wang, M. Pan, Y. Ji, and P. Zhang, “Light Dark Routing
and Z. Turányi, “Design aspects of network assisted device-to-device Protocol: An innovative Infrastructure-assisted Ad hoc Routing Proto-
communications,” IEEE Communications Magazine, vol. 50, no. 3, pp. col,” in IEEE Global Telecommunications Conference (GLOBECOM),
170–177, 2012. 2006, pp. 1–5.
[67] B. Coll-Perales and J. Gozalvez, “On the capability of multi-hop cel- [89] F. S. Shaikh and R. Wismüller, “Centralized Adaptive Routing
lular networks with mobile relays to improve handover performance,” in Multihop Cellular D2D Communications,” in 2nd International
in 8th International Symposium on Wireless Communication Systems Conference on Computer and Communication Systems (ICCCS), 2017,
(ISWCS), 2011, pp. 207–211. pp. 158–162.
[68] I. Gruber and H. Li, “A novel ad hoc routing algorithm for cellular cov- [90] G. Kannan, S. N. Merchant, and U. B. Desai, “Cross layer routing
erage extension,” in 11th International Telecommunications Network for multihop cellular networks,” in 21st International Conference
Strategy and Planning Symposium. NETWORKS, 2004, pp. 163–168. on Advanced Information Networking and Applications Workshops
[69] H. Tullberg, Z. Li, A. Hoglund, P. Fertl, D. Gozalvez-Serrano, (AINAW), vol. 2, 2007, pp. 165–170.
K. Pawlak, P. Popovski, G. Mange, and O. Bulakci, “Towards the [91] H. Yuan, W. Guo, Y. Jin, S. Wang, and M. Ni, “Interference-aware
METIS 5G concept: First view on Horizontal Topics concepts, booktitle multi-hop path selection for device-to-device communications in a
= European Conference on Networks and Communications (EuCNC),” cellular interference environment,” IET Communications, 2017.
2014, pp. 1–5. [92] V. Bhardwaj and C. R. Murthy, “On optimal routing and power
[70] P. Li and S. Guo, “Incentive mechanisms for device-to-device commu- allocation for D2D communications,” in IEEE International Conference
nications,” IEEE Network, vol. 29, no. 4, pp. 75–79, 2015. on Acoustics, Speech and Signal Processing (ICASSP), 2015, pp. 3063–
[71] Y. Zhang, L. Song, W. Saad, Z. Dawy, and Z. Han, “Contract-based 3067.
incentive mechanisms for device-to-device communications in cellu- [93] P. Ren, Q. Du, and L. Sun, “Interference-aware routing for hop-count
lar networks,” IEEE Journal on Selected Areas in Communications, minimization in wireless D2D networks,” in IEEE/CIC International
vol. 33, no. 10, pp. 2144–2155, 2015. Conference on Communications in China-Workshops (CIC/ICCC),
[72] N. Mastronarde, V. Patel, and L. Liu, “Device-to-device relay assisted 2013, pp. 65–70.
cellular networks with token-based incentives,” in IEEE International [94] Q. Du, H. Song, Q. Xu, P. Ren, and L. Sun, “Interference-controlled
Conference on Communication Workshop (ICCW), 2015, pp. 698–704. D2D routing aided by knowledge extraction at cellular infrastructure
[73] F. Tian, B. Liu, J. Xiong, and L. Gui, “Movement-based incen- towards ubiquitous CPS,” Personal and Ubiquitous Computing, vol. 19,
tive for cellular traffic offloading through D2D communications,” in no. 7, pp. 1033–1043, 2015.
IEEE International Symposium on Broadband Multimedia Systems and [95] Z. Jingyi, L. Xi, and X. Quansheng, “Multi-hop routing for energy-
Broadcasting (BMSB), 2016, pp. 1–5. efficiency enhancement in relay-assisted device-to-device communica-
[74] N. Mastronarde, V. Patel, J. Xu, and M. van der Schaar, “Learning tion,” The Journal of China Universities of Posts and Telecommunica-
relaying strategies in cellular D2D networks with token-based incen- tions, vol. 22, no. 2, pp. 1–51, 2015.
tives,” in IEEE Globecom Workshops (GC Wkshps), 2013, pp. 163–169. [96] B. Coll-Perales and J. Gozalvez, “Energy efficient routing protocols for
[75] F. Malandrino, C. Casetti, and C.-F. Chiasserini, “Toward D2D- multi-hop cellular networks,” in IEEE 20th International Symposium on
enhanced heterogeneous networks,” IEEE Communications Magazine, Personal, Indoor and Mobile Radio Communications (PIMRC), 2009,
vol. 52, no. 11, pp. 94–100, 2014. pp. 1457–1461.
[76] M. N. Tehrani, M. Uysal, and H. Yanikomeroglu, “Device-to-device [97] B. Coll-Perales and J. Gozalvez, “Contextual optimization of location-
communication in 5G cellular networks: challenges, solutions, and based routing protocols for multi-hop cellular networks using mobile
future directions,” IEEE Communications Magazine, vol. 52, no. 5, relays,” Telecommunication Systems, pp. 1–13, 2015.
pp. 86–92, 2014. [98] D. B. Johnson, “The dynamic source routing protocol for mobile ad
[77] H.-C. Jang and Y.-J. Lee, “QoS-constrained resource allocation hoc networks,” draft-ietf-manet-dsr-09. txt, 2003.
scheduling for LTE network,” in International Symposium on Wireless [99] T. Finke, “SEREMA: Self-Organized Routing in Heterogeneous Mobile
and Pervasive Computing (ISWPC), 2013, pp. 1–6. Ad Hoc Networks,” Ph.D. dissertation, Ilmenau University of Technol-
[78] G. Ding, J. Wang, Q. Wu, Y.-D. Yao, F. Song, and T. A. Tsiftsis, ogy, Germany, 2016.
“Cellular-base-station-assisted device-to-device communications in tv [100] T. A. Ramrekha, G. P. Millar, and C. Politis, “A model for designing
white space,” IEEE Journal on Selected Areas in Communications, scalable and efficient adaptive routing approaches in emergency ad hoc
vol. 34, no. 1, pp. 107–121, 2016. communications,” in IEEE Symposium on Computers and Communi-
[79] I. Koutsopoulos, E. Noutsi, and G. Iosifidis, “Dijkstra goes social: cations (ISCC), 2011, pp. 916–923.
Social-graph-assisted routing in next generation wireless networks,” in [101] S. Madabhushi, G. R. Gopal, and C. R. Murthy, “Optimal routing and
20th European Wireless Conference, 2014, pp. 1–7. data transmission for multi-hop D2D communications under stochastic
[80] Y. Zhao, Y. Li, H. Mao, and N. Ge, “Social community aware long- interference constraints,” in 23rd National Conference on Communica-
range link establishment for multi-hop D2D communication networks,” tions (NCC), 2017, pp. 1–6.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


[102] C. Casetti, C.-F. Chiasserini, L. C. Pelle, C. Del Valle, Y. Duan, and in IEEE 66th Vehicular Technology Conference (VTC Fall), 2007, pp.
P. Giaccone, “Content-centric Routing in Wi-Fi Direct Multi-group 6–10.
Networks,” in IEEE 16th International Symposium on a World of [124] C. Tata and M. Kadoch, “Secure network coding based data splitting
Wireless, Mobile and Multimedia Networks (WoWMoM), June 2015, for public safety D2D communications over LTE heterogeneous net-
pp. 1–9. works,” in 14th International Conference on Computer and Information
[103] H.-H. Choi and D.-H. Cho, “Fast and reliable route discovery protocol Technology (CIT), 2014.
considering mobility in multihop cellular networks,” in IEEE 63rd [125] A. Mitschele-Thiel, “Quality of Service (in Mobile Networks),”
Vehicular Technology Conference (VTC Spring), vol. 2, 2006, pp. 886– Lecture Slides for course Advanced Mobile Communication
890. Networks, Technische Universitat Ilmenau, Germany . [Online].
[104] R.-G. Cheng and J.-Y. Hung, “Power-Strength-Based Selective For- Available:
warding (PSF) Mechanism for Multihop Cellular Networks,” in IEEE mobilkommunikationsnetze/AMCN%20-%20QoS%20-%20ss18.pptx.
Wireless Communications and Networking Conference (WCNC), 2007, pdf
pp. 3613–3618. [126] C. R. Lin, “On-demand QoS routing in multihop mobile networks,” in
[105] R.-G. Cheng, J.-Y. Hung, and Y.-Y. Liu, “Power-strength-based Se- 20th Annual Joint Conference of the IEEE Computer and Communi-
lective Forwarding (PSF) mechanism for multihop cellular networks,” cations Societies (INFOCOM), vol. 3, 2001, pp. 1735–1744.
IEEE Transactions on Vehicular Technology, vol. 59, no. 7, pp. 3455– [127] Y. Xing and H. Seferoglu, “Device-aware routing and scheduling
3467, 2010. in multi-hop device-to-device networks,” in Information Theory and
[106] H. Meshgi and B. Abolhasani, “High throughput interference-based Applications Workshop (ITA), 2017, pp. 1–7.
routing in multihop CDMA cellular networks,” in 14th International
[128] S. Wang, W. Guo, Z. Zhou, Y. Wu, and X. Chu, “Outage probability
CSI Computer Conference (CSICC), 2009, pp. 206–212.
for multi-hop D2D communications with shortest Path routing,” IEEE
[107] Z. Li and H. Shen, “An efficient routing protocol to increase the
Communications Letters, vol. 19, no. 11, pp. 1997–2000, 2015.
communication reliability of multi-hop cellular networks,” in IEEE
Military Communications Conference (MILCOM), 2008, pp. 1–7. [129] O. Bello and S. Zeadally, “Intelligent device-to-device communication
[108] L. L. Andrew, S. V. Hanly et al., “On routing in CDMA multihop in the internet of things,” IEEE Systems Journal, vol. 10, no. 3, pp.
cellular networks,” in IEEE Global Telecommunications Conference 1172–1182, 2016.
(GLOBECOM), vol. 5, 2004, pp. 3063–3067. [130] R.-C. Marin, R.-I. Ciobanu, and C. Dobre, “Improving opportunistic
[109] A. R. Wolff and C.-C. Lee, “Large scale routing in a multi-hop cellular networks by leveraging device-to-device communication,” IEEE Com-
network using a radial geometric approach,” in IEEE Wireless Commu- munications Magazine, vol. 55, no. 11, pp. 86–91, 2017.
nications and Networking Conference (WCNC), 2007, pp. 4446–4451. [131] C. Ma, Y. Li, H. Yu, X. Gan, X. Wang, Y. Ren, and J. J. Xu,
[110] Y. Park and E.-S. Jung, “Resource-aware routing algorithms for multi- “Cooperative Spectrum Sharing in D2D-Enabled Cellular Networks,”
hop cellular networks,” in International Conference on Multimedia and IEEE Transactions on Communications, vol. 64, no. 10, pp. 4394–4408,
Ubiquitous Engineering (MUE), 2007, pp. 1164–1167. 2016.
[111] S.-H. Lee, E. Choi, and D.-H. Cho, “Energy-efficient routing using [132] F. Mansourkiaie and M. H. Ahmed, “Cooperative routing in wireless
timer-based MAC protocol in power-controlled multihop cellular net- networks: A comprehensive survey,” IEEE Communications Surveys &
works,” in IEEE 60th Vehicular Technology Conference (VTC Fall), Tutorials, vol. 17, no. 2, pp. 604–626, 2015.
vol. 5, 2004, pp. 3201–3205. [133] M. Z. Farooqi, S. M. Tabassum, M. H. Rehmani, and Y. Saleem,
[112] L. Cao, K. Sharif, Y. Wang, and T. Dahlberg, “Adaptive multiple “A survey on network coding: From traditional wireless networks to
metrics routing protocols for heterogeneous multi-hop wireless net- emerging cognitive radio networks,” Journal of Network and Computer
works,” in 5th Consumer Communications and Networking Conference Applications, vol. 46, pp. 166–181, 2014.
(CCNC). IEEE, Jan 2008, pp. 13–17. [134] M. A. Iqbal, B. Dai, B. Huang, A. Hassan, and S. Yu, “Survey of
[113] C. A. St Jean, A. N. Zadeh, and B. Jabbari, “Combined routing, channel network coding-aware routing protocols in wireless networks,” Journal
scheduling, and power control in packet radio ad hoc networks with of Network and Computer Applications, vol. 34, no. 6, pp. 1956–1970,
cellular overlay,” in IEEE 55th Vehicular Technology Conference (VTC 2011.
Spring), vol. 4, 2002, pp. 1960–1964. [135] A. Naeem, M. H. Rehmani, Y. Saleem, I. Rashid, and N. Crespi,
[114] I. Ioannidis and B. Carbunar, “Scalable routing in hybrid cellular and “Network Coding in Cognitive Radio Networks: A Comprehensive
ad-hoc networks,” in IEEE International Conference on Mobile Ad-hoc Survey,” IEEE Communications Surveys & Tutorials, 2017.
and Sensor Systems, 2004, pp. 522–524. [136] M. Youssef, M. Ibrahim, M. Abdelatif, L. Chen, and A. V. Vasilakos,
[115] I. Ioannidis, B. Carbunar, and C. Nita-Rotaru, “High throughput routing “Routing metrics of cognitive radio networks: A survey,” IEEE Com-
in hybrid cellular and ad-hoc networks,” in 6th IEEE International munications Surveys & Tutorials, vol. 16, no. 1, pp. 92–109, 2014.
Symposium on a World of Wireless Mobile and Multimedia Networks [137] J. Zuo, C. Dong, H. V. Nguyen, S. X. Ng, L.-L. Yang, and L. Hanzo,
(WoWMoM). IEEE, 2005, pp. 171–176. “Cross-layer aided energy-efficient opportunistic routing in ad hoc
[116] X. S. J. Riekki, “A Cluster-based Routing Method for D2D Com- networks,” IEEE Transactions on Communications, vol. 62, no. 2, pp.
munication Oriented to Vehicular Networks,” in IEEE International 522–535, 2014.
Conference on Systems, Man, and Cybernetics (SMC), 2017, pp. 2772–
[138] L. Junhai, Y. Danxia, X. Liu, and F. Mingyu, “A survey of multicast
routing protocols for mobile ad-hoc networks,” IEEE communications
[117] Y.-J. Choi, J. H. Jun, and S. Bahk, “Profit-based routing for centralized
surveys & tutorials, vol. 11, no. 1, pp. 78–91, 2009.
multi-hop wireless networks,” in IEEE 59th Vehicular Technology
Conference (VTC Spring), vol. 4, 2004, pp. 2131–2135. [139] F. Hajiaghajani and S. Biswas, “Device-to-device coupon distribution
[118] Y. Wu, K. Yang, and J. Shen, “Source selection routing algorithms in using economic routing utilities,” in IEEE 12th International Confer-
integrated cellular networks,” IET Communications, vol. 2, no. 1, pp. ence on Wireless and Mobile Computing, Networking and Communi-
98–106, 2008. cations (WiMob), 2016, pp. 1–8.
[119] Q. Du, P. Ren, H. Song, Y. Wang, and L. Sun, “On P2P-Share [140] M. Pushpalatha, T. Ramesh, C. Giriraja, and S. K. Konda, “Power
Oriented Routing over Interference-Constrained D2D Networks,” in efficient multicast routing protocol for dynamic intra cluster device
10th International Conference on Mobile Ad-hoc and Sensor Networks to device communication,” in International Conference on Electrical,
(MSN), 2014, pp. 138–143. Electronics, and Optimization Techniques (ICEEOT), 2016, pp. 1853–
[120] S. K. Singh, T. Das, and A. Jukan, “A survey on internet multipath 1856.
routing and provisioning,” IEEE Communications Surveys & Tutorials, [141] S. Sridhara and B. Ramesh, “Energy efficient device to device multicast
vol. 17, no. 4, pp. 2157–2175, 2015. routing for wireless cellular ad-hoc network,” in International Con-
[121] A. Jamalipour, “Proxy discovery and resource allocation for coop- ference on Applied and Theoretical Computing and Communication
erative multipath routing in cellular networks,” in IEEE Wireless Technology (iCATccT), 2015, pp. 708–712.
Communications and Networking Conference (WCNC), 2011, pp. 552– [142] M. Ramalho, “Intra-and inter-domain multicast routing protocols: A
556. survey and taxonomy,” IEEE Communications Surveys & Tutorials,
[122] H. Okada, N. Nakagawa, W. Tadahiro, T. Yamazato, and M. Katayama, vol. 3, no. 1, 2000.
“Multi-route coding in wireless multi-hop networks,” IEICE transac- [143] S. Batabyal and P. Bhaumik, “Mobility models, traces and impact
tions on communications, vol. 89, no. 5, pp. 1620–1626, 2006. of mobility on opportunistic routing algorithms: A survey,” IEEE
[123] H. Okada, H. Imai, T. Yamazato, M. Katayama, and K. Mase, “A route Communications Surveys & Tutorials, vol. 17, no. 3, pp. 1679–1707,
selection scheme for multi-route coding in multihop cellular networks,” 2015.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/COMST.2018.2848108, IEEE
Communications Surveys & Tutorials


[144] M. Madhavan, H. Ganapathy, M. Chetlur, and S. Kalyanaraman,

“Adapting cellular networks to whitespaces spectrum,” IEEE/ACM
Transactions on Networking, vol. 23, no. 2, pp. 383–397, 2015.
[145] Y. Saleem, F. Salim, and M. H. Rehmani, “Routing and channel selec-
tion from cognitive radio networks perspective: A survey,” Computers
& Electrical Engineering, vol. 42, pp. 117–134, 2015.
[146] Y. Saleem, F. Salim, M. H. Rehmani, and B. Rashid, “Routing through
efficient channel assignment in cognitive radio networks,” in Handbook
of Research on Software-Defined and Cognitive Radio Technologies for
Dynamic Spectrum Management. IGI Global, 2015, pp. 275–299.
[147] F. Akhtar, M. H. Rehmani, and M. Reisslein, “White space: Defini-
tional perspectives and their role in exploiting spectrum opportunities,”
Telecommunications Policy, vol. 40, no. 4, pp. 319–331, 2016.
[148] H. Nishiyama, M. Ito, and N. Kato, “Relay-by-smartphone: realizing
multihop device-to-device communications,” IEEE Communications
Magazine, vol. 52, no. 4, pp. 56–65, 2014.
[149] N. Eshraghi, B. Maham, and V. Shah-Mansouri, “Millimeter-wave
device-to-device multi-hop routing for multimedia applications,” in
IEEE International Conference on Communications (ICC), 2016, pp.
[150] J. Kim and A. F. Molisch, “Quality-aware millimeter-wave device-to-
device multi-hop routing for 5G cellular networks,” in IEEE Interna-
tional Conference on Communications (ICC), 2014, pp. 5251–5256.
[151] S. Riolo, D. Panno, and A. Di Maria, “A new centralized access control
for mmWave D2D communications,” in 13th International Conference
on Wireless and Mobile Computing, Networking and Communications
(WiMob), 2017, pp. 1–8.

Farrukh Salim Shaikh is currently a PhD candidate

at Universität Siegen, Germany. He was a lecturer
in NED University, Pakistan from June 2010 to
August 2014. He completed his MS in Computer
Engineering from Hanyang University, South Korea
in 2010. He received his Bachelors of Engineering
(BE) degree in Computer Systems Engineering from
Mehran University of University and Technology
(MUET), Pakistan in 2006. His current research
interests include wireless communications, specif-
ically, D2D networks, cellular networks, cognitive
radio networks, wireless sensor networks and wireless ad-hoc networks.

Roland Wismüller received the Ph.D. degree in

computer science from the Technical University of
Munich in 1994, and the State doctorate in 2001.
He is a full Professor at the Department of Electri-
cal Engineering and Computer Science, University
of Siegen, Germany. His research interests include
energy efficiency in multihop wireless networks,
pattern recognition in network and sensor data, and
security mechanisms for mobile and component-
based systems.

1553-877X (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See for more information.