Content uploaded by Nikolaos Nomikos
Author content
All content in this area was uploaded by Nikolaos Nomikos on Apr 25, 2018
Content may be subject to copyright.
1
Relay Selection for Secure
5G Green Communications
Nikolaos Nomikos, Ana Nieto, Prodromos Makris,
Dimitrios N. Skoutas, Demosthenes Vouyioukas,
Panagiotis Rizomiliotis, Javier Lopez and Charalambos Skianis
Abstract
In this article, we present relay selection policies in applications with secrecy requirements which are
of interest in the fifth generation (5G) of wireless networks. More specifically, we provide a classification
of relays based on their distinct communication attributes, such as processing, multiple antennas, storage,
channel estimation, density and security level. In addition, we discuss the level of efficiency exhibited
by each relay class, regarding their impact in delay-critical applications and green communications
applications, while aiming at a specific security level at the physical layer. Then, relay selection policies
are proposed taking into consideration the goals set by each application. Numerical evaluation of the
proposed policies in terms of the average secrecy rate, average delay and power reduction show improved
performance compared to other state-of-the-art solutions.
Index Terms
5G relays, relay selection, delay-critical, power-efficient communications, security
N. Nomikos , D. N. Skoutas, D. Vouyioukas, P. Rizomiliotis and C. Skianis are with the Department of Informa-
tion and Communication Systems Engineering, University of the Aegean, Karlovassi 83200, Samos, Greece (E-mails:
{nnomikos,d.skoutas,dvouyiou,prizomil,cskianis}@aegean.gr).
A. Nieto and J. Lopez are with the Department of Computer Science, University of Malaga, Campus de Teatinos 29071,
Malaga, Spain (E-mails: {nieto,jlm}@lcc.uma.es).
P. Makris is with the Computer Technology Institute, B’ Building, University Campus 26504, Patras, Greece (E-mail:
pmakri@ceid.upatras.gr).
August 26, 2014 DRAFT
2
I. INTRODUCTION
5G wireless networks aim at the seamless connectivity of all types of devices, networks
and communication protocols fulfilling the goals which have been set by regulatory authorities
and standardization bodies. In this way, the Internet of Things (IoT) can be enabled providing
improved Quality of Service (QoS) [1] compared to current networks. Moreover, increased
spectral and spatial efficiency is harvested through novel networking paradigms involving the
use of unused radio frequencies through cognitive radio [2] and the deployment and association
with relays which allow multi-hop transmissions [3].
Moreover, towards economic and environmental sustainability, green communications is a
major research topic [4]. In this field, relay selection aiming at efficient power usage, leads
to two distinct gains; on the one hand, the carbon footprint reduction of telecommunication
networks leads to environmentally friendly communications and decreased power expenses by
mobile operators and users. On the other hand, reduced public exposure to Electromagnetic Fields
(EMF) is achieved as pursued in forthcoming international research initiatives (e.g. Horizon2020
[5]). Furthermore, satisfying delay requirements without sacrificing the energy efficiency of green
communications poses challenges in developing relay selection policies [6].
These novel characteristics can prove potentially dangerous, if the security threats are not
sufficiently considered during the development of new communication mechanisms. In particular,
these scenarios take advantage of the proliferation of small devices, enabling them to be used
as relays. From the network communication point of view, these devices, used as relays, can
increase the coverage of the network and the convergence of heterogeneous networks. However,
from an attacker’s point of view, potential security holes can be exploited. This assumption is
based on the certainty that small devices are resource constrained, and security mechanisms
cannot be easily applied without severely affecting their operation.
Consequently, in these scenarios the problem is twofold; on the one hand, security and
communication mechanisms have to coexist at the physical layer without making the devices
unusable. That means that relay selection policies have to be aware of the available resources at
the devices, and look for a trade-off between the data transmission and the desired security [7],
[8]. On the other hand, the coexistence of heterogeneous devices becomes a problem when they
are selected without considering their capabilities or security issues (e.g. misbehavior probability).
August 26, 2014 DRAFT
3
Instead, a more reasonable approach is to take advantage of the diversity of relays to enhance,
as far as possible, the quality of the communications and to satisfy the requirements imposed by
the applications that make use of the network. In other words, the suitability of different devices
to enhance security has to be carefully considered as part of the relay selection process.
Regarding the role of relays in the currently deployed fourth generation (4G) networks
many limitations are observed. First, hardware constraints held back the deployment of Full-
Duplex (FD) relaying, as the Loop Interference (LI) between the relay antennas could not be
avoided to a satisfactory degree. Also, mobile relays which provide the potential of ubiquitous
connectivity were not adequately studied. In addition, cognitive relaying and Device-to-Device
(D2D) communications were partly investigated but their implementation was left for the next
generation of wireless networks.
Based on the previous statements, novel relay concepts need to be further exploited in order
to enable 5G green and secure communications. As various types of User Equipments (UEs) and
smart objects could be connected through cellular infrastructure [9], it is of great importance that
the network provides improved access links to support the increased traffic demand. Also, node
density increases and various devices can act as relays to forward traffic from the end-nodes
to the core network and vice versa. These include small cells installed through uncontrolled
network deployment by users to provide improved indoor coverage and UEs resulting in ad-
hoc topologies. In a similar sense, vehicles (either private or public) may act as moving relays
in order to fill coverage gaps in urban settings. Last but not least, fixed relays with increased
processing capabilities are destined to provide improved connectivity to groups of nodes that can
not communicate directly with the base station by employing advanced interference mitigation
techniques [10], [11].
As 5G relaying aims to support a multitude of devices and applications, it is important to adjust
relay functionality on an application basis. Extending the opportunistic relaying paradigm [12],
[13], we propose relay selection policies that take into consideration the type the requirements
of each application. More specifically, the contributions of this work are as follows:
•Two different relay selection policies are proposed for two different applications. More
specifically:
A. A delay minimization policy is given for secure delay-critical applications.
B. A low power consumption policy is proposed for power-efficient communications.
August 26, 2014 DRAFT
4
Additionally, security issues are investigated by incorporating elements of physical-layer
security in these selection policies. As we consider the presence of an eavesdropper which
can overhear all the links, we propose for each application:
i. An optimal secure version which takes into consideration the instantaneous knowledge
of the eavesdropper’s links
ii. A suboptimal version which uses the statistical knowledge of the eavesdropper’s links.
•Numerical results for each selection policy are given employing different types of relays.
The structure of this paper is as follows. Section II presents contributions in the area of green
cooperative and secure cooperative communications. For the green communications, we include
works which study the power-efficiency achieved in communications as well as delay-critical
communications in the context of smart-grids. Then, for secure cooperative communications,
we provide an extensive survey for physical-layer security approaches. In Section III, we give a
detailed description of the various relay classes for 5G wireless networks, as well as the capability
of each class to support specific applications. Next, in Section IV we present the system model
and the necessary preliminaries for the development of the relay selection policies. Subsequently,
the relay selection policies are given in detail in Section V while in Section VI, performance
evaluation and comparisons between relay classes are performed for each selection policy. Finally,
conclusions and future directions are discussed in Section VII.
II. STATE OF THE ART
Here, we provide a detailed state-of-the-art review of works that study aspects related to green
communications and physical-layer security in relay networks.
A. Green cooperative communications
The term “green” has been used in different contexts over the last few years (e.g. computing,
IT) [14] [15], advocating that the use of energy-efficient technologies is fundamental to achieve
a sustainable system. This is motivated by the growing number of devices, and the widespread
concerns about the impact of technology on the environment and health of individuals. In
our study, green communications include two distinct areas; on the one hand, we provide
relevant works regarding power-efficient communication protocols. On the other hand, we present
contributions for delay-critical smart-grid applications that are based on dynamic supply and
August 26, 2014 DRAFT
5
demand of electricity. Such a holistic approach on green communications has been proposed in
[16], where energy-efficiency was not examined only for energy savings of the communication
infrastructure but also as a way to reduce the carbon footprint through smart environments such
as the smart-grids. Moreover, the “green” paradigm includes the cost of the technology.
Several studies agree that the deployment of low-energy consumption mechanisms is essential,
if 5G green communications are becoming a reality [17]. More specifically, the Self-Organized
Networking (SON) of 4G networks has to be extended to enable the next generation of wireless
communication. In [18] smart techniques such as adaptive Base Station (BS) activation according
to data traffic has been proposed. As a result, by introducing the concept of energy-partitions
i.e. associations between activated and deactivated BSs, significant energy savings can be har-
vested. Also, energy-efficiency is a well known concept in resource-constrained networks such
as Wireless Sensor Networks (WSN). Additionally, In Automatic Repeat Request (ARQ) relay
networks, the authors in [19] have increased the energy-efficiency by applying network coding
and improving the QoS.
In general, the biggest improvements regarding power savings aim battery-dependent devices,
because they more than anything else require extended lifetime, especially, small mobile devices.
Indeed, as UEs’ specifications improve, power-consumption becomes a big problem. However,
while users always have the option to periodically charge the battery of their personal devices,
this is not always possible for other battery-dependent devices such as sensors or relays located
in isolated areas (e.g. in rural areas or in forests). As a result, the deployment of energy-
efficient communication mechanisms depends heavily on the purpose or the application driven
by devices [11]. Recently, energy harvesting has been introduced as a promising technique to
extend the lifetime of battery-dependent devices by gathering energy through renewable resources
and electromagnetic waves [20] and in [21] network coding has been combined with energy
harvesting in order to further extend network lifetime.
So, the maximization of the network lifetime, as well as the node lifetime, are discussed widely
in papers related with WSN and other similar networks employing battery-dependent devices
[22] where power allocation is performed based on energy pricing and a more balanced use of the
available relays is achieved. Also, the authors of [23] consider the selection of links which require
the minimum power for successful transmission and the remaining energy of the nodes. In these
studies, the energy consumption, and, therefore, energy-efficient mechanisms, are critical factors,
August 26, 2014 DRAFT
6
and this holds for relay networks, where the relays might be battery-dependent. Furthermore, in
a multi-source environment, the authors of [24] provided a multi-player game theoretic approach
to model the access of the sources and achieve energy-efficient data dissemination. The main
goal of the article is the estimation of equilibrium points in order to optimally handle the trade-
off between energy conservation and data dissemination. In addition, these types of devices are
taking an important role within critical infrastructures, where delay restrictions are usually the
main factor that has to be ensured [25] [14], and also in green cellular networks, where delay-
energy tradeoffs are considered simultaneously . In [26] user association with BSs is based on
energy minimization while policies which activate and de-activate BSs based on their coverage
area and the avoidance of additional signaling overheads.
Power-efficient relay selection proposes power allocation schemes which help to maintain the
energy-efficiency factor under a valid QoS threshold. Fig. 1 shows a summary of the properties
for energy-efficient adaptation according to the current literature. In particular, the maximization
of throughput is a widely discussed issue, as well as power selection based on the Signal-to-
Noise Ratio (SNR) [27] [28]. Some approaches also consider the degrading effect of Additive
white Gaussian Noise (AWGN) in the channel and how it can be mitigated [28]. In other cases,
the minimization of the co-channel interference is an objective which can be achieved through
the maximization of the Signal-to-Interference-plus-Noise Ratio (SINR). The outage probability
is closely related to the co-channel interference [29], and the spectral efficiency also plays an
important role [30]. Moreover, it was demonstrated in [11], that the Inter-Relay Interference
(IRI) can be mitigated with buffer-aided relays and power adaptation, while minimizing the
power expenditure through relay-pair selection. Note that, in general, optimal power adaptation
relies on the availability of Channel State Information (CSI). In addition, the density of the
relays, and, even more so if these relays are personal devices, is the key when defining different
policies to improve the final decisions on the employed power levels [27].
Furthermore the maximization of the capacity is seen as an interesting objective to be achieved
in conjunction with energy-awareness. In this respect, the use of multiple interfaces can be
exploited to improve the communication mechanisms, and opportunistic relaying schemes help
achieve outage-optimality [31]. These techniques can be combined with delay constraints in order
to determine power allocation policies for different types of users [32] and applications such as
smart-grids. In [33], delay bounds that facilitate the design of wireless networks that serve smart-
August 26, 2014 DRAFT
7
Fig. 1. Properties for energy-efficiency in two-hop relay networks.
grids were provided. Furthermore, wireless access and relay schemes were presented that improve
the delay performance when critical messages must be transmitted in the context of the smart-grid
application. Also, the authors of [34] presented relaying strategies that fulfill the requirements of
smart-grids’ data transmission. As relays provide increased diversity, the reliability and spectral
efficiency of smart-grid communications is significantly improved. Cooperative transmission for
meter data collection in a smart-grid was the subject of [35]. A game-theoretic approach was
developed for a scenario consisting of relays that belong to communities that help the data
aggregator unit to transmit smart-grid data to the control center.
B. Secure cooperative communications
Security can be deployed at different layers, using diverse mechanisms. While security-
based cryptosystems are understood to be the best way to ensure confidentiality, data integrity,
August 26, 2014 DRAFT
8
authentication, non-repudiation and other high-layer requirements needed in 5G, at the physical
layer, the integration of such mechanisms can be unaffordable in terms of power consumption,
which is so critical in green computing scenarios. Moreover, it is unrealistic to assume that
all the devices in a heterogeneous network have been provided with security capabilities, or
that there is a willingness to use them. In consequence, cooperative relaying based on security
at the physical layer has emerged as an alternative to cryptographic mechanisms [36]. The
focus is, that there are some physical attacks that can be mitigated based on the selection of
adequate interface parameters. Moreover, as physical-layer security depends on selecting the
communication parameters, the relay that is involved in the communication is enforced to use
these values, in order to be activated.
The most relevant attacks in relay cooperative networking, due to their effect on the com-
munication, are eavesdropping and jamming. First, eavesdroppers can be defined as passive
devices with the sole purpose of collecting data from other participants in the network. So, the
eavesdropping activity affects the confidentiality, privacy, and, therefore, the trustworthiness in
the communication infrastructure. Even worse, eavesdropping is the first step before perpetrating
an active attack. Once the eavesdropper has been able to collect the data he needs in order to
deduce the location of critical devices or sectors in the network, such as the location of the BSs,
he can consider his passive role as insufficient and become an active intruder [37] [38]. If this
happens, the intruder has sufficient knowledge of the behavior of the network to perform several
actions with unpredictable consequences. For example, the attacker may try to impersonate a
relay aiming to drop packets, or to modify the data exchanged between two or more nodes, thus
breaking the integrity of the data. Depending on the infrastructure, these attacks can trigger a
cascade effect with dire consequences. The Denial of Service (DoS) in critical infrastructures
not only threatens business interests, but also human lives.
Therefore, several approaches have focused their efforts on avoiding the eavesdropper based on
selecting the optimal channel and transmit power to improve the secrecy rate [39] [40], defined
as the rate at which data can be transmitted secretly from the source to its intended destination
[36]. The main drawback of these techniques, is that they are usually built on the assumption
that the quality of the eavesdropper’s channel is worse than the channel between the relay and
the source. Therefore, some of these approaches also consider reactive mechanisms to expel the
eavesdropper from the network, like for example, cooperative jamming [41]. These mechanisms
August 26, 2014 DRAFT
9
use artificial Gaussian noise, built under the common knowledge of the Gaussian variance in
order to be able to recover the signal at the end point [36]. For this reason, the Gaussian variance
could be considered as a secret value between the relays. Indeed, some authors consider the use of
Gaussian codebooks in the relays [42]. The disadvantage is that although the Gaussian variance
can be selected taking into consideration power constraints [43], in general, cooperative jamming
requires additional hardware and increases the overhead due to extra-traffic [36]. Furthermore,
depending on the network resources, it may increase the Packet Delivery Ratio (PDR).
Precisely, the PDR and the Packet Sent Ratio (PSR) are two performance metrics that can be
considered for the identification of the possible jammers [44]. A jammer is an active intruder
able to interfere with the signal between two or more devices. The main focus of these types
of attacks is to degrade the communication between the devices, causing DoS. These types of
attacks are widely explained in [44], where the authors provide strategies to detect different types
of jammers in a network, and the effectiveness of PDR and PSR in detecting them. They also
provide some mechanisms to protect the network against jamming, depending on the capabilities
of the nodes. For example, evasive techniques use channel surfing or spatial retreats. The latter
requires the node to be moved, something that is not always possible.
As the last point of discussion, in Fig. 2 a summary of the threats that can be mitigated at
the physical layer based on the current literature is shown. Although the vast majority of the
approaches related to eavesdropping use the secrecy rate as the main metric to avoid eaves-
droppers, there are alternatives, such as the secrecy capacity or the secrecy outage probability.
Nevertheless, both are based on the secrecy rate: the secrecy capacity is the maximum achievable
secrecy rate, while the secrecy outage is defined as the probability that the instantaneous secrecy
capacity is less than a target threshold [45]. In [46] the achievable ergodic secrecy rate, which
is similar to the inverse of the secrecy rate, is considered. In other papers, the eavesdropper is
considered to know the CSI in both hops and relay selection aims to avoid the eavesdropper [42],
[47]. Along these lines, in [47] the authors consider buffer-aided relays with max-link selection
based on instantaneous and statistical CSI.
Finally, it is important to remark, that physical-layer security cannot avoid all the attacks
considered in 5G networks, but it can be very useful in reducing the cost of mitigating some
physical-layer attacks, and it is fundamental to take advantage of this fact, in order to take into
account feasible alternatives to resource-constrained devices without cryptographic capabilities.
August 26, 2014 DRAFT
10
Fig. 2. Security threats and countermeasures in two-hop relay networks.
Thus, in this paper we combine relay selection with physical-layer security in order to avoid the
interception of the transmitted signals by an eavesdropper who can overhear all the links in the
network.
III. RELAY CLASSES AND APPLICATIONS
In this Section, we discuss in detail different relay classes and their potential to support various
applications. Specifically, we introduce a relay classification based on the type of devices, and
we characterize the level that power-efficient and delay-critical applications can be supported by
each class.
A. Relay classification
The classification of 5G relays depends foremost on the type of devices which are deployed.
In Table I, we demonstrate the capabilities of each relay class, ranging from low to high for
various communication attributes.
August 26, 2014 DRAFT
11
Relay Class / Capability Processing MIMO Storage CE Density Trust Confidentiality
User Equipment Low / Med. Low / Med. Low / Med. Low High Low / Med. Low / Med.
Battery-dependent Mobile Relay Med. / High Med. Med. Low / Med. Low / Med. Med. Low / Med.
Battery-dependent Fixed Relay Med. / High High High Med. / High Low / Med. High / Med. Med.
Power-supplied Fixed Relay High High High High Low High High
TABLE I
RELAY CLASSES AND THEIR DISTINCT CAPABILITIES FOR VARIOUS COMMUNICATION ATTRIBUTES.
First, processing capabilities vary significantly, as relay classes range from UEs such as
smartphones and tablets to battery-dependent mobile and fixed relays, as well as power-supplied
fixed relays. Although UEs’ processing power has increased dramatically in recent years, their
battery dependence and small size puts them at a disadvantage against the other relay classes.
In addition, Multiple-Input Multiple-Output (MIMO) capability is expected to be a basic
element in 5G setups. Again, as relay size increases, it becomes easier to install more antennas
on each relay. A subject related to MIMO is the potential for FD relaying, where the same
channel is used to relay the signal in both hops at the expense of LI from the relay’s output
to its input. On the one hand, the small size of UEs prohibits them from employing isolated
antennas, while directional antennas can not be used due to the need for isotropic radiation in
order to maintain connection with BSs. In the same spirit, antenna installation on mobile relays
may face difficulties compared to fixed relays that are easier to configure, as fading conditions
change faster and this reflects on the performance of LI mitigation as well [10].
Recently, the storage capability and buffer management has been investigated in various works
[11], [48]–[51], as buffer-aided relays are able to increase the diversity of the network, as long
as delay can be tolerated. As mobile relays change connections more frequently compared to
fixed relays we consider them to employ medium-sized buffers, as packet expiration increases
due to frequent handovers and disconnections. On the other hand, fixed relays can employ larger
queues as their relative positions to UEs and other devices remain constant for extended periods.
Another critical characteristic for 5G networks is the accurate and fast Channel Estimation
(CE) which provides the means for efficient resource utilization, as the relay’s behavior is adapted
to its environment. As spectrum and channel sensing depend on the quality of the antenna
August 26, 2014 DRAFT
12
sensors and hardware, small-sized UEs are expected to exhibit decreased performance, while
more advanced mobile and fixed devices can provide improved results.
The last characteristic which motivated the use of heterogeneous devices as potential relays is
the increased node density offering spatial diversity. Here, UEs have the advantage compared to
operator-deployed relays and their usage is envisioned in D2D communications. Regarding the
other classes, battery-dependent mobile and fixed relays can be installed more easily compared
to power-supplied relays, and their density is expected to be larger.
Moreover, considering the heterogeneity of relays, trust issues arise. Bearing in mind that the
UEs are personal devices and, thus, can be modified at the software layer, this opens the door to
the personalization of the device, not only by the user, but also by a hacker, who can be seen as
just another user in this type of networks. Unlike UEs, the misbehavior of the relays controlled
by the operators, is most unlikely. Besides, if the relay is power-supplied and fixed these relays
should be located within surveilled areas. Also, security mechanisms in power-supplied relays
can be applied more easily as they are not power-constrained.
Regarding fixed battery-dependent relays, as they have increased degrees of freedom for their
installation it is more difficult for them to be located in well-protected areas e.g. energy-harvesting
relays which are installed in places offering increased energy availability [20], [21]. Additionally,
if the relay is mobile, it is very difficult to maintain reputation systems or any proof about the
behavior of the relays in previous communications. It is well known that, for example, in the case
of Vehicular Ad-Hoc Networks (VANET), there are approaches to mitigate these problems, using
anti-tampering mechanisms, such as the Trusted Platform Module (TPM), in order to provide a
core-of-trust. However, in general, these mechanisms have not yet been widely applied.
Finally, the ability to provide confidentiality depends on the resources of the relay and on
the willingness to use these features. UEs are special cases, because the user might prefer to
customize his device or prolong the battery life as much as possible, at the cost of accepting a
secure platform. Again, relays under the control of operators are more controlled and it is easy
to apply security mechanisms to protect data confidentiality in these cases, more so in the case
of fixed relays.
August 26, 2014 DRAFT
13
Relay Class / Application Performance Security
Delay-critical Power-efficient Delay-critical Power-efficient
User Equipment Low/Medium Medium Low Low
Battery-dependent Mobile Relay Low / Medium Medium Low / Medium Medium
Battery-dependent Fixed Relay Medium Medium / High Low / Medium Low / Medium
Power-supplied Fixed Relay Medium / High High High Medium
TABLE II
PERFORMANCE &SECURITY CAPABILITIES ON APPLICATION BASIS FOR VARIOUS RELAY CLASSES
B. Application performance classification
Table II depicts the level of performance that each relay class can offer to power-efficient and
delay-critical applications. It is obvious that each application may aim at multiple performance
goals that require separate investigation.
In many cases, delay minimization is prioritized e.g. emergency applications, high-precision
industrial applications and various types of smart-grid applications. For a fixed packet size, the
average delay scales with the achieved throughput [52]. To this end, the relay must be capable
of handling delays by increasing the throughput and relays that offer low-latency processing,
full-duplexity and efficient buffer management, have an advantage in such applications. On the
contrary, when the number of relays increases, so does the average delay as the increased density
of the UEs might lead to the distribution of packets in many nodes, thus causing packet expiration
[11], [48], [50], [51]. Similarly, mobile relays offer frequent handovers to their associated devices
and might not offer the same performance as fixed relays. Additionally, the average delay can also
be intentionally increased, due to the participation of misbehaving relays in the communication.
So, the selection of the most trustworthy relays is fundamental in order to avoid intentional delays
in the communication. At the physical-layer, security can be ensured by selecting the relay that
minimizes the delay, according to the performance requirements, while also maximizing the
secrecy rate [42], [45].
Finally, techniques that rely on smart power adaptation are of utmost importance. In this
area, exploiting the increased density of UEs might leverage their lower processing potential
and offer extra channels which can route the signals with more efficient power consumption as
August 26, 2014 DRAFT
14
the offered diversity is greater [11]. It is obvious that advanced nodes with increased MIMO
and CE capabilities might better support power-efficient applications, at the expense of reduced
diversity and costlier deployments. However, once again, the implementation of secure relay
selection is threatened by the existence of misbehaving users. Therefore, the increased density
of UEs also increases the probability of misbehaving. Moreover, when security mechanisms are
widely applied to protect the communications, the overhead increases, and, therefore there is a
final impact on the energy consumption.
IV. SYSTEM DESCRIPTION
In this part, we present in detail the system and the assumptions where we base our study.
More specifically, we describe the system model, its parameters, the relaying strategies that we
employ in this work and consist the basis for the proposed relaying policies and finally, we
provide useful preliminaries regarding secrecy rates for these strategies.
A. System model
We consider a two-hop network where a source Scommunicates with a destination Dvia a
pool of relays C. An eavesdropper Eis located nearby and is able to intercept messages at both
SR and RD links. As we have various classes of relays with diverse characteristics, we provide
their specifications for processing, MIMO (or FD capability), channel estimation and transmit
power.
The UE relay class consists of single-antenna terminals and no FD capability. Moreover, they
use the lowest transmit power equal to 0.2P, where Pis the maximum transmit power that can be
used by a network node. Then, the battery-dependent mobile relays are considered to be MIMO
equipped and FD capable. However, due to the varying fading environment and difficulties in
antenna isolation compared to fixed relays, increased amount of LI affects the FD operation, while
their power output is equal to 0.8Pdue to battery limitations. For the class of battery-dependent
fixed relays, MIMO and FD operation is assumed, as well as an amount of LI comparable to
noise levels due to optimized antenna isolation. Their transmit power is equal to that of mobile
battery-dependent relays. Finally, the power-supplied relay class supports multiple antennas and
has FD characteristics, instantaneous channel estimation and has the highest transmit power P.
August 26, 2014 DRAFT
15
For the transmit power ratios we use the output power values of LTE-Advanced [53] while
lowering the power of battery-dependent nodes in order to justify power supply limitations.
Multiple antenna relays are equipped with N= 2 antennas one for reception and one for
transmission. Each relay Rkcan operate in Half-Duplex (HD) and in FD mode except for the
UEs which operate only in HD. In FD mode, information can be transmitted to and from a
point in two directions simultaneously on the same physical channel [10]. In this work, we not
only consider ideal FD operation, but also LI at the relay. As an alternative to FD operation,
successive relaying is adopted [30], where in each time-slot two relays are activated. In this
strategy, one relay receives the source’s signal while another forwards a previously received
signal to the destination. In this way, FD operation is mimicked, but at the cost of IRI.
All wireless links exhibit fading and we assume an AWGN channel. For simplicity, the variance
of the AWGN is assumed to be normalized with zero mean and unit variance. We also assume
frequency non-selective Rayleigh block fading following a complex Gaussian distribution with
zero mean and variance σ2
ij for the ito jlink and thus, the fading coefficient is constant
during one time slot, but changes independently from one slot to another. The channel gains are
gij ,|hij |2and exponentially distributed. In addition, we consider a clustered relay topology
which offers equivalent average SNR in the SR and RD links (σ2
SRk=σ2
RkD). More specifically,
the relays are positioned relatively close together based on location-based clustering and through
a long-term routing process, variations due to pathloss and shadowing effects are tracked. On the
contrary, the eavesdropper’s average SNR is not the same as that for the i.i.d. clustered relays
(σ2
SE 6=σ2
SRkand σ2
SE 6=σ2
RkD). The power level chosen by the transmitter iis denoted by Pi.
njdenotes the variance of thermal noise at the receiver j, which is assumed to be AWGN and
for simplicity, it is equal to nat all the receivers.
Each relay Rkholds a buffer (data queue) Qkof length Lk(number of data elements) where it
can store source data that has been decoded at the relay and can be forwarded to the destination.
The parameter lk∈Z+,lk∈[0, Lk]denotes the number of data elements measured in bits that
are stored in buffer Qk; at the beginning, each relay buffer is empty (i.e., lk= 0 for all k).
In line with Table I we assume that for UEs Lk= 0.25L, for battery-dependent mobile relays
Lk= 0.5Land for the two types of fixed relays Lk=L.
In Fig. 3, we depict the network topology where each time, a pool of relays supports the SD
communication. The presence of the eavesdropper is illustrated, as well as his ability to overhear
August 26, 2014 DRAFT
16
Power-Efficient
Delay-Critical
UE Relay
Mobile Relay
Fixed Relay
Eavesdropper
Relay Pool
Fig. 3. A network where two different applications take place and end-to-end communication is established via relay selection.
In each application, an eavesdropper is located nearby and overhears the SR and RD links.
all the links between Sand D. Also, we denote the different types of applications that take place.
Each application prioritizes different performance metrics targeting at the same time, specific
secrecy rate goals. More specifically, the delay-critical application prioritizes delay minimization
and the transmitters use fixed powers, while the number of transmitted codewords is matched
to the capacity achieved in each time-slot. Finally, the green communications application uses
power adaptation aiming at a specific rate threshold r0in order to achieve power reduction.
B. Relaying strategies
Next, we analyze the three relaying strategies which form the basis of our work:
•max −link selection [48], where only one link is selected among the available SR and
RD links, resulting in improved diversity. In [11], max −link with power adaptation was
proposed to lower the power expenditure.
•Successive Opportunistic Relaying (SOR), where two selected relays mimic FD operation,
as one receives the source’s signal while the other forwards a previously received signal to
the destination, providing spectral efficiency. Fixed-power SOR has been studied in [51], for
August 26, 2014 DRAFT
17
throughput maximization. Moreover, SOR with power adaptation was the subject of [11],
for power minimization.
•FD opportunistic relaying, where one selected relay transmits and receives on the same
channel using its two antennas, offering throughput enhancement. FD with power adaptation
was studied in [49], to reduce the power expenditure of the network.
As FD and SOR result in FD operation, we refer to these schemes as F D, while max −link
is denoted by ML.
C. Link rates
Here, we provide the rate expressions that are required for the development of the proposed
secure relay selection policies. First, we establish some notation. More specifically, we use r
to denote the rate achieved in a link between a transmitter and a receiver when instantaneous
CSI is available and r†to denote the rate in a link when statistical CSI is known. Similarly, r∗
denotes the secrecy rate based on instantaneous CSI of the eavesdropper’s channels and r†∗ the
secrecy rate based on statistical CSI of the eavesdropper’s channels.
The secrecy rate is defined as the difference between the rate achieved in the link between
a transmitter and the legitimate receiver and the link rate between the transmitter and the
eavesdropper. In our model, a transmission is successful when the secrecy rate r∗(resp. r†∗)
is above a secrecy rate threshold r∗
0(resp. r†∗
0).
The SD link is divided into two links: the link SRibetween the source and a relay Riand
the link between a relay Rjand the destination. The SRirate is given by:
rSRi= log2(1 + γS Ri) = log21 + gS RiPSRi
gRjRiPRjD+n(1)
where gSRi,SRi, and gRjRiare the channel powers of the links between the source and relay
Ri, relay Rjand relay Riand the source. When ML selection is performed, the rate in the SRi
link is equal to
rSRi= log2(1 + γS Ri) = log21 + gS RiPSRi
n(2)
August 26, 2014 DRAFT
18
as there is no IRI or LI in the first hop. Similarly, the RjDrate is given by:
rRjD= log21 + γRjD= log21 + gRjDPRjD
n(3)
We note that in FD relaying i=j, while in SOR i6=j.
In the following subsections, we compute the secrecy rate for a FD scheme when instantaneous
or statistical CSI is available of the eavesdropper’s channels.
1) Instantaneous CSI knowledge: In this case, the instantaneous CSI values γSE and γRjE
are known. The rate of the source-eavesdropper SE link and the Rj-eavesdropper RjElinks is
given by,
rSE = log2(1 + γS E ) = log21 + gSE PSRi
n(4)
and
rRjE= log21 + γRjE= log21 + gRjEPRjD
n(5)
Thus, from the difference between Eq. (1) and Eq. (4), the SRisecrecy rate is given by
r∗
SRi=rSRi−rSE = log2 gRjRiPRjD+n+gSRiPSRin
gRjRiPRjD+n(n+gSE PSRi)!(6)
and
r∗
SRi=rSRi−rSE = log2n+gSRiPSRi
n+gSE PSRi(7)
for ML operation.
Finally, from the difference between Eq. (3) and Eq. (5), the RjDsecrecy rate is given by
r∗
RjD=rRjD−rRjE= log2n+gRjDPRjD
n+gRjEPRjD(8)
where gRjDand gRjEare the channel powers of the Rjto destination and Rjto eavesdropper
links, respectively.
2) Statistical CSI knowledge: For the scenario where only statistical CSI for the SE and RjE
channels is available, the average SNR values (γ†
SE and γ†
ERj) are used in the calculations. More
August 26, 2014 DRAFT
19
precisely, the rates of the eavesdroppers links become,
r†
SE = log21 + γ†
SE = log21 + σ2
SE PSRi
n(9)
and
r†
RjE= log21 + γ†
RjE= log2 1 + σ2
RjEPRjD
n!.(10)
Again, from the difference between Eq. (1) and Eq. (9), the SRisecrecy rate is given by
r†∗
SRi=rSRi−r†
SE = log2 gRjRiPRjD+n+gS RiPSRin
gRjRiPRjD+n(n+σ2
SE PSRi)!(11)
and for ML operation, it becomes
r†∗
SRi=rSRi−r†
SE = log2n+gS RiPSRi
n+σ2
SE PSRi.(12)
Finally, the RD link’s secrecy rate is the rate difference between the RjDand RjElinks, i.e.
Eq. (3) and Eq. (10), it holds that
r†∗
RjD=rRjD−r†
RjE= log2 n+gRjDPRjD
n+σ2
RjEPRjD!.(13)
V. SELECTION POLICIES
In this section, we present the proposed relay selection policies for each application based on
three different CSI assumptions:
i. The CSI of the eavesdropper is not known and thus, relay selection does not consider the
SE and RjElinks. This case is denoted as conventional relay selection.
ii. The instantaneous CSI of the eavesdropper is known and relay selection considers the SE
and RjElinks. This case is denoted as optimal secure relay selection.
iii. The statistical CSI of the eavesdropper is known and relay selection considers the average
channel powers of SE and RjElinks. This case is denoted as suboptimal secure relay
selection.
In the following, we provide the exact details of each policy and we denote by bthe best choice
of the selection policy which is a relay-pair in the case of successive transmission, a relay in
the case of FD transmission or the best link in the case of max −link.
August 26, 2014 DRAFT
20
A. Delay-critical
In many applications, low delay is the main target and relay selection must be adjusted towards
delay minimization. As the average delay scales with the achieved throughput [52] for constant
packet size, relay selection searches for a FD relay or a relay-pair which operates in successive
mode, in order to maximize the end-to-end secrecy rate so that increased secure transmission
rates can be achieved.
1) Conventional relay selection: When the presence of the eavesdropper is not known, relay
selection considers only the links between the source, the Krelays and the destination. This
selection policy is formulated as
bF D = arg max
i,j∈C min{rS Ri, rRjD}= arg max
i,j∈C min{rS Ri, rRjD}(14)
where i=jif one FD relay is selected and i6=jwhen a relay-pair is selected for successive
relaying.
When these modes fail, max −link is used, as it improves the robustness of communication
through increased diversity
bML = arg max
i∈C max{rSRi, rRiD}.(15)
2) Optimal secure relay selection: On the contrary, when the instantaneous CSI knowledge
of the eavesdropper is available, the secure selection policy includes the channels to the eaves-
dropper and the best relay is selected as
b∗
F D = arg max
i,j∈C min{r∗
SRi, r∗
RjD}= arg max
i,j∈C min{r∗
SRi, r∗
RjD}(16)
When FD operation can not be performed, max −link is employed by considering the channels
of the eavesdropper in both hops
b∗
ML = arg max
i∈C max{r∗
SRi, r∗
RiD}.(17)
August 26, 2014 DRAFT
21
3) Suboptimal secure relay selection: The knowledge of the instantaneous CSI for the eaves-
dropper’s channels limits the practicality of relay selection so, we propose the use of statistical
channel knowledge in line with [45], [47], [54]. However, we extend these schemes, as we
propose a hybrid scheme based on three relaying strategies. In contrast to optimal secure relay
selection, in the suboptimal scheme, we use the average SNR expressions for the eavesdropper’s
channels
b†∗
F D = arg max
i,j∈C min{r∗
SRi, r∗
RjD}= arg max
i,j∈C min{r†∗
SRi, r†∗
RjD}(18)
where the corresponding secrecy rates are given by Eqs. (11)-(13).
When FD modes are infeasible, max −link is employed by considering the channels of the
eavesdropper in both hops
b∗
ML = arg max
i∈C max{r†∗
SRi, r†∗
RiD}.(19)
where the corresponding secrecy rates are given by Eqs. (12)-(13).
B. Power-efficient communications
As wireless networks migrate towards power-efficient deployments, power minimization is
targeted. In these cases, power adaptation adjusts the power level of the transmitters. As proposed
in [11], relay selection should take into account the minimum sum of powers required in a time-
slot vif FD or successive relaying takes place. The power levels are calculated to fulfill a secrecy
rate threshold r∗
0set by the application.
1) Conventional relay selection: In this scenario, the CSI of the eavesdropper is not available
and so, relay selection takes into account only the links between the source, the Krelays and
the destination and the rate threshold r0set by the application. As this is the first variation
of the selection policy, we denote the calculated power levels as P1
i. Here, the source and the
transmitting relay are able to perform power adaptation so, the relay which transmits to the
destination defines its power level as
August 26, 2014 DRAFT
22
rRjD≥r0⇒log2 1 + gRjDP1
RjD
n!≥r0⇒
P1
RjD≥n(2r0−1)
gRjD
.(20)
where superscript 1 indicates that the power levels correspond to the traditional relay selection.
In the case of interference at the relay, the source power level P1
SRi, is matched to a rate
threshold r0. Thus, solving for P1
SRiwe have
rSRi≥r0⇒log21 + gSRiP1
Si
gRjRiPRjD+n≥r0
⇒P1
SRi≥gRiRj(2r0−1) P1
RjD+n(2r
0−1)
gSRi
.(21)
For the case of interference-free reception at the relay, the source power level is found by solving
Eq. (21) where the transmitting relay’s power level does not affect the source’s output power
rSRi≥r0⇒log21 + gSRiP1
SRi
n≥r0
⇒P1
SRi≥n(2r
0−1)
gSRi
.(22)
We note that Eq. (22) holds when max −link is adopted.
So, for FD or SOR transmissions selection is defined by
bF D = arg min
i,j∈C(P1
SRi+P1
RjD).(23)
When FD operation is infeasible, max −link selection exhibits power adaptation
bML = arg min
i∈C min{P1
SRi, P 1
RjD}.(24)
We remark here that, power adaptation improves the performance of FD and SOR compared
August 26, 2014 DRAFT
23
to fixed-power schemes, as LI and IRI levels decrease due to the optimal power levels in the
RD links.
2) Optimal secure relay selection: For the case of instantaneous CSI of the eavesdropper’s
channels, the power level calculations are more complex since in both the SR and RD links
we consider their effect in the secrecy rate. Additionally, we must calculate two possible power
levels at the transmitting relay, the power level P2
RjDwhich is required to satisfy the secrecy
rate threshold r∗
0and the power level P1
RjDwhich is needed to satisfy the rate threshold r0as
calculated in the traditional relay selection policy in Eq. (20). We note that the superscript 1
denotes a power level based on conventional relay selection while superscript 2 denotes a power
level based on optimal secure relay selection. First, we calculate the required power of RjD
based on the secrecy rate threshold r∗
0
r∗
RjD≥r∗
0⇒log2
n+gRjDP2
RjD
n+gRjEP2
RjD
≥r∗
0
⇒P2
RjD≥n2r∗
0−1
gRjD−2r∗
0gRjE
,⇔P2
Rj≥n2r∗
0−1
gRjD−2r∗
0gRjE
,(25)
As we require to satisfy both r∗
0and r0, the utilized power level at Rj,Pu
Rjis derived as
Pu
RjD= arg max{P1
RjD, P 2
RjD}(26)
where the superscript udenotes the utilized power level. Then, we calculate Pu
SRifor the case
of FD schemes
r∗
SRi≥r∗
0⇒log2gRjRiPu
RjD+n+gSRiP2
SRin
gRjRiPu
RjD+nn+gSE P2
SRi≥r∗
0
⇒P2
SRi≥gRiRjn−2r∗
0Pu
RjD+n2−2r∗
0gRiRjn−2r∗
0n−2r∗
0n2
2r∗
0gRiRjgSE +gS E Pu
RjD+ 2r∗
0gSE −gSRin(27)
Again, both r∗
0and r0thresholds must be satisfied and the power level at S,Pu
SRiis derived
as
August 26, 2014 DRAFT
24
Pu
SRi= arg max{P1
SRi, P 2
SRi}(28)
When max −link selection is used, the corresponding power level at the source is found as
r∗
SRi≥r∗
0⇒log2
n+gSRiPu
SRi
n+gSE Pu
SRi
≥r∗
0
⇒Pu
SRi≥2r∗
0(n−1)
gSRi−gSE
(29)
Again, for FD or SOR transmissions selection is defined by the minimum sum of powers but,
this time the secrecy rate threshold r∗
0is targeted
bF D = arg min
i,j∈C(Pu
SRi+Pu
RjD).(30)
When FD operation is infeasible, max −link selection exhibits power adaptation while aiming
for r∗
0
bML = arg min
i∈C min{Pu
SRi, P u
RjD}.(31)
where Pu
SRiis derived as
Pu
SRi= arg max{P1
SRi, P 2
SRi},(32)
with P1
SRigiven by Eq. (22).
3) Suboptimal secure relay selection: For the suboptimal secure relay selection policy for
power-efficient communications, we follow a similar analysis as with the instantaneous CSI
selection. So, we provide the power levels which are required to satisfy both the rate threshold
and the secrecy rate threshold
August 26, 2014 DRAFT
25
P†u
RjD= arg max{P1
RjD, P †2
RjD}(33)
where P†2
Rjis selected so as
P†2
RjD≥n2r∗
0−1
gRjD−2r∗
0σ2
RjE
.(34)
Moreover, the source’s power level is chosen as
P†u
SRi= arg max{P1
SRi, P †2
SRi},(35)
where P†2
SRiis defined by the following inequality
P†2
SRi≥gRiRjn+n2−2r∗
0(Pu
RjD+gRiRjn+n+n2)
2r∗
0gRiRjσ2
SE +σ2
SE Pu
RjD+ 2r∗
0σ2
SE −gSRin,(36)
As a result, FD relay selection is performed based on
b†
F D = arg min
i,j∈C(P†u
SRi+P†u
RjD).(37)
Also, max −link selection relies on the statistical CSI of the eavesdropper
b†
ML = arg min
i∈C min{P†u
SRi, P †u
RjD},(38)
where for this case P†u
SRiis derived as
P†u
SRi= arg max{P1
SRi, P †2
SRi},(39)
with P1
SRigiven by Eq. (22) and P†2
SRiby
August 26, 2014 DRAFT
26
P†2
SRi≥2r∗
0(n−1)
gSRi−σ2
SE
(40)
C. Complexity
As we develop relay selection policies aiming to offer improved performance and physical-
layer security, it is critical to calculate the complexity of each policy. In this way, the trade-offs
between improved performance and complexity become clear and it is easier to decide which
policy is chosen based on the relays’ capabilities and the network settings. More specifically, for
both applications CSI at the transmitter is required as in the delay-critical case rate adaptation
is performed by the transmitters, while in the power-efficient communications, power adaptation
is adopted.
Regarding the number of possible relay evaluations that must be calculated each time, assuming
Kavailable relays, for the conventional relay selection in the worst case scenario (in which all
the queues are neither empty nor full), there will be KX(K−1) combinations. Hence, the worst
case complexity of the scheme is O(K2)as the eavesdropper’s links are not known. An identical
complexity is required in the suboptimal case as long as the eavesdropper’s links’ statistics do
not change. Otherwise, a period where the eavesdropper’s CSI i.e. one SE link and K RjE
links must be estimated is needed. As this estimation has to be done in each time-slot (or in the
time period that the fading coefficient does not change) in the optimal selection the complexity
of the optimal policy is O(K3).
VI. NUMERICAL RESULTS
At this point, we provide the performance evaluation of the proposed relay selection policies
for the delay-critical and power-efficient applications. More specifically, we study the secrecy rate
for the delay-critical application and the power reduction for the power-efficient communications.
To examine the impact of the heterogeneous relay classes, three different pools of relays are
compared. The first consists of eight UEs, the second is a mixture of three UEs, two battery-
dependent mobile relays, two battery-dependent fixed relays and one power-supplied fixed relay,
while the last pool provides four Fixed Power-Supplied (FPS) relays, denoted as 8 UE, 8 MIX
and 4 FIX, respectively. The maximum transmit power of each relay class is taken from the
August 26, 2014 DRAFT
27
system model’s description in section IV-A and is given as a transmit SNR value. The buffer
size for a UE relay is set at L= 75, for a battery-dependent mobile relay L= 150 and for the
fixed relays L= 300. To provide comparisons for the proposed policies, we include the max-
ratio secure relay selection of [47], as well as conventional relay selection without considering
the eavesdropper’s CSI.
A. Delay-critical application
In the first set of comparisons, we examine the secrecy rate performance performance of the
delay-critical relay selection policy. We impose a secrecy rate threshold r∗
0= 0.1bps/Hz for a
spectral efficiency r0= 1 bps/Hz and since we use fixed power at the transmitters, the secrecy
threshold scales as the rate increases. The results are depicted in Fig. 4.
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
1.2
1.4
SNR [dB]
Secrecy Rate [bps/Hz]
Optimal 4 FIX
Optimal 8 MIX
Optimal 8 UE
Max−Ratio 4 FIX
Max−Ratio 8 MIX
Max−Ratio 8 UE
Fig. 4. Secrecy rate comparison for various relay classes for the delay-critical application with a secrecy threshold 0.1bps/Hz.
We observe that the proposed optimal relay selection policy outperforms the max-ratio policy
of [47] for all relay pools. In this case, we consider all channels to be i.i.d. including the channels
of the eavesdropper so, the decisive factor to avoid the interception of the signal is the number
of available relays. Also, in the low SNR regime, relays with increased transmission power
August 26, 2014 DRAFT
28
increase the secrecy performance as noise limits the communication. This is evident for the case
of K= 4 FPS relays. However, when the transmit SNR increases, so does the probability of the
eavesdropper to intercept the signal and thus, we see that the heterogeneous pools and the pool
consisting of UEs perform better as they include relays with lower transmit power. Furthermore,
these two pools consist of K= 8 relays and we see that the heterogeneous pool performs better
throughout this comparison as it has all three relay strategies available and combines both the
advantages of UEs, FPS and battery-dependent relays. From the curves, we observe that when
selection relies on the pool of 4 FPS relays, the increased transmit power results in more severe
IRI and LI and the secrecy rate performance decreases after 15 dBs.
Next, in Fig. 5, we impose a more strict secrecy rate threshold r∗
0= 0.2bps/Hz for a spectral
efficiency r0= 1 bps/Hz. The results show that the secrecy rate performance degrades for all
schemes. Still, the proposed optimal policy is superior to max-ratio due to the hybrid combination
of three relay strategies. Also, we see that the K= 4 FPS perform better in the low SNR
regime and then their low number does not allow them to offer the secrecy rate required in this
comparison. The other two cases with K= 4 relay show improved performance after 15 dBs
with the heterogeneous pools having the upper hand. Again, severe performance degradation is
shown by the pool of 4 FPS relays due to the increased amount of interference which is difficult
to avoid due to the smaller number of available relays compared to the other two pools.
In the last comparison for the delay-critical application, the secrecy rate performance of the
conventional, suboptimal and optimal relay selection policies is depicted in Fig. 6 for a secrecy
rate threshold r∗
0= 0.1bps/Hz and a spectral efficiency of r0= 1 bps/Hz. In this case, we have
assumed that the relay channels are i.i.d. but the eavesdropper’s channels are n.i.i.d. and as a result
different variance is assumed in each RjEfading channel. This assumption allows us to capture
the effect of a more realistic environment for all relay selection policies. We see that the optimal
policy exploits the full CSI knowledge to achieve the best secrecy rate performance. Also, due
to n.i.i.d. environment, performance degradation is avoided here by selecting the links which are
less limited by the eavesdropper’s presence. Also, the suboptimal policy provides an interesting
trade-off between CSI reduction at the cost of reduced performance. Finally, conventional relay
selection does not have any knowledge on the eavesdropper’s channels, thus offering the worst
secrecy rate performance.
August 26, 2014 DRAFT
29
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
1.2
1.4
SNR [dB]
Secrecy Rate [bps/Hz]
Optimal 4 FIX
Optimal 8 MIX
Optimal 8 UE
Max−Ratio 4 FIX
Max−Ratio 8 MIX
Max−Ratio 8 UE
Fig. 5. Secrecy rate comparison for various relay classes for the delay-critical application with a secrecy threshold 0.2bps/Hz.
B. Power-efficient communications application
The second set of comparisons examines the performance of transmit power reduction for
different relay pools and CSI knowledge of the eavesdropper’s links. Again, we include results
for the three relay pools (8 UE, 8 MIX and 4 FIX) and the three CSI cases for the 4 FIX pool.
To calculate the power levels, we assume fixed rate transmissions with r0= 1 bps/Hz and we
perform power adaptation according to the CSI of the SR,RD,SE and RE channels using the
power levels of equations (20)–(22) for conventional relay selection, (25)–(27), (29) and (32),
for optimal relay selection and finally, (33)–(36) and (40) for suboptimal relay selection. The
power reduction results are obtained by comparing the utilized power levels to a fixed-power
equivalent scheme when power adaptation is not adopted.
In Fig. 7, results for a secrecy rate threshold r∗
0= 0.1bps/Hz are depicted for the proposed
optimal policy and the max-ratio policy. We observe that the increased diversity provided by the
8 MIX relay pool improves the power reduction performance. Then, the 4 FIX case with power
adaptation follows closely. Although only 4 relays are used in this case, all of them are power-
supplied and thus, they can provide increased transmission power compared to UEs and battery-
August 26, 2014 DRAFT
30
0 5 10 15 20 25 30
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
SNR [dB]
Secrecy Rate [bps/Hz]
Optimal 4 FIX
Suboptimal 4 FIX
Conventional 4 FIX
Fig. 6. Secrecy rate comparison for the three different relay selection policies for the delay-critical application.
dependent relays. So, to calculate the power reduction, we compare with the maximum power
value that can be used by the fixed-power relays each time. We see that these two cases have a
performance gap to their benefit compared to the other four cases. Since, they employ FD and
successive relay strategies, when an FD transmission is performed, power reduction is calculated
by comparing to the fixed-power scheme where the source and the relay use their maximum
transmit powers. On the contrary, the schemes that use the max-ratio policy only, employ one
transmission in each time-slot and, thus, power reduction performance is compromised.
Then, in Fig. 8, results for a secrecy rate threshold r∗
0= 0.2bps/Hz are illustrated. Again, we
see that the 8 MIX and 4 FIX relay pools for the proposed optimal relay selection have the best
performance since FD transmissions provide increased power reduction potential. Then, the 4
FIX max-ratio is distinguished from the rest of the curves as each time, the chosen power level is
compared to an equivalent fixed power transmission for relays that have the capability to use the
maximum available transmit power. The other schemes exhibit an almost identical performance.
Since the proposed 8 UE scheme can not use the FD relay strategy and all the relays have the
lowest maximum power, comparisons with the equivalent fixed-power scheme limits the power
August 26, 2014 DRAFT
31
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
SNR [dB]
Power Redyction [dB]
Optimal 8 MIX
Optimal 4 FIX
Max−Ratio 4 FIX
Optimal 8 UE
Max−Ratio 8 MIX
Max−Ratio 8 UE
Fig. 7. Power reduction comparison for various relay classes for the power-efficient communications application with a secrecy
threshold 0.1bps/Hz.
reduction performance. This holds for the rest two max-ratio schemes 8 UE and 8 MIX, with
the latter having a slightly improved performance due to the use of battery-dependent and FPS
relays in the relay pool.
Finally, in Fig. 9, we compare the three relay selection policies considering different CSI for
the eavesdropper’s channels. The secrecy rate threshold is set at r∗
0= 0.1bps/Hz and while the
relay channels are i.i.d., the eavesdropper’s channels are n.i.i.d. We see that when full CSI for
the eavesdropper is available, optimal relay selection exhibits the best performance as the power
levels are set according to the instantaneous CSI of the eavesdropper’s links. So, power-efficient
and secure transmissions are performed optimally. However, we see that the suboptimal policy
based on the statistical knowledge of the eavesdropper’s links offers a good alternative as it
reflects a more practical case. The worst performance is observed by the conventional relay
selection which lacks the eavesdropper’s CSI.
August 26, 2014 DRAFT
32
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
SNR [dB]
Power Reduction [dB]
Optimal 8 MIX
Optimal 4 FIX
Max−Ratio 4 FIX
Optimal 8 UE
Max−Ratio 8 UE
Max−Ratio 8 MIX
Fig. 8. Power reduction comparison for various relay classes for the power-efficient communications application with a secrecy
threshold 0.2bps/Hz.
VII. CONCLUSIONS AND FUTURE DIRECTIONS
In this survey and performance evaluation paper, we investigated the potential of secure relay
selection in 5G networks supporting green applications. First, we provided a classification of
various types of relays based on their distinct characteristics. Then, we discussed the level
at which each relay type can support different kinds of applications which are of interest
in 5G networks i.e. delay-critical and power-efficient communications under specific security
goals at the physical-layer. Furthermore, relay selection policies were presented which were
linked to specific performance and security metrics set by the aforementioned applications. To
show the efficiency of the proposed policies, numerical results on the secrecy throughput and
power reduction performance for different relay pools were given. The performance evaluation
showed that there is a trade-off between selecting the optimal relay secure transmissions and the
requirement for CSI of the eavesdropper channels. Also, relays’ heterogeneity can be exploited
by selecting each time a transmission strategy (ML, SOR or FD) that achieves the security and
performance requirements of the application.
An interesting extension to this work is the combination of the proposed relay selection
August 26, 2014 DRAFT
33
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
SNR [dB]
Power Reduction [dB]
Optimal 4 FIX
Suboptimal 4 FIX
Conventional 4 FIX
Fig. 9. Power reduction comparison for the three different relay selection policies for the green-communications application.
policies in order to satisfy applications, where more metrics are targeted such as power-efficient
low-delay applications. Also, characteristics such as multi-band relaying, shared deployments by
operators and cognitive relaying can be examined.
REFERENCES
[1] P. Demestichas, A. Georgakopoulos, D. Karvounas, K. Tsagkaris, V. Stavroulaki, J. Lu, Chunshan Xiong and J. Yao, “5G
on the Horizon: Key challenges for the radio-access network,” IEEE Veh. Tech. Mag., vol. 8, no. 3, pp. 47–53, Sept. 2013.
[2] S. Pandit and G. Singh, ”Throughput maximization with reduced data loss rate in cognitive radio network,” Springer
Telecom. Systems, DOI 10.1007/s11235-013-9858-z, Aug. 2013.
[3] P. Mach, R. Bestak and Z. Becvar, ”Optimization of association procedure in WiMAX networks with relay stations,”
Springer Telecom. Systems, vol. 52, no. 3 , pp. 1697–1704, March 2013.
[4] L. Zhou, ”Green service over Internet of Things: a theoretical analysis paradigm,” Springer Telecom. Systems, vol. 52, no.
2 , pp. 1235–1246, Feb. 2013.
[5] “Horizon 2020 – Work programme 2014-2015,” http://ec.europa.eu/research/participants/portal/doc/call/h2020/common/
1587758-05i._ict_wp_2014-2015_en.pdf.
[6] A. Papapostolou, V. Friderikos, T. A. Yahiya and H. Chaouchi, ”Path selection algorithms for fault tolerance in wireless
mesh networks,” Springer Telecom. Systems, vol. 52, no. 4 , pp. 1831-1844, April 2013.
[7] A. Nieto and J. Lopez, “Analysis and taxonomy of security/QoS tradeoff solutions for the future internet,” Wiley Online
Library, Sec.and Commun. Net., In Press.
August 26, 2014 DRAFT
34
[8] L. M. Varalakshmi, G. Florence Sudha and G. Jaikishan, “A selective encryption and energy efficient clustering scheme
for video streaming in wireless sensor networks,” Springer Telecom. Systems, Aug. 2013.
[9] K. Zheng, F. Hu, W. Wang, W. Xiang and M. Dohler, “Radio resource allocation in LTE-advanced cellular networks with
M2M communications,” IEEE Comm. Mag., vol. 50, no. 7, pp. 184–192, July 2012.
[10] T. Riihonen, S. Werner and R. Wichman, “Mitigation of loopback self-interference in full-duplex MIMO relays,” IEEE
Trans. Signal Process. vol. 59, no. 12, pp. 5983–5993, Dec. 2011.
[11] N. Nomikos, T. Charalambous, I. Krikidis, D. N. Skoutas, D. Vouyioukas and M. Johansson, “Buffer-aided successive
opportunistic relaying with inter-relay interference cancellation,” IEEE Pers. Indoor. Mob. Rad. Comm. (PIMRC), London,
United Kingdom, pp. 1321–1325, Sept. 2013.
[12] A. Bletsas, A. Khisti, D. Reed and A. Lippman, “A simple cooperative diversity method based on network path selection,”
IEEE J. Select. Areas Commun., vol. 24, pp. 659–672, March 2006.
[13] D. S. Michalopoulos and G. K. Karagiannidis, “Performance analysis of single relay selection in Rayleigh fading,” IEEE
Trans. Wireless Commun., vol. 7, pp. 3718–3724, Oct. 2008.
[14] V. Kharchenko, A. Gorbenko, V. Sklyar and C. Phillips, “Green computing and communications in critical application
domains: Challenges and solutions,” IEEE Dig. Tech. (DT), pp. 191–197, 2013.
[15] T. El Maliki and J.M. Seigneur, “Security adaptation based on autonomic and trust systems for ubiquitous mobile network
and green IT,” Conf. on Mobile Ubiq. Comp., Syst., Serv. and Techn. (UBICOMM), Porto, Portugal, pp. 152–158, Sept.
2013.
[16] S. Aleksic, “Energy-efficient communication networks for improved global energy productivity,”, Springer Telecom.
Systems, vol. 54, pp. 183-199, Oct. 2013.
[17] M. Olsson, C. Cavdar, P. Frenger, S. Tombaz, D. Sabella and R. Jantti, “5GrEEn: Towards green 5G mobile networks. In
wireless and mobile computing,” IEEE Networking and Communications (WiMob), pp. 212–216, Oct. 2013.
[18] K. Samdanis, D. Kutscher and M. Brunner, “Self-organized energy efficient cellular networks,” IEEE Pers. Indoor. Mob.
Rad. Comm. (PIMRC), Istanbul, Turkey, pp.1665–1670, Sept. 2010.
[19] A. Antonopoulos, C. Verikoukis, C. Skianis and O. Akan, ”Energy efficient network coding-based MAC for cooperative
ARQ wireless networks,“ Elsevier Ad Hoc Networks, vol. 11, no. 1, pp. 190–200, Jan. 2013.
[20] I. Krikidis, T. Charalambous and J.S Thompson, “Stability analysis and power optimization for energy harvesting
cooperative networks,” IEEE Sign. Proc. Letters, vol. 19, no. 1, pp. 20–23, Jan. 2012.
[21] P. V. Mekikis, A. S. Lalos, A. Antonopoulos, L. Alonso and C. Verikoukis, C., “Wireless energy harvesting in two-way
network coded cooperative communications: A stochastic approach for large scale networks,” IEEE Comm. Letters, vol.
18, no. 6, pp. 1011–1014, June 2014.
[22] F. Ke, S. Fenf and H. Zhuang, “Relay selection and power allocation for cooperative network based on energy pricing,”
IEEE Comm. Letters, vol. 14, no. 5, pp. 396–398, May 2010.
[23] B. Panigrahi, B. S. Panda and J. D. Luk, “Energy-efficient greedy forwarding protocol for wireless sensor networks,” IEEE
Vehic. Techn. Conf. (VTC), Taipei, Taiwan, pp. 1–5, May 2010.
[24] A. Antonopoulos and C. Verikoukis, “Multi-player game theoretic MAC strategies for energy efficient data dissemination,”
IEEE Trans. Wireless Commun., vol. 13, no. 2, pp. 592–603, Feb. 2014.
[25] C. H. Hauser, D. E. Bakken, I. Dionysiou, K. H. Gjermundrod, V. Irava, J. Helkey and A. Bose, “Security, trust, and QoS
in next-generation control and communication for large power systems,” Critical Infrastructures, vol. 4, no. 1, pp. 3–16,
2008.
August 26, 2014 DRAFT
35
[26] K. Son, H. Kim, Y. Yi and B. Krishnamachari,“Base station operation and user association mechanisms for energy-delay
tradeoffs in green cellular networks,” IEEE Sel. Areas in Commun., vol. 29, no. 8, pp. 1525–1536, Sept. 2011.
[27] C. Y. Ho and C. Y. Huang, “Energy-efficient subcarrier power allocation and relay selection scheme for OFDMA-based
cooperative relay networks,” IEEE Int. Conf. Comm. (ICC), Kyoto, Japan, pp. 1–6, June 2011.
[28] S. S. Ikki, P. Ubaidulla and S. Aissa, “Regenerative cooperative diversity networks with co-channel interference:
performance analysis and optimal energy allocation,” IEEE Trans. Vehicular Tech., vol. 62, no. 2, pp. 896–902, Feb.
2013.
[29] H. Yu, I. H. Lee and G. L. Stuber, “Outage probability of decode-and-forward cooperative relaying systems with co-channel
interference,” IEEE Trans. Wireless Commun., vol. 11, no. 1, pp. 266–274, Jan. 2012.
[30] B. Rankov and A. Wittneben, “Spectral efficient protocols for half-duplex fading relay channels,” IEEE Sel. Areas in
Commun., vol. 25, no. 2, pp. 379–389, Feb. 2007.
[31] N. Nomikos, D. N. Skoutas, D. Vouyioukas, C. Verikoukis and C. Skianis, “Capacity maximization through energy-aware
multi-mode relaying,” Springer Wireless Personal Comm., vol. 74, no. 1, pp. 83–99, Jan. 2014.
[32] L. Musavian, S. Aïssa and S. Lambotharan, “Effective capacity for interference and delay constrained cognitive radio
relay channels,” IEEE Trans. Wireless Commun., vol. 9, no. 5, pp. 1698–1707, May 2010.
[33] Yi Xu and W. Wang, “Wireless mesh network in smart grid: Modeling and analysis for time critical communications,”
IEEE Trans. Wireless Commun., vol. 12, no. 7, pp.3360–3371, July 2013.
[34] H. Sun, A. Nallanathan, Bo Tan, J. S. Thompson, J. Jiang and H. V. Poor, “Relaying technologies for smart grid
communications,” IEEE Wireless Commun., vol. 19, no. 6, pp.52–59, Dec. 2012.
[35] D. Niyato and P. Wang, “Cooperative transmission for meter data collection in smart grid,” IEEE Commun. Mag., vol. 50,
no. 4, pp. 90–97, April 2012.
[36] L. Dong, H. Yousefi’zadeh and H. Jafarkhani, “Cooperative jamming and power allocation for wireless relay networks in
presence of eavesdropper,” IEEE Int. Conf. Comm. (ICC), Kyoto, Japan, pp. 1–5, June 2011.
[37] J. Kong, X. Hong and M. Gerla, “A new set of passive routing attacks in mobile ad hoc networks,” IEEE Mil. Commun.
Conf. (MILCOM), Boston, USA, vol. 2, pp. 796–801, Oct. 2003.
[38] H. Deng, W. Li and D. P Agrawal, “Routing security in wireless ad hoc networks,” IEEE Commun. Mag., vol. 40, no. 10,
pp. 70–75, Oct. 2002.
[39] H. Khodakarami and F. Lahouti, “Link Adaptation with Untrusted Relay Assignment: Design and Performance Analysis,”
Arxiv, Nov. 2012.
[40] Y. Zou, X. Wang and W. Shen “Optimal relay selection for physical-layer security in cooperative wireless networks,” IEEE
J. Select. Areas Commun., vol. 31, no. 10, pp. 2099–2111, Oct. 2013.
[41] J. Huang, A. Mukherjee and A. L. Swindlehurst, “Secure communication via an untrusted non-regenerative relay in fading
channels,” IEEE Trans. Signal Process., vol. 61, no. 10, pp.2536–2550, April 2013.
[42] R. Bassily and S. Ulukus “Deaf cooperation and relay selection strategies for secure communication in multiple relay
networks,” IEEE Trans. Signal Process., vol. 61, no. 6, pp. 1544–1554, Feb. 2013.
[43] J. Li, A.P. Petropulu and S. Weber “On cooperative relaying schemes for wireless physical layer security,” IEEE Trans.
Signal Process., vol. 59, no. 10, pp. 4985–4997, Sept. 2011.
[44] W. Xu, K. Ma, W. Trappe and Y. Zhang, “Jamming sensor networks: attack and defense strategies,” IEEE Networks, vol.
20, no. 3, pp. 41–47, June 2006.
August 26, 2014 DRAFT
36
[45] I. Krikidis, “Opportunistic relay selection for cooperative networks with secrecy constraints,” IET Communications, vol.
4, no. 15, pp. 1787–1791, Oct. 2010.
[46] Z. Ding, K.K. Leung, D.L. Goeckel and Towsley, “Opportunistic relaying for secrecy communications: Cooperative jamming
vs. relay chatting,” IEEE Trans. Wireless Commun., vol. 10, no. 6, pp. 1725–1729, June 2011.
[47] G. Chen, Z. Tian, Y. Gong, Z. Chen and J. Chambers, “Max-ratio relay selection in secure buffer-aided cooperative wireless
networks,” IEEE Trans. Inf. Forens. Secur., vol. 9, no. 4, pp. 719–729, April 20142.
[48] I. Krikidis, T. Charalambous and J. S. Thompson, “Buffer-aided relay selection for cooperative diversity systems without
delay constraints,” IEEE Trans. Wireless Commun., vol. 11, pp. 1957–1967, May 2012.
[49] N. Nomikos, T. Charalambous, I. Krikidis, D. Vouyioukas and M. Johansson, “Hybrid cooperation through full-duplex
opportunistic relaying and max-link relay selection with transmit power adaptation,” IEEE Int. Conf. Comm. (ICC), Sydney,
Australia, June 2014.
[50] A. Ikhlef, D. S. Michalopoulos and R. Schober, “Max-max relay selection for relays with buffers,” IEEE Trans. Wireless
Commun., vol. 11, pp. 1124–1135, March 2012.
[51] N. Nomikos, D. Vouyioukas, T. Charalambous, I. Krikidis, P. Makris, D. N. Skoutas, M. Johansson and C. Skianis, “Joint
relay-pair selection for buffer-aided successive opportunistic relaying,” Wiley-Blackwell Trans. Emerg. Telecom. Techn.,
DOI: 10.1002/ett.2718, Nov. 2013.
[52] A. El Gamal, J. Mammen, B. Prabhakar and D. Shah, “Optimal throughput-delay scaling in wireless networks – Part II:
Constant-size packets,” IEEE Trans. on Inf. Theory, vol. 52, no. 11, pp. 5111–5116, Nov. 2006.
[53] 3GPP TR 36.814 v.9.0.0 “Technical specification group radio access network; evolved universal terrestrial radio access
(E-UTRA); Further advancements for E-UTRA physical layer aspects,” March 2010.
[54] I. Krikidis, J.S. Thompson and S. McLaughlin “Relay selection for secure cooperative networks with jamming,” IEEE
Trans. Wireless Commun., vol. 8, no. 10, pp. 5003–5011, Oct. 2009.
August 26, 2014 DRAFT
