INVITED

PAPER

Physical-Layer Security in the

Internet of Things: Sensing and
Communication Confidentiality
Under Resource Constraints
This paper offers a comprehensive survey of present advances and remaining
challenges in resource constrained secrecy coding and secret-key generation suitable
for applications in the Internet of Things.
By Amitav Mukherjee, Member IEEE

ABSTRACT | The Internet of Things (IoT) will feature pervasive KEYWORDS | Internet of Things (IoT); low complexity; physical-
sensing and control capabilities via a massive deployment of layer security; resource constraints; secure sensing
machine-type communication (MTC) devices. The limited hard-
ware, low-complexity, and severe energy constraints of MTC
devices present unique communication and security chal- No me nc la tu re
lenges. As a result, robust physical-layer security methods 3G Third generation (cellular).
that can supplement or even replace lightweight cryptographic 3GPP Third Generation Partnership Project.
protocols are appealing solutions. In this paper, we present an 5G Fifth generation (cellular).
overview of low-complexity physical-layer security schemes AN Artificial noise.
that are suitable for the IoT. A local IoT deployment is modeled BER Bit error rate.
as a composition of multiple sensor and data subnetworks, with CAE Channel-aware encryption.
uplink communications from sensors to controllers, and CS Compressive sensing.
downlink communications from controllers to actuators. The CSI Channel state information.
state of the art in physical-layer security for sensor networks is CSIT Channel state information at the transmitter.
reviewed, followed by an overview of communication network D2D Device-to-device.
security techniques. We then pinpoint the most energy-efficient EFC Eavesdropping fusion center.
and low-complexity security techniques that are best suited for GEVD Generalized eigenvalue decomposition.
IoT sensing applications. This is followed by a discussion of GSVD Generalized singular value decomposition.
candidate low-complexity schemes for communication secu- HSPA High speed packet access (cellular).
rity, such as ON–OFF switching and space-time block codes. The IoT Internet of Things.
paper concludes by discussing open research issues and LFC Legitimate fusion center.
avenues for further work, especially the need for a theoretically LLR Log-likelihood ratio.
well-founded and holistic approach for incorporating com- LRT Likelihood ratio test.
plexity constraints in physical-layer security designs. LTE-A Long-term evolution-advanced.
MIMO Multiple-input–multiple-output.
MISO Multiple-input–single-output.
ML Maximum likelihood.
Manuscript received January 27, 2015; revised June 23, 2015; accepted August 2, 2015. MSE Mean squared error.
Date of publication September 11, 2015; date of current version September 16, 2015.
The author is with the Radio Access Technologies Group, Ericsson Research, MTC Machine-type communication.
San Jose, CA 95134 USA (e-mail: amitav.mukherjee@ericsson.com). NFC Near-field communication.
Digital Object Identifier: 10.1109/JPROC.2015.2466548 OFDMA Orthogonal frequency-division multiple access.
0018-9219  2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1747
Mukherjee: Physical-Layer Security in the Internet of Things
RAT Radio access technology.
RFID Radio-frequency identification.
RSSI Received signal strength indicator.
SINR Signal-to-interference-plus-noise ratio.
SISO Single-input–single-output.
TBMA Type-based multiple access.
VLC Visible light communications.
I. INTRODUCTION
The notion of an IoT is an inevitable offshoot of the ongo-
ing advances in communication technology and network-
ing services. The IoT is expected to provide ubiquitous
connectivity and information-gathering capabilities span-
ning home, vehicular, and industrial environments. It is
envisioned that billions of physical devices will be outfitted
with different kinds of sensors and actuators and con-
nected to the Internet via heterogeneous access networks
[1]. Thus, the main driver behind IoT will be the large-
scale deployment of MTC devices or machine-to-machine
links that perform sensing and actuation tasks with mini-
mal human intervention [2]–[4].
IoT is synonymous with the notion of a cyber–physical
system, or a ‘‘network of networks’’ [5]. The scope of the
IoT extends far beyond autonomously adapting air condi-
tioning and heating levels in a smart home or reporting
smart meter readings. Assistive technologies for the elderly
and disabled, in vivo and in situ biological monitoring,
adaptive irrigation for agriculture, self-driving vehicles,
emergency and disaster response, and many other such
applications will be enabled and facilitated by the IoT. Such
pervasive sensing and control capabilities will lead to a
transformative change in daily life. The IoT is expected to
be a centerpiece of upcoming 5G communication tech-
nologies that will be commercially deployed from 2020
onwards [6], [7].
An IoT infrastructure is rendered operational by a
communication network that collects and exchanges useful
information to fully leverage the advantages of IoT. The air
interface used for wireless connectivity may range from
NFC, Bluetooth, GSM, HSPA, IEEE 802.15.4 (e.g.,
ZigBee), IEEE 802.11ah, 3GPP LTE-A, or a proprietary
system (e.g., Weightless, SigFox, On-Ramp). It is possible
that multiple such RATs are present within the same local
IoT deployment. As the IoT evolves, future connectivity
solutions may encompass VLC [8], acoustic [9], and mole-
cular communications [10].
Communication security is clearly a requirement for
IoT applications, given their wide scope encompassing
commercial, industrial, governmental, and military appli-
cations [11], [12]. Traditional cryptographic protocols [13],
[14] that require key distribution or certificate manage-
ment can be challenging to implement in IoT systems with
a very large number of MTC devices, coupled with hetero-
geneous RATs and different subsystems being controlled
by distinct operators. As a result, robust physical-layer
1748 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
security methods that can supplement lightweight crypto-
graphic protocols [15]–[17] are appealing solutions for the
IoT. Roughly speaking, these methods exploit differences
in channel conditions and interference environments to
boost signal reception at intended recipients while degrad-
ing the reception of unauthorized users. Physical-layer se-
curity methods are generally agnostic to the RAT, and offer
‘‘built-in’’ security that is information-theoretically un-
breakable [18], [19].
However, not all existing physical-layer security tech-
niques are suitable for IoT applications. This is because
MTC devices have some unique characteristics compared
to smartphones and tablets that are powered by 3G and
LTE-A broadband networks today. They generally have low
data rate requirements, periodic data traffic arrivals, li-
mited hardware and signal processing capabilities, limited
storage memory, compact form factors, and significant
energy constraints [20]. As an example, a battery life of ten
years at a quarter of the cost of wideband LTE-A devices is
one of the objectives of the Release 13 LTE-A MTC stan-
dardization effort [21]. Thus, the complexity, energy effi-
ciency, and CSI requirements of the chosen security
methods are critical aspects that determine their feasibility
in the IoT [23]. These aspects have received relatively li-
mited attention in the literature on physical-layer security.
The remainder of this work is organized as follows.
Section II introduces simple mathematical models for up-
link and downlink communications in an IoT environment.
The state of the art in physical-layer security for sensor
networks and data communications is reviewed in succes-
sion in Section III, along with the challenges unique to IoT
security. Secure transmission methods tailored for resource-
constrained IoT sensors and actuators are analyzed in detail
in Sections IV and V, respectively. Finally, conclusions and
future directions are discussed in Section VI.
Furthermore, the notation used hereafter signifies the
following. Lowercase boldface letters denote vectors, up-
percase boldface letters represent matrices, and C is the
complex domain. We will use N ð0; ZÞ to denote a circular
symmetric complex multivariate Gaussian distribution
with zero mean and covariance matrix Z. We also use E
to denote expectation, HðÞ for entropy, ðÞT for the trans-
pose, ðÞH for the complex conjugate (Hermitian) trans-
pose, k  k for the vector norm, detðÞ to denote the matrix
determinant, and I represents an identity matrix of ap-
propriate dimension.
I I. IoT T AXONOMY AND MODEL
We consider an abstraction of an IoT system with four
categories of components.
• Sensors: They monitor some phenomenon and
report their observations to a data collection point.
The data collection is performed by a controller.
The majority of the traffic flow is from the sensors
to the collection point (uplink).
 Mukherjee: Physical-Layer Security in the Internet of Things

Fig. 1. Conceptual visualization of an IoT realization with heterogeneous wireless communication links. Each subsystem has an associated
controller that controls transmission modes and scheduling.

• Actuators: They receive directions from a control-
ler and perform corresponding actions, either phy-
sical or electronic. The majority of the traffic flow
is from the controller to the actuators (downlink).
• Controllers: They perform scheduling of a subset
of IoT devices (sensors and actuators), and trans-
mit or receive data to them.
• Eavesdroppers: They are unauthorized receivers
that seek to passively intercept communications
between controllers and sensors or actuators.
Eavesdroppers are modeled as being passive, i.e., they do
not inject false data or jamming signals into the system.
Similar to worst case assumptions in the literature, eaves-
droppers are assumed to have unbounded computational
power, which implies that they are not a compromised
sensor and are external to the IoT network. We also as-
sume that communications are hierarchical. In the hierar-
chical IoT model, only controllers can communicate with
spectively. Thus, n~k > 1 would represent a multiple access
channel with multiple sensors transmitting on the same
time–frequency resources to the controller, while n~k ¼ 1 if
orthogonal resources (in time or frequency) are available
to form parallel access channels. Similarly, a~k ¼ 1 models a
downlink system where separate time or frequency re-
sources are allocated to different actuators, for, e.g., an
OFDMA system with one actuator served on a particular
set of frequency tones. An example is depicted in Fig. 2.
For purposes of illustration, assume all sensors are
equipped with a single antenna. The sensors observe their
environment and compute a digitally modulated informa-
tion signal to be sent to the LFC. The baseband signals
received from the sensors at their controller for an arbi-
trary channel use can be represented as
X
n~k

controllers of adjacent subsystems and the external wide yc;k ¼ hi;k xi;k þ wk (1)
i¼1
area network. A subsystem is a localized group of nodes
tasked with a common objective, such as video surveillance
or factory automation, as shown in Fig. 1. Thus, components
of a subsystem cannot directly communicate with those of
another subsystem, due to the likely usage of different RATs
or frequency channels. For this reason, we focus on
infrastructure-based communication architectures in this
work, and do not explicitly consider security challenges in
D2D communication scenarios [24]. However, many of the
challenges and potential solutions described herein also
apply to D2D communications [25]–[28].
Consider an arbitrary IoT subsystem k comprising nk
sensors and ak actuators, with the corresponding controller
equipped with mk antennas. The parameters nk and ak are
the number of active nodes at a particular instant. While
the number of deployed devices may be much larger, many
of them are likely to be in sleep mode for energy conserva-
tion [20]. Furthermore, let n~k and a~k denote the number of
sensors and actuators that transmit or receive data at a Fig. 2. IoT sensor network with SISO sensors and multiantenna EFC
particular instant, with 1  n~k  nk and 1  a~k  ak , re- and LFC.

Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1749
Mukherjee: Physical-Layer Security in the Internet of Things

where yc;k 2 Cmk 1 is observed by the controller, hi;k 2

Cmk 1 is the vector of complex fading channel coefficients
from sensor i to the controller, xi;k is the information
symbol transmitted by sensor i, and wk  N ð0; IÞ is zero-
mean circular symmetric complex Gaussian noise. The jth
element of the channel vector hi;k represents the wireless
channel coefficient between antenna j of the controller
and sensor i. Each sensor obeys an average transmit power
constraint of Ejxi;k j2  Ps . The above model is easily gener-
alized to multiantenna sensors, however, as discussed be-
fore, hardware complexity and energy consumption are
minimized when IoT sensors have a single RF chain.
Assuming a single re -antenna eavesdropper for simpli-
city in the vicinity of this controller, its intercepted signal
can be written as
Fig. 3. IoT controller-to-actuators communications overheard
by an EFC.
X
n~k
ye;k ¼ gi;k xi;k þ zk (2)
i¼1
I II . STATE OF THE ART IN PHY SECURITY
where gi;k 2 Cre 1 is the vector of complex fading channel
A. Background
coefficients from sensor i to the eavesdropper, and
The two main categories of physical-layer security tech-
zk  N ð0; IÞ is zero-mean circular symmetric complex
niques are: 1) those based on intelligent design of transmit
Gaussian noise.
coding strategies that do not need a secret key; and 2) those
The general downlink signal from the controller to ac-
that exploit the wireless communication medium to dev-
tuator j can be represented as
elop secret keys over public channels [29]–[32]. In this
work, we will mostly focus on the first category, namely,
X
a~k keyless secrecy methods, since they do not require error-
~ j tj;k þ
ya;j ¼ h ~ j td;k þ w
h ~j (3) free two-way public channels and are more easily extended
d6¼j to large-scale multiuser networks.
Security aspects of sensor and communications net-
works will be considered separately in subsequent parts of
where h ~ i;k 2 C1mk is the vector of complex fading channel
the paper. Most distributed detection problems in sensor
coefficients from the controller to actuator j, ti;k 2 Cmk 1 networks involve a fusion center trying to infer the state of
is the information signal vector intended for actuator j, and a target phenomenon based on reports from noncolocated
~j  N ð0; 1Þ is zero-mean circular symmetric complex
w sensors [33], [34]. Security issues arise when an LFC would
Gaussian noise. The signal for an arbitrary actuator does like to prevent an EFC from performing the same inference
not need to be kept confidential from the ð~ ak  1Þ other [35]. A binary hypothesis test is the simplest such case,
actuators. A special case of (3) would be a multicast where the unknown state H belongs to one of two possible
scenario where a single, common information signal vector hypotheses:  2 f0 ; 1 g. The LFC performs data fusion if
tk is to be received by all a~k actuators. the sensors report their (quantized) observations, or
The corresponding received signal at the eavesdropper is decision fusion if the sensors report their individual
decisions (for, e.g., 1-bit feedback to indicate the local
X
ak state estimate). In the case of data fusion, the LFC generally
~ e;k ¼
y ~ e;k ti;k þ ~
G zk (4) performs a likelihood ratio test to arrive at its overall state
i¼1 estimate H.^ The LFC detection probability and probability
of false alarm are then defined as Pd ¼ PrðH^ ¼ H1 jH1 Þ and
~ e;k 2 Cre mk is the matrix of complex fading chan- Pf ¼ PrðH^ ¼ H1 jH0 Þ, respectively. In the case of distributed
where G
detection, security can be quantified using a divergence
nel coefficients from controller to eavesdropper, z ~k 
criterion1 [36], conditional eavesdropper entropy [38], the
N ð0; IÞ is zero-mean circular symmetric complex Gauss-
1
ian noise, and it is assumed that the eavesdropper would As an example, the Kullback–Liebler divergence of Rtwo continuous
like to decode all ak signals. All channels and noise vectors probability densities p and q is defined as DðqjjpÞ ¼ qðxÞ logðqðxÞ=
pðxÞÞdx, which is also the relative entropy. It is desirable to maximize the
are mutually independent across (1)–(4). A pictorial over- divergence of the LFC, while a divergence of zero at the EFC implies that
view is shown in Fig. 3. its decision is equivalent to flipping a coin.

1750 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
 Mukherjee: Physical-Layer Security in the Internet of Things

relative detection probabilities [40], or relative error processing speed, storage capacity, and energy resources in
probabilities [49] at the EFC and LFC. In one definition, sensor networks [22], asymmetric cryptography such as
information-theoretic perfect secrecy is said to be achieved if the Rivest–Shamir–Adleman algorithm or the Diffie–
the conditional entropy at the EFC after its observations is Hellman key agreement protocol is often considered too
the same as the a priori entropy demanding in terms of processing power [23]. In recogni-
tion of this, Kundur et al. examined cross-layer security
methodologies based on lightweight symmetric key cryp-
~ e;k Þ ¼ HðÞ
HðjY (5) tography for multimedia sensor networks [37]. However,
their approach requires that each network node have an
~ e;k are the random variable counterparts of individual key that is shared with the LFC and that pair-
where  and Y
wise keys be available between adjacent sensors in a sub-
state  and EFC signal y ~ e;k , respectively. Satisfying the
system. In addition, every network entity has two
above condition is also known as maximizing the equivo-
predeployed network-wide keys. The burden of secure
cation of the EFC.
key exchange and management are not negligible even in
The physical-layer security of data communication in a
lightweight key-based systems when the number of nodes
wiretap channel with a single legitimate link and one or
can be very large as in the IoT. This motivates the use of
more eavesdroppers is usually quantified by the secrecy
physical-layer security methods that are reviewed next.
rate. Other metrics for communication security include
the relative SINRs, relative BER, and relative MSE of the
1) Censoring: Marano et al. [36] examined optimal sensor
authorized and eavesdropping receivers.
censoring strategies in an energy-constrained sensor net-
In the wiretap channel, the secrecy rate is a transmis-
work where the EFC has a degraded channel that can only
sion rate that can be reliably supported on the primary
distinguish if sensor transmissions are present or absent.
channel, but which is undecodable on the eavesdropper’s
Censoring entails comparing a sensor’s local LLR to a pair
channel.2 For Gaussian channels, it is calculated as the
of thresholds, and transmitting to the LFC only if the ratio
difference between the mutual information on the primary
is either very high or very low. The EFC therefore does not
and eavesdropper’s channels. For example, the secrecy rate
have access to the sensors’ transmitted data, but can moni-
obtained from (3) and (4) with a single actuator ðak ¼ 1Þ
tor the transmission activity of the channel and exploit the
and fixed channel states is
busy-idle state of the channel for detecting the hypothesis.
h   2) Channel-Based Bit Flipping: Now consider the case
~ jh
Rs ¼ log2 1 þ Pc;k h ~H
j where the EFC is able to directly overhear the actual sensor
 iþ n~k
~ e;k t1;k tH G~H transmissions fxi;k gi¼1 . In [38], channel fading gains are
 log2 det I þ Pc;k G 1;k e;k (6)
used in a secure TBMA scheme where the sensors follow
different reporting rules depending on the strength of their
channel gains to the LFC. The LFC announces two thresh-
where ½xþ ¼ maxf0; xg. Secrecy capacity is achieved
olds s and w to the sensors, which then compare their
when the secrecy rate is maximized with respect to input
channel amplitudes against these thresholds and autono-
distribution and power allocation.
mously classify themselves into strong and weak sets. Sen-
In fast fading channels, the secrecy outage probability
sors from each set then randomly decide to transmit their
is another metric of interest that represents the probability
quantized measurements; the activation probabilities are
that a certain target secrecy rate is not achieved for a given
provided by the LFC. Active sensors with a weak channel to
communication link. When multiple communication links
the LFC flip their local decisions in order to confuse the
are present, for example, as in broadcast or interference
EFC, while the LFC discards the flipped reports. Active
channels, then one is typically interested in defining the
sensors are allowed to transmit simultaneously ð~ nk > 1Þ by
achievable secrecy rate region or secrecy capacity region,
transmitting on orthogonal waveforms. The sensors must
or the aggregate secrecy sum rate/capacity. Metrics other
however obtain precise CSI of their links to the LFC,
than secrecy rate or secrecy outage probability, such as
including both phase and amplitude information, so as to
relative SINR or BER, do not provide any information-
perform channel phase precompensation and enable
theoretic guarantees of security, but are often more trac-
coherent demodulation at the LFC.
table for purposes of system design.
As an improvement over [38], Jeon et al. [39] and
Choi et al. [40] proposed a CAE scheme where at each
B. Security in Sensor Networks instant a sensor has three possible actions: stay dormant,
Radio communications between sensors and the LFC report a ‘‘flipped’’ decision, or report its unaltered local
are inherently vulnerable to eavesdropping. Due to limited decision. The choice of action depends on where its in-
2
More rigorous definitions and examples can be found in [79]–[81] stantaneous channel fading gain to the LFC falls between a
and the extensive references within. set of publicly known thresholds f1 ; 2 ; 3 g, with

Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1751
Mukherjee: Physical-Layer Security in the Internet of Things
1 > 1  2  3 . That is, sensor i reports its unaltered
local decision if khi;k k2 > 1 , reports a bit-flipped decision
if 3  khi;k k2  2 , and stays dormant otherwise. The
channel fading gains, which depend on the location of the
sensors and LFC, are known to the LFC (assuming channel
reciprocity) but are unknown to the EFC due to the sta-
tistical independence of the sensor-to-LFC and sensor-to-
EFC channels, and thus the EFC cannot identify which
sensors are flipping their decisions. The optimal thresholds
are computed such that the numbers of flipping and non-
flipping sensors are equal, which turns out to satisfy the
perfect secrecy condition in (5).
The LFC performs square law combining and an LLR
decision fusion rule, since only statistical CSI of the sensor-
to-LFC channel gains is assumed to be known for additional
robustness. At high SNR, the LFC decision is quite simple:
the sensor-to-LFC channel gains are first approximated to
be the received signal magnitude. Then, the LFC compares
the approximate channel gains with the thresholds f1 ;
2 ; 3 g to determine whether the sensor report was flipped
via a hard decision rule. Once flipped bits have been re-
aligned, the ensuing high-SNR LLR decision rule has a
closed-form expression. The system efficiency is improved
compared to [38] in two ways: by allowing the LFC to also
exploit the flipped sensor reports for its LLR-based global
decision, and by using noncoherent digital modulation
which does not require exact CSI estimation. However, or-
thogonal reporting channels are required ð~ nk ¼ 1Þ, in order
for the LFC to estimate individual sensor channel gains.
3) Probabilistic Ciphering: Another related category of
techniques is based upon probabilistic ciphering. In [41],
the sensor observations are randomly mapped to a set of
discrete quantization levels, with the corresponding map-
ping probabilities known only to the LFC and not the EFC.
The suboptimal mapping probabilities and LFC decision rule
that jointly minimize its error probability subject to a con-
straint on the EFC error probability are then derived. This
approach is made more rigorous in [42], where the optimal
cipher matrices are obtained based on a divergence metric,
again assuming that they remain unknown to the EFC. Aysal
and Barner [43] investigated the problem of secure distri-
buted estimation by incorporating a stochastic cipher as an
additional block to the existing sensor networks to improve
secrecy. They showed a significant deterioration in the EFC’s
performance (in terms of bias and MSE) at the cost of a
marginal increase in the estimation variance at the LFC.
4) Optimal Quantization: It is also possible to optimize the
feedback reports emanating from the sensors. As an exam-
ple, Li et al. [44] investigated the problem of Bayesian dis-
tributed detection for the special case of nk ¼ 2 sensors in
the network, where the EFC has access to only one of the
sensor’s transmissions. The authors proved that LRT-based
tests (declaring an output of 1 or 0 depending upon whether
the received LLR is above or below some threshold) were
1752 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
optimal at the sensors if the network is designed to minimize
the expected detection cost at the LFC such that the mini-
mum average cost at the EFC is no greater than a prescribed
nonnegative value . It turns out that the sensor that is
being intercepted manipulates its feedback so that the EFC
gains practically no information about the true hypothesis,
while the LFC still extracts some useful information. The
same network scenario was studied assuming Neyman–
Pearson detectors in [45], where security was modeled using
a constraint placed on the EFC detection probability. Here,
it was shown that the optimal local quantizer is a determi-
nistic LRT, while the fusion rule may still be a randomiza-
tion between two or more LRTs.
A more general scenario with multiple eavesdropped
sensors was considered in [46], where an optimal 1-bit sen-
sor quantization rule was designed to maximize the
difference between the Kullback–Leibler divergences of
the LFC and EFC. Denoting the sensor quantization rule for
sensor i’s observation by i ðÞ, the output of the quantizer is a
single bit: xi;k ¼ 1 if the local hypothesis test decision is 1 ,
and xi;k ¼ 0 if the local decision is 0 . Alternatively, in [46]
and [47], the quantizer seeks to maximize the divergence at
the LFC while constraining the EFC’s divergence, under the
assumption of binary symmetric channels between sensors
and LFC/EFC. Numerical algorithms are then presented to
determine the optimal sensor quantization thresholds for
identical and nonidentical channel statistics. While security
is directly integrated into the sensor likelihood ratio test
procedure, these algorithms require knowledge of the EFC
channel statistics which is hard to obtain.
5) Compressive Sensing: In [48] and [49], the CS signal
processing technique is utilized for physical-layer security.
In CS, a linear transformation is applied to compress sparse
vectors by multiplying with a measurement matrix [50].
Reconstruction of the sparse vectors is possible in polyno-
mial time from fewer samples than suggested by the
Nyquist sampling theorem, using an optimization or algo-
rithmic framework. A single sensor ð~ nk ¼ 1Þ, LFC, and EFC
system is considered in [48], where both the sensor and
LFC extract secret bits from RSSI values of packets ex-
changed between them. These bits are used as seeds to feed
a shift register to generate an m-sequence. The output
m-sequences are later reorganized to form an ðM  NÞ CS
measurement matrix %i;k by the sensor and LFC without
coordination. The EFC cannot compute the same measure-
ment matrix since its channels and RSSI values are
uncorrelated with the legitimate nodes. Then, the sensor’s
M-dimensional transmit signal is constructed as
xi;k ¼ %i;k di;k
where di;k is the N-dimensional sparse vector of sensor
observations. The LFC recovers di;k by numerically

minimizing the ‘1 norm of its received signal, which is
feasible since the measurement matrix is known to it. In
[49], the channel matrix between multiple sensors and
multiple relays is directly used as the CS measurement
matrix in a two-hop network, and it shown that the eaves-
droppers’ probability of signal recovery is zero.
6) Artificial Noise: The final category of mechanisms is
based on the use of AN. AN denotes the transmission of
noise-like jamming signals by a subset of legitimate nodes
in order to degrade the SINR of the EFC. The sensors that
transmit AN may also simultaneously transmit actual data,
such that their transmit signal is given by xi;k ¼ di;k þ si;k
with appropriate power division between actual data di;k
and AN si;k . The identity of the nodes that transmit AN is
assumed to be unknown at the EFC. The AN signals there-
fore do not convey any useful information, and can also
degrade the LFC performance if not designed carefully. AN
was proposed in [51] and [52] to ‘‘confuse’’ the EFC about
the aggregate sensor observations sent to the LFC. The
related technique of cooperative jamming was used in [53]
for a two-hop system, where a relay transmit a jamming
signal simultaneously with the legitimate sensors in order
to confuse the EFC. In [54] and [55], the CS concept is
combined with AN in order to reduce the reporting over-
head of the sensors. A fraction of the sensors send falsified
data to confuse the EFC, and the LFC and EFC perfor-
mances are characterized using the deflection coefficient
metric.3 AN techniques are usually best effort and cannot
guarantee any particular secrecy level since they often
operate without EFC CSI.
A performance comparison of the weighted error pro-
bability (WEP) of CAE, TBMA, and probabilistic ciphering
with EFC error constraint of 0.4 [41] is shown in Fig. 4,
assuming n~k ¼ 18 sensors, unit noise variance, and equally
likely hypotheses PrðH0 Þ ¼ PrðH1 Þ ¼ 0:5. The WEP is a
weighted combination of the false alarm and missed detec-
tion probabilities: WEP¼ 0:5Pf þ 0:5ð1Pd Þ. It is observed
that only the channel-based bit flipping methods guarantee
perfect secrecy at all SINR levels or number of sensors, at
the cost of a slight degradation in LFC performance at low
SINR. The LFC performance of CAE is improved over
TBMA due to the additional exploitation of flipped sensor
reports. Note that a direct head-to-head comparison of all
the different techniques listed in this section is very
difficult and arguably unfair, since each of them is designed
to optimize a different performance metric under differing
eavesdropper and channel assumptions.
C. Communication Network Security
Compared to distributed detection in sensor networks, a
much broader range of transmission techniques have been
3
For a received signal vector y, the deflection coefficient of a binary
hypothesis test is defined as d ¼ ðEfyjH1 g  EfyjH0 gÞH 21
EfyjH0 gÞ, where 20 is the covariance matrix of y given the null hypo-
thesis. The higher is the deflection d, the better the detection performance.
Mukherjee: Physical-Layer Security in the Internet of Things
Fig. 4. EFC and LFC error probabilities for CAE, TBMA, and
probabilistic ciphering.
proposed for data secrecy, as exemplified in Fig. 5. In signal
processing terms, the optimal secrecy scheme (if known) is
determined by the number of legitimate receivers, the
number of antennas at the concerned nodes, and the
precision and extent of CSI known to them. While physical-
layer security encompasses both keyless secrecy and secret-
key agreement techniques, in this paper, we focus on keyless
security mechanisms. A brief review of the most common
secure transmission strategies is provided below.
1) Single User, Full CSIT: First consider the scenario of a
single legitimate receiver and a single eavesdropper with
additive Gaussian noise; the instantaneous CSIT of both is
known to the legitimate transmitter. When all users are
SISO (single antenna), the optimal power allocation that
maximizes the secrecy capacity is given in [57]. Transmis-
sion occurs only at instances when the intended receiver
has a higher SINR compared to the eavesdropper. If the
eavesdropper has a higher SINR than the legitimate re-
ceiver, then a positive secrecy rate cannot be achieved in
most cases. Exceptions to this rule of thumb include key-
based secrecy methods, such as the scenario where a two-
way, error-free public channel is available between the
sender and the receiver and is used to iteratively derive a
secret key [29]. A second key-based example is where
the transmitter rapidly modulates its transmit power based
on a preshared secret key so as to induce clipping and
saturation effects only at the EFC analog-to-digital con-
verter; the nonlinear nature of the receiver processing
implies that revealing the preshared key after the A-to-D
phase still does not allow the EFC to recover the secret
information [58].
0 ðEfyjH1 g 
For the MISO wiretap channel, the optimal strategy is
transmit beamforming along the direction of the
Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1753
Mukherjee: Physical-Layer Security in the Internet of Things
Fig. 5. Various transmission schemes for security and their associated
antenna and complexity requirements. Higher complexity generally
implies the need for more extensive channel state information at the
transmitter, along with increased encoding/decoding complexity.
generalized eigenvector corresponding to the largest gene-
ralized eigenvalue of the receiver and eavesdropper chan-
nel covariance matrices [59]. Of a number of heuristic
approaches, Li et al. [60] deliberately induced rapid time-
varying fluctuations in the eavesdropper’s channel to
prevent it from acquiring CSI. For the MIMO wiretap
channel, the asymptotically optimal linear precoding
strategy is GSVD-based spatial multiplexing [61], with
transmit power allocated over multiple generalized singu-
lar values of the receiver–eavesdropper channel matrix
pencil. The GEVD- and GSVD-based transmit solutions
must be recomputed for each new realization of channel
states.
2) Single User, Partial CSIT: A more realistic scenario is
when only the eavesdropper’s statistical CSIT (fading dis-
tribution) is known, in addition to legitimate CSIT. The
general approach in such cases is to optimize the average
secrecy rate with an expectation over the eavesdropper’s
channel state distribution. For the SISO case, the use of
AN injection and power bursting can achieve a positive
secrecy rate even when the main channel is arbitrarily
worse than the eavesdropper’s average channel [62]. AN
transmission is a widely used strategy in partial CSIT cases,
and refers to using part of the available transmit power to
send noise-like jamming signals simultaneously with data
in an attempt to degrade eavesdropper SINR.
In the SISO case, the AN also impacts the SINR of the
legitimate receiver and the power allocated for AN needs
to be chosen judiciously. When multiple antennas are
available at the transmitter, it can avoid degrading the
SINR of its desired receiver by transmitting AN in the
nullspace of the main channel [63]. The simplest example
would be when the transmitter has more antennas than its
desired receiver, thus, a nontrivial nullspace would exist
for the main channel matrix. The transmit beamformer or
precoder without eavesdropper CSIT is generally chosen so
as to maximize the rate between the transmitter and the
1754 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
legitimate receiver, for example, via maximum ratio trans-
mission. Similar results hold when optimizing the outage
probability of the secrecy rate [64]. In [65], it is shown that
below a threshold SINR, on–off power allocation without
AN is sufficient to achieve optimal scaling in secrecy capa-
city for a MISO system. The efficacy of AN schemes is, of
course, highly dependent on the accuracy of the legitimate
link’s CSIT; the use of AN based on inaccurate or outdated
CSIT can severely degrade the SINR of the intended
receiver [66].
3) Multiuser Systems: Next, we consider the scenario of
multiple eavesdroppers, and multiple legitimate receivers
or transmitters, such as in a broadcast or multiple access
channel. Multiple eavesdroppers are generally handled by
considering the eavesdropper with the ‘‘strongest’’ channel
gain from the legitimate transmitter, and choosing trans-
mission rate and power with respect to this worst case
node. In a two-user MIMO broadcast channel with sepa-
rate confidential messages for each intended receiver, a
nonlinear dirty paper coding technique can achieve the
respective maximum secrecy rates of both messages simul-
taneously, under a general power-covariance constraint
[67]. In a broadcast channel if the messages do not need to
be mutually confidential among the downlink receivers,
but must be protected only from external eavesdroppers, a
combination of AN and conventional downlink transmis-
sion methods can be applied [70].
In the case of multiple access systems with multiple
legitimate transmitters and a common desired receiver,
the simultaneous transmission of multiple signals makes it
inherently more difficult for eavesdroppers to isolate par-
ticular signals. When full CSIT is available, sophisticated
techniques that make use of the idea of interference align-
ment [68] can be applied, as espoused in [69] for example.
Here, transmitters repeat their symbols in two symbol
instants with appropriate scaling so as to create a full-rank
channel matrix at the main receiver and a unit-rank
channel matrix at the eavesdropper, in every two conse-
cutive instants. These coordinated actions create a 2-D
space for the signal received by the legitimate receiver,
while sustaining the interference in a 1-D space at the
eavesdropper.
Finally, large-scale systems with a multitude of legi-
timate nodes and eavesdroppers can be analyzed using
tools from stochastic geometry [71]–[76]. Here, the loca-
tions of receivers and eavesdroppers are modeled using
2-D point processes, and the statistical properties of the
SINR of different categories of users and secrecy rate are
derived subsequently. Furthermore, new metrics to re-
present the collective security of multiple links per unit
area of the legitimate network can be derived, such as
the network secrecy rate density and corresponding net-
work secrecy rate outage density [77], or the distributed
network secrecy throughput in a large-scale network with
hierarchy [78].
 Mukherjee: Physical-Layer Security in the Internet of Things
D. Challenges Specific to IoT Applications
It is prohibitively difficult in our view to employ many
of the above sensor and communication security schemes
in an IoT setting. The principal barrier is the problem of
accurate CSIT acquisition that includes both channel am-
plitude and phase information. The typical approach to ac-
quire CSIT is for one device to transmit predefined training
or pilot symbols. If the channels between a transmitter and
a receiver are reciprocal, for example, due to operating on
the same carrier frequency, then the nominal receiver can
send the training signals to allow the transmitter to esti-
mate the channel and acquire CSIT. Otherwise, the trans-
mitter needs to send the training signals so that the receiver
estimates the CSI, which is then quantized and fed back to
the transmitter. In the IoT, the acquisition of accurate le-
gitimate CSIT is prevented by limited channel training
opportunities and the lack of high-rate feedback channels.
Transmitting frequent training signals for channel estima-
tion is highly energy inefficient and wastes spectrum access
occasions in dense IoT deployments. Second, eavesdropper
CSIT is also difficult to acquire when eavesdroppers are
external to the IoT system and remain completely passive.
IV. IoT SENSOR SECURITY
Having briefly introduced the principal sensor network
security strategies in Section III-B, we now examine how
they compare in terms of complexity, energy efficiency,
and scalability, which are the key considerations for IoT
applications.
A. Complexity
The computational load of encoding-decoding opera-
tions and the cost of CSI acquisition required at the sensors
and controller are the two major factors that we define as
the overall complexity of a particular mechanism. The
methods based on CS in [48] fare poorly on both counts.
Constructing the CS measurement matrix at both sensor
and LFC requires both parties to arrive at the same, precise
CSI estimates. Furthermore, the CS methods are block
transmission schemes where the sensor has to collect N
observations before transmitting, while decoding at the
LFC involves numerically solving a convex norm-minimi-
zation problem. AN methods require the judicious splitting
of transmit power between sensor data and the AN, and
precise CSI is required if the LFC performance is to remain
unaffected. The 1-bit quantization method and bit-flipping
algorithms in TBMA, CAE, and probabilistic ciphering
clearly have the lowest transmission complexity. Some
element of numerical computation is required for all of
them, namely for the optimal quantization rule, sensor
activation thresholds, and optimal ciphering parameters.
Among these, CAE has the lowest computational complex-
ity and CSI requirement, since it can work with noncoher-
ent digital modulation and only requires coarse channel
magnitude information at the sensors and LFC.
Vol. 103, No. 10, October 2015 | Proceedings of the IEEE
B. Energy Efficiency
AN methods have the poorest energy efficiency of any
technique, since multiple sensors expend energy on send-
ing noise-like jamming signals. In probabilistic ciphering,
secure TBMA, and CS transmission, all the sensors in the
subsystem are assumed to be active at once. In 1-bit quan-
tization for nonidentical sensors and channels, a subset of
the ‘‘strongest’’ sensors are activated while the remainder
sleep, but the selection process involves a numerical search
procedure at the LFC. Censoring and CAE perform well in
terms of energy efficiency, since a simple threshold
comparison indicates to the sensors if they should transmit
or stay silent.
C. Scalability
Scalability of any security protocol as the number of
sensors n~k increases is an important aspect, given the large-
scale deployments expected in the IoT. It is desirable to
minimize the recomputation of protocol parameters if new
sensors are added or some sensors are rendered inoperable.
As long as the sensor observations are independent
identically distributed (i.i.d.), the censoring protocol does
not require new LLR transmit thresholds if n~k changes,
since the optimal thresholds are derived to maximize a per-
sensor divergence metric. AN methods are also arguably
quite scalable if we assume that sensors make autonomous
decisions regarding how much power to allocate for jam-
ming. This may however give rise to uncontrollably large
cochannel interference to legitimate nodes. The remaining
methods all require recomputation of key parameters such
as TBMA activation rates or CAE thresholds as n~k increases.
D. Summary
Table 1 summarizes the comparison of the various sen-
sor security schemes. From Table 1, it is observed that the
CAE scheme offers the best combination of low complexity,
coarse CSI requirements, energy efficiency, and perfect
secrecy. Unlike probabilistic ciphering, it does not require
the exchange of secret encryption matrices that are un-
known to eavesdroppers. Unlike type-based multiple access
or AN methods, it does not require a subset of sensors to
waste energy by transmitting dummy or jamming signals
that are only used to confuse the eavesdropper.
For the sake of comparison, consider the sensor secu-
rity scheme proposed for IoT based on public key crypto-
graphy in [56]. It requires the implementation of public
key infrastructure (PKI) using a certificate authority (CA)
at the controller, and identity-based cryptography (IBC) at
the resource-constrained sensors. In PKI, the CA issues a
certificate which provides a trusted link between the pub-
lic key and the identity of a user by the signature of the CA,
but requires certificate management and validation. In
IBC, a user’s public key is derived directly from its identity
information, such as IP addresses, and secret keys are
generated for users by a trusted third party known as the
private key generator (PKG). However, the secret keys
1755
Mukherjee: Physical-Layer Security in the Internet of Things
Table 1 Comparison of Sensor Network Physical-Layer Security Strategies
generated by the PKG still need to be sent to the sensors in
a secure manner. Furthermore, Li and Xiong [56] propose
a two-stage signcryption generation method that involves
both offline and online computations and the offline results
need to be stored at the sensors, which increases the com-
putational complexity. This suggests that physical-layer
security techniques can at least further augment crypto-
graphic protocols (e.g., IBC secret-key exchange phase),
and alleviate computational complexity since fewer offline
computations are required, if any.
V. IoT COMMUNICATIONS S ECURITY
It is evident that low complexity, high energy efficiency,
and scalability are also desirable attributes of security
schemes for IoT data communications. For this reason,
transmission schemes that are predicated upon instanta-
neous eavesdropper CSIT, such as the GEVD- and GSVD-
based methods, are not deemed to be viable in the IoT. In
addition, AN-based methods are also not justifiable due to
their higher energy expenditure and increased cochannel
interference to any adjacent user. Instead, in this section,
we examine possible alternative solutions for physical-layer
security that avoid these pitfalls, namely:
• they require minimal or no eavesdropper CSIT;
• they need coarse legitimate CSIT;
• they have simple CSI estimation requirements at
the receivers;
• they have low-complexity encoding and decoding;
• they do not employ AN.
It is understood that information-theoretic perfect secrecy
may not be achieved if such constraints are imposed on the
transmission methods. However, achieving an eavesdrop-
per error probability of 0.5 is sufficient as a proxy for per-
fect secrecy in practical scenarios such as fading channels.
TABLE 2 Comparison of Communication Physical-Layer Security Strategies
1756 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
A natural question is if the CAE scheme designed for
sensor networks can also be used for controller-to-actuator
downlink communications. Extending CAE to the general
broadcast channel in (3) with separate messages for the
actuators is not straightforward. However, we observe that
the CAE principle of selective message flipping is applica-
ble to downlink multicast, when a common message is
sent to all actuators. This new application of CAE to low-
complexity secure multicasting at the physical layer is
proposed next. The remainder of this section then reviews
candidate low-complexity solutions for point-to-point data
transmission scenarios case, namely, on–off switching,
noncoherent communication, and space-time codes. All of
these candidates have various pros and cons as summa-
rized in Table 2, and it would be premature at this stage to
choose one of them as a definitive solution perfect for IoT
applications. Furthermore, note that it is prohibitively
difficult to design secure broadcast schemes without
precise instantaneous CSIT, since interuser interference
cannot be mitigated otherwise.
A. Secure Multicast Based on CAE
Physical-layer multicasting requires that a common
information symbol be sent to multiple downlink recei-
vers. Consider the simple case of a binary digital
modulation scheme employed at the transmitter
(controller) to send one multicast bit to a~k actuators.
Based on CAE, the controller first estimates the channel
magnitudes khi;k k2 to the actuators, by observing uplink
signals for example. The actuators also possess knowledge
of their channel magnitudes from the controller, assuming
channel reciprocity. The controller then announces three
thresholds f1 ; 2 ; 3 g, after which the actuators infer in
which region between these thresholds does their channel
magnitude belong.

The actual multicast of the data symbol, say ‘k , is div-
ided into two phases. In the first phase, the controller
transmits the modulation symbol corresponding to the bi-
nary input ‘‘1’’ as tk . In the second phase, the controller
sends the ‘‘flipped’’ symbol corresponding to the binary
input ‘‘0.’’ Assume the true message bit is ‘‘0.’’ Then, the
thresholds are designed such that in the first phase, a~k =2
actuators have channel magnitudes satisfying 3 
khi;k k2  2 and flip their demodulation decision to the
correct bit value of ‘‘0.’’ In the second phase, the remaining
a~k =2 actuators have channel magnitudes larger than 1 and
do not flip their detected bit. The EFC’s equivocation after
the two phases is then maximized since it cannot infer in
which phase the bit flipping is performed. Of course, a
bandwidth penalty is incurred due to the use of two channel
uses to send one bit, but this scheme preserves the advan-
tages of CAE such as low complexity and coarse CSI
requirements.
B. Optimized ON–OFF Switching
A downlink analog to the censoring strategy for sensors
would be an on–off switching strategy, where a threshold
is used by the controller to determine whether to transmit
to a particular receiver at a given time instance. Nguyen
and Shin presented an on–off strategy [65] for scenarios
with a~k ¼ 1 in (3), i.e., a downlink with orthogonal
resource allocation based on OFDMA or time-division
multiple access. The scheme requires knowledge of the
legitimate SINRs and legitimate channel vector h ~ j , as well
as the eavesdropper SINR and the statistical distribution of
eavesdropper channel matrix ~Ge;k . The optimal on–off
switching that maximizes the average secrecy rate is then
given by [65, eq. 9]
 plete CSIT of the main channel.
1; ~ j k2  
if kh
? ¼
0; otherwise
where the transmit signal when active is beamformed in
the direction of h ~ j , such that tj;k ¼ pffiffiffiffiffiffi
Pc;j h ~ H x~j and x~j is the
j ceiver and the EFC will be different, since with high pro-
unit-power confidential information symbol intended for
actuator j. The switching threshold  is a function of the
eavesdropper SINR and the number of antennas mk and re ,
and needs to be computed once. However, the transmit
beamformer must be recomputed every time the legitimate
channel changes, which incurs additional complexity. A
solution to further reduce complexity would be to compute
the transmit beamformer based on the spatial correlation
matrix of h~ j , which can be estimated over time and would
not require beamformer recomputations.
C. Noncoherent Communication
Noncoherent communication in the classical sense
refers to the instantaneous CSI not being known to either
the transmitter or the receiver in a point-to-point link
Mukherjee: Physical-Layer Security in the Internet of Things
(statistical CSI is known to both). Thus, channel estima-
tion and tracking do not need to be performed continu-
ously at the receiver, which theoretically leads to a simpler
implementation. However, not knowing the instantaneous
CSIT of the legitimate channel leads to a zero secrecy rate
since the transmission rate cannot be set accurately.
Therefore, no prior work exists on keyless secrecy for
noncoherent systems, though secret-key agreement for
this scenario was studied in [82]. On the other hand, it
may be possible to design noncoherent security schemes if
a metric less stringent than secrecy rate is adopted to
quantify security, such as BER. Interestingly, there is no
prior work in this direction to our best knowledge, which
makes it a fruitful area for further research.
D. Secure Space-Time Coding
Space-time block codes (STBCs) gained prominence in
the 1990s as multiantenna transmission schemes that
could offer spatial diversity together with low encoding/
decoding complexity, without the need for CSIT [83]. The
most famous example of a STBC is the Alamouti code,
which is a rate-1 full diversity scheme for the case of two
transmit antennas. The fact that STBCs were designed to
operate without CSIT and have low complexity raises the
question if they can also be applied to secrecy scenarios.
The limited work in this area includes [84]–[86], for
MIMO scenarios with multiple transmit and receive
antennas. Fakoorian et al. [84] proposed a rate-one secure
STBC that allows for separable, low complexity (symbol-
wise) decoding at the intended receiver but not at the EFC
(which must perform ML detection). This is achieved by
ensuring that the effective STBC precoding matrix over
two time slots has orthogonal columns when seen at the
intended receiver, however, this scheme assumes com-
A transmit antenna selection scheme combined with
(7)
Alamouti coding was presented in [85] with reduced
legitimate CSIT requirements: only the indices of the two
strongest antennas and the average SINR of the main
channel. The decoding performance of the intended re-
bability the two strongest transmit antennas on the main
channel will not be the two strongest antennas to the EFC.
However, unlike [84] which used BER to characterize EFC
performance, the average SINR to the EFC is also assumed
to be known at the transmitter so as to achieve a nonzero
secrecy rate.
Finally, Ferdinand et al. [86] analyzed an orthogonal
STBC on spatially correlated MIMO channels, with the
assumption that no instantaneous CSIT of the main or EFC
channels is available. A general, closed-form expression for
the secrecy outage probability was derived, and it was
observed numerically that increasing antenna correlation
raised the secrecy outage probability. In summary, while
secure STBC schemes have several merits, the lack of
designs for single-antenna receivers and the enhanced
Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1757
Mukherjee: Physical-Layer Security in the Internet of Things
CSIT requirements compared to classical STBCs make
them less appealing for IoT applications.
VI . CONCLUDING REMARKS
Thus far we have reviewed several security techniques for
distributed detection and communication scenarios in an
IoT context. Most of these techniques, particularly those in
Section V, were generally developed for sophisticated
devices in networks of limited scale. Clearly this would no
longer be true in a dense IoT deployment for the various
reasons discussed until now. The ad hoc way in which
energy and complexity constraints are usually factored into
security studies at present also invites further scrutiny. In
this concluding section, we offer our views on how to
progress toward a more general understanding of physical-
layer security aspects, followed by a discussion of potential
new research topics.
A. Need for a New Framework
At a fundamental level, we perceive a need for a
theoretically well-founded and holistic approach for
incorporating complexity and energy constraints into
physical-layer security designs. For example, there is no
universal, commonly understood metric that precisely
characterizes the computational complexity of a chosen
encoding and decoding scheme. Most analyses resort to
counting floating point operations required at the trans-
mitter for one channel use, but that does not provide a
meaningful measure of complexity since processing capa-
bilities and clock rates are not taken into account. In [87],
measures of the complexity of decoding were analyzed by
considering the number of logic elements and clock cycles
needed to decode one codeblock, but it is unclear how to
apply these measures to general communication systems.
Regarding energy constraints in communication security,
these are commonly incorporated using a secure bit-per-
Joule formulation (secure communication rate normalized
by the energy consumption) [88]. The energy–secrecy
tradeoff has been explored partially for single-user net-
works [89], [90], but multiuser networks remain largely
unexplored in this context.
Additionally, the realization that the IoT is inherently a
massive network of MTC devices should spur new, funda-
mental definitions of secrecy metrics for point-to-multi-
point systems with a very large number of downlink
receivers. The same need arises for a multipoint-to-point
system with a very large number of uplink transmitters.
The information-theoretic concepts of [91] and [92] can be
a starting point in this direction. In their work, a new
paradigm in multiuser information theory models is con-
sidered where the number of users can grow arbitrarily
large together with the coding blocklength, which is re-
ferred to as the many-user regime. This is distinct from
prior work where the number of users was taken to infinity
after the blocklength was taken to infinity, and generally
1758 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
led to different results. As a simple example, for a large
number of users, the classical capacity/user/channel use
tends toward zero, even though the total amount of trans-
mitted information can be arbitrarily large. In the explo-
ratory work of [91] and [92], initial results such as the
array of achievable message lengths that can be reliably
decoded in the large-user regime (as opposed to channel
coding rate) have been presented. Extending these ideas to
secret message lengths in the many-user regime would be a
worthwhile first step toward a sound theory of physical-
layer security in large MTC networks.
B. New Applications
The application of physical-layer security to a system
with multiple practical constraints, such as the IoT, is seen
to be a challenge. The IoT will employ a wide array of
conventional RATs and new access methods to achieve
connectivity. Nontraditional communication channels can
have very different propagation characteristics as opposed
to the canonical Rayleigh and Rician multipath channel
models for broadband microwave systems. A step in this
direction was taken in [93], which studied the physical-
layer security of RFID systems and proposed an AN-based
security scheme for such backscatter communication
systems. Nonetheless, low-complexity secrecy schemes
tailored for VLC [94], acoustic, and molecular communi-
cation channels remain largely open problems, as well as
studies on how to jointly manage and allocate resources
securely among such multi-RAT systems.
Moving from a link layer to a network perspective, the
application of stochastic geometry to study security in
large-scale communication networks has been remarked
upon in Section III-C. The typical approach is to assume
that legitimate users, eavesdropping nodes, and possibly in-
terfering transmitters are all spatially distributed according
to independent 2-D point processes. Interestingly, a similar
analysis of security in distributed detection systems has yet
to be performed. A basic analytical model in this case would
be a sensor network with locations drawn from an isotropic
2-D point process, a single LFC and EFC, Gaussian-noise
channel links with only path loss, and the metric of interest
being one of those described in Section III.
Finally, relaxing the hierarchical communication con-
straint assumed in this work opens the door for a much
richer set of security case studies that include D2D and
ad hoc communication models. For example, the interplay
between distributed in-network signal processing algorithms
such as gossiping [95] and information security is a fertile
avenue for further study. Gossip algorithms feature repeated
localized message exchanges between neighboring sensor
nodes in order to arrive at a global consensus regarding a
parameter of interest, potentially without the assistance of a
fusion center. The additional consideration of secrecy is
expected to impact both the structure of messages
exchanged in the network, as well as the decision fusion
rules used to reach a consensus. h

REFERENCES
[1] Ericsson, ‘‘Ericsson mobility report
on the pulse of the networked society,’’
Jun. 2015. [Online]. Available: www.
ericsson.com.
[2] I. Stojmenovic, ‘‘Machine-to-machine
communications with in-network data
aggregation, processing and actuation for
large scale cyber-physical systems,’’ IEEE
Internet Things J., vol. 1, no. 2, pp. 122–128,
Apr. 2014.
[3] R. Lu, X. Li, X. Liang, X. Shen, and
X. Lin, ‘‘GRS: The green, reliability,
security of emerging machine to machine
communications,’’ IEEE Commun. Mag.,
vol. 49, no. 4, pp. 28–35, Apr. 2011.
[4] M. Weyrich, J.-P. Schmidt, and C. Ebert,
‘‘Machine-to-machine communication,’’
IEEE Software, vol. 31, no. 4, pp. 19–23,
Jul./Aug. 2014.
[5] A. Manzalini, R. Minerva, and C. Moiso,
‘‘Towards resource-aware network of
networks,’’ in Proc. 5th IEEE Int. Symp.
Wireless Perv. Comput., May 2010,
pp. 221–225.
[6] S. Talwar et al., ‘‘Enabling technologies
and architectures for 5G wireless,’’ in Proc.
IEEE Int. Microw. Symp., Tampa, FL, USA,
Jun. 2014, DOI: 10.1109/MWSYM.2014.
6848639.
[7] Huawei, ‘‘5G: A technology vision,’’ 2013.
[Online]. Available: http://www.huawei.com/
5gwhitepaper/.
[8] S. Wu, H. Wang, and C.-H. Youn, ‘‘Visible
light communications for 5G wireless
networking systems: From fixed to mobile
communications,’’ IEEE Network, vol. 28,
no. 6, pp. 41–45, Nov./Dec. 2014.
[9] B. Zhang et al., ‘‘PriWhisper: Enabling
keyless secure acoustic communication for
smartphones,’’ IEEE Internet Things J., vol. 1,
no. 1, pp. 33–45, Feb. 2014.
[10] T. Nakano et al., ‘‘Externally controllable
molecular communication,’’ IEEE J. Sel. Areas
Commun., vol. 32, no. 12, pp. 2417–2431,
Dec. 2014.
[11] L. Zhou and H. Chao, ‘‘Multimedia traffic
security architecture for the Internet of
Things,’’ IEEE Network, vol. 25, no. 3,
pp. 35–40, May 2011.
[12] J. Granjal, E. Monteiro, and J. S. Silva,
‘‘Security for the Internet of Things:
A survey of existing protocols and open
research issues,’’ IEEE Commun. Survey
Tuts., vol. 17, no. 3, pp. 1294–1312,
3rd Quart., 2015.
[13] C. E. Shannon, ‘‘Communication theory
of secrecy systems,’’ Bell Syst. Tech. J.,
vol. 28, pp. 656–715, 1949.
[14] W. Diffie and M. E. Hellman, ‘‘New
directions in cryptography,’’ IEEE Trans.
Inf. Theory, vol. IT-22, no. 6, pp. 644–654,
Nov. 1976.
[15] A. F. Skarmeta, J. L. Hernandez-Ramos, and
M. V. Moreno, ‘‘A decentralized approach
for security and privacy challenges in
the Internet of Things,’’ in Proc. IEEE
World Forum Internet Things, Mar. 2014,
pp. 67–72.
[16] K. Zhang, X. Liang, R. Lu, and X. Shen,
‘‘Sybil attacks and their defenses in
the Internet of Things,’’ IEEE Internet
Things J., vol. 1, no. 5, pp. 372–383,
Oct. 2014.
[17] M. Abomhara and G. M. Koien, ‘‘Security
and privacy in the Internet of Things: Current
status and open issues,’’ in Proc. IEEE Int.
Conf. Privacy Security Mobile Syst., May 2014,
DOI: 10.1109/PRISMS.2014.6970594.
Mukherjee: Physical-Layer Security in the Internet of Things
[18] A. D. Wyner, ‘‘The wire-tap channel,’’
Bell Syst. Tech. J., vol. 54, pp. 1355–1387,
1975.
[19] S. L. Y. Cheong and M. E. Hellman, ‘‘The
Gaussian wire-tap channel,’’ IEEE Trans.
Inf. Theory, vol. IT-24, no. 4, pp. 451–456,
Jul. 1978.
[20] J.-M. Liang, J.-J. Chen, H.-H. Cheng, and
Y.-C. Tseng, ‘‘An energy-efficient sleep
scheduling with QoS consideration in
3GPP LTE-Advanced networks for Internet
of Things,’’ IEEE J. Emerg. Sel. Top.
Circuits Syst., vol. 3, no. 1, pp. 13–22,
Mar. 2013.
[21] Ericsson, ‘‘LTE release 13,’’ White Paper,
Apr. 2015. [Online]. Available: www.ericsson.
com/res/docs/whitepapers/150417-wp-lte-
release-13.pdf.
[22] D. Dardari, A. Conti, C. Buratti, and
R. Verdone, ‘‘Mathematical evaluation
of environmental monitoring estimation
error through energy-efficient wireless
sensor networks,’’ IEEE Trans. Mobile
Comput., vol. 6, no. 7, pp. 790–802,
Jul. 2007.
[23] K. Piotrowski, P. Langendoerfer, and
S. Peter, ‘‘How public key cryptography
influences wireless sensor node lifetime,’’
in Proc. 4th ACM Workshop Security Ad Hoc
Sensor Netw., Alexandria, VA, USA, 2006,
pp. 169–176.
[24] J. Liu, N. Kato, J. Ma, and N. Kadowaki,
‘‘Device-to-device communication in
LTE-Advanced networks: A survey,’’ IEEE
Commun. Survey Tuts., , 2014, DOI: 10.1109/
COMST.2014.2375934.
[25] D. Zhu, A. L. Swindlehurst, A. Fakoorian,
W. Xu, and C. Zhao, ‘‘Device-to-device
communications: The physical layer security
advantage,’’ in Proc. IEEE Int. Conf. Acoust.
Speech Signal Process., Florence, Italy,
May 2014, DOI: 10.1109/ICASSP.2014.
6853869.
[26] J. Yue, C. Ma, H. Yu, and W. Zhou,
‘‘Secrecy-based access control for
device-to-device communication
underlaying cellular networks,’’ IEEE
Commun. Lett., vol. 17, no. 11, pp. 2068–2071,
Nov. 2013.
[27] H. Zhang, T. Wang, L. Song, and
Z. Han, ‘‘Radio resource allocation for
physical-layer security in D2D underlay
communications,’’ in Proc. IEEE Int.
Conf. Commun., Sydney, Australia, 2014,
pp. 2319–2324.
[28] C. Ma et al., ‘‘Interference exploitation
in D2D-enabled cellular networks:
A secrecy perspective,’’ IEEE Trans.
Commun., vol. 63, no. 1, pp. 229–242,
Jan. 2015.
[29] U. Maurer, ‘‘Secret key agreement by public
discussion from common information,’’
IEEE Trans. Inf. Theory, vol. 39, no. 3,
pp. 733–742, May 1993.
[30] R. Ahlswede and I. Csiszár, ‘‘Common
randomness in information theory and
cryptographyVPart I: Secret sharing,’’
IEEE Trans. Inf. Theory, vol. 39, no. 4,
pp. 1121–1132, Jul. 1993.
[31] L. Lai, Y. Liang, and H. V. Poor, ‘‘A
unified framework for key agreement
over wireless fading channels,’’ IEEE
Trans. Inf. Forensics Security, vol. 7, no. 2,
pp. 480–490, Apr. 2012.
[32] Y. Shen and M. Z. Win, ‘‘Intrinsic information
of wideband channels,’’ IEEE J. Sel. Areas
Commun., vol. 31, no. 9, pp. 1875–1888,
Sep. 2013.
[33] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam,
and E. Cayirci, ‘‘A survey on sensor networks,’’
Vol. 103, No. 10, October 2015 | Proceedings of the IEEE
IEEE Commun. Mag., vol. 40, no. 8,
pp. 102–114, Aug. 2002.
[34] D. Estrin, L. Girod, G. Pottie, and
M. Srivastava, ‘‘Instrumenting the world
with wireless sensor networks,’’ in Proc.
IEEE Int. Conf. Acoust. Speech Signal
Process., Salt Lake City, UT, USA,
May 2001, pp. 2033–2036.
[35] B. Kailkhura, V. S. S. Nadendla, and
P. K. Varshney, ‘‘Distributed inference
in the presence of eavesdroppers: A survey.’’
[Online]. Available: http://arxiv.org/abs/
1502.05448 .
[36] S. Marano, V. Matta, and P. K. Willett,
‘‘Distributed detection with censoring
sensors under physical layer secrecy,’’
IEEE Trans. Signal Process., vol. 57, no. 5,
pp. 1976–1986, May 2009.
[37] D. Kundur, W. Luh, U. N. Okorafor, and
T. Zourntos, ‘‘Security and privacy for
distributed multimedia sensor networks,’’
Proc. IEEE, vol. 96, no. 1, pp. 112–130,
Jan. 2008.
[38] H. Jeon, D. Hwang, J. Choi, H. Lee, and
J. Ha, ‘‘Secure type-based multiple access,’’
IEEE Trans. Inf. Forensics Security, vol. 6,
no. 3, pp. 763–774, Sep. 2011.
[39] H. Jeon, J. Choi, S. McLaughlin, and
J. Ha, ‘‘Channel aware encryption
and decision fusion for wireless sensor
networks,’’ IEEE Trans. Inf. Forensics
Security, vol. 8, no. 4, pp. 619–625,
Apr. 2013.
[40] J. Choi, J. Ha, and H. Jeon, ‘‘Physical layer
security for wireless sensor networks,’’ in
Proc. IEEE Int. Symp. Pers. Indoor Mobile
Radio Commun., 2013, DOI: 10.1109/PIMRC.
2013.6666094.
[41] R. Soosahabi and M. Naraghi-Pour,
‘‘Scalable PHY-layer security for distributed
detection in wireless sensor networks,’’ IEEE
Trans. Inf. Forensics Security, vol. 7, no. 4,
pp. 1118–1126, Aug. 2012.
[42] R. Soosahabi, M. Naraghi-Pour, D. Perkins,
and M. Bayoumi, ‘‘Optimal probabilistic
encryption for secure detection in wireless
sensor networks,’’ IEEE Trans. Inf. Forensics
Security, vol. 9, no. 3, pp. 375–385,
Mar. 2014.
[43] T. C. Aysal and K. E. Barner, ‘‘Sensor
data cryptography in wireless sensor
networks,’’ IEEE Trans. Inf. Forensics
Security, vol. 3, no. 2, pp. 273–289,
Jun. 2008.
[44] Z. Li, T. Oechtering, and K. Kittichokechai,
‘‘Parallel distributed Bayesian detection
with privacy constraints,’’ in Proc. IEEE
Int. Conf. Commun., Sydney, Australia,
Jun. 2014, pp. 2178–2183.
[45] Z. Li, T. Oechtering, and J. Jalden, ‘‘Parallel
distributed Neyman-Pearson detection
with privacy constraints,’’ in Proc. IEEE
Int. Conf. Commun., Sydney, Australia,
Jun. 2014, pp. 765–770.
[46] V. S. S. Nadendla, H. Chen, and
P. K. Varshney, ‘‘Secure distributed
detection in the presence of
eavesdroppers,’’ in Proc. 44th Asilomar
Conf. Signals Syst. Comput., Nov. 2010,
pp. 1437–1441.
[47] V. S. S. Nadendla and P. K. Varshney,
‘‘Design of binary quantizers for distributed
detection under secrecy constraints,’’ 2014.
[Online]. Available: http://arxiv.org/abs/
1410.8100v1.
[48] R. Dautov and G. R. Tsouri, ‘‘Securing while
sampling in wireless body area networks
with application to electrocardiography,’’
IEEE J. Biomed. Health Inf., 2014,
DOI: 10.1109/JBHI.2014.2366125.
1759
Mukherjee: Physical-Layer Security in the Internet of Things
[49] J. E. Barceló-Lladó, A. Morell, and
G. Seco-Granados, ‘‘Amplify-and-forward
compressed sensing as a physical-layer
secrecy solution in wireless sensor
networks,’’ IEEE Trans. Inf. Forensics
Security, vol. 9, no. 5, pp. 839–850,
Apr. 2014.
[50] D. Donoho, ‘‘Compressed sensing,’’
IEEE Trans. Inf. Theory, vol. 52, no. 4,
pp. 1289–1306, Apr. 2006.
[51] M. Anand, Z. Ives, and I. Lee,
‘‘Quantifying eavesdropping vulnerability
in sensor networks,’’ in Proc. 2nd Int.
Workshop Data Manage. Sensor Netw.,
Aug. 2005, pp. 3–9.
[52] A. Araujo, J. Blesa, E. Romero, and
O. Nieto-Taladriz, ‘‘Artificial noise
scheme to ensure secure communications
in CWSN,’’ in Proc. 8th Int. Wireless
Commun. Mobile Comput. Conf., Aug. 2012,
pp. 1023–1027.
[53] V. M. Rohokale, N. R. Prasad, and
R. Prasad, ‘‘Cooperative jamming for
physical layer security in wireless sensor
networks,’’ in Proc. IEEE Int. Symp. Wireless
Pers. Multimedia Commun., Taipei, Taiwan,
Sep. 2012, pp. 458–462.
[54] B. Kailkhura, T. Wimalajeewa, and
P. K. Varshney, ‘‘On physical layer secrecy
of collaborative compressive detection,’’ in
Proc. 48th Asilomar Conf. Signals Syst. Comput.,
Nov. 2014, pp. 51–55.
[55] B. Kailkhura, S. Liu, T. Wimalajeewa, and
P. K. Varshney, ‘‘Measurement matrix design
for compressive detection with secrecy
guarantees,’’ 2015. [Online]. Available:
http://arxiv.org/abs/1506.00238.
[56] F. Li and P. Xiong, ‘‘Practical secure
communication for integrating wireless
sensor networks into the Internet of
Things,’’ IEEE Sensors J., vol. 13, no. 10,
pp. 3677–3683, Oct. 2013.
[57] P. K. Gopala, L. Lai, and H. El-Gamal,
‘‘On the secrecy capacity of fading channels,’’
IEEE Trans. Inf. Theory, vol. 54, no. 10,
pp. 4687–4698, Oct. 2008.
[58] A. Sheikholeslami, D. Goeckel, and
H. Pishro-Nik, ‘‘Everlasting secrecy
by exploiting non-idealities of the
eavesdropper’s receiver,’’ IEEE J. Sel. Areas
Commun., vol. 31, no. 9, pp. 1828–1839,
Sep. 2013.
[59] A. Khisti and G. W. Wornell, ‘‘Secure
transmission with multiple antennas I:
The MISOME wiretap channel,’’ IEEE Trans.
Inf. Theory, vol. 56, no. 7, pp. 3088–3104,
Jul. 2010.
[60] X. Li, M. Chen, and E. P. Ratazzi,
‘‘Array-transmission based physical-layer
security techniques for wireless sensor
networks,’’ in Proc. IEEE Int. Conf.
Mechatron. Autom., 2005, pp. 1618–1623.
[61] A. Khisti and G. W. Wornell, ‘‘Secure
transmission with multiple antennas II:
The MIMOME wiretap channel,’’
IEEE. Trans. Inf. Theory, vol. 56, no. 11,
pp. 5515–5532, Nov. 2010.
[62] Z. Li, R. Yates, and W. Trappe,
‘‘Secret communication with a fading
eavesdropper channel,’’ in Proc. IEEE
Int. Symp. Inf. Theory, Nice, France,
Jul. 2007, pp. 1296–1300.
[63] S. Goel and R. Negi, ‘‘Guaranteeing secrecy
using artificial noise,’’ IEEE Trans. Wireless
Commun., vol. 7, no. 6, pp. 2180–2189,
Jun. 2008.
[64] S. Gerbracht, C. Scheunert, and
E. A. Jorswieck, ‘‘Secrecy outage in
MISO systems with partial channel
information,’’ IEEE Trans. Inf. Forensics
1760 Proceedings of the IEEE | Vol. 103, No. 10, October 2015
Security, vol. 7, no. 2, pp. 704–716,
Apr. 2012.
[65] T. V. Nguyen and H. Shin, ‘‘Power
allocation and achievable secrecy
rates in MISOME wiretap channels,’’
IEEE Commun. Lett., vol. 15, no. 11,
pp. 1196–1198, Nov. 2011.
[66] A. Mukherjee and A. L. Swindlehurst,
‘‘Robust beamforming for secrecy in
MIMO wiretap channels with imperfect
CSI,’’ IEEE Trans. Signal Process., vol. 59,
no. 1, pp. 351–361, Jan. 2011.
[67] R. Liu, T. Liu, H. V. Poor, and S. Shamai,
‘‘Multiple-input multiple-output Gaussian
broadcast channels with confidential
messages,’’ IEEE Trans. Inf. Theory, vol. 56,
no. 9, pp. 4215–4227, Sep. 2010.
[68] V. R. Cadambe and S. A. Jafar, ‘‘Interference
alignment on the degrees of freedom of the K
user interference channel,’’ IEEE Trans. Inf.
Theory, vol. 54, no. 8, pp. 3425–3441,
Aug. 2008.
[69] R. Bassily and S. Ulukus, ‘‘Ergodic
secret alignment,’’ IEEE Trans. Inf.
Theory, vol. 58, no. 3, pp. 1594–1611,
Mar. 2012.
[70] A. Mukherjee and A. L. Swindlehurst,
‘‘Utility of beamforming strategies
for secrecy in multiuser MIMO wiretap
channels,’’ in Proc. 47th Annu. Allerton
Conf. Commun. Control Comput.,
Oct. 2009, pp. 1134–1141.
[71] X. Zhou, R. Ganti, J. Andrews, and
A. Hjørungnes, ‘‘On the throughput
cost of physical layer security in decentralized
wireless networks,’’ IEEE Trans. Wireless
Commun., vol. 10, no. 8, pp. 2764–2775,
Aug. 2011.
[72] H. Wang, X. Zhou, and M. C. Reed,
‘‘Physical layer security in cellular
networks: A stochastic geometry
approach,’’ IEEE Trans. Wireless Commun.,
vol. 12, no. 6, pp. 2776–2787, Jun. 2013.
[73] P. Pinto, J. Barros, and M. Z. Win,
‘‘Secure communication in stochastic
wireless networksVPart I: Connectivity,’’
IEEE Trans. Inf. Forensics Security,
vol. 7, no. 1, pp. 125–138, Feb. 2012.
[74] P. Pinto, J. Barros, and M. Z. Win,
‘‘Secure communication in stochastic
wireless networksVPart II: Maximum
rate and collusion,’’ IEEE Trans.
Inf. Forensics Security, vol. 7, no. 1,
pp. 139–147, Feb. 2012.
[75] M. Z. Win, A. Rabbachin, J. Lee, and
A. Conti, ‘‘Cognitive network secrecy
with interference engineering,’’
IEEE Netw., vol. 28, no. 5, pp. 86–90,
Sep./Oct. 2014.
[76] P. Pinto and M. Z. Win, ‘‘Percolation and
connectivity in the intrinsically secure
communications graph,’’ IEEE Trans. Inf.
Theory, vol. 58, no. 3, pp. 1716–1730,
Mar. 2012.
[77] A. Rabbachin, A. Conti, and M. Z. Win,
‘‘Wireless network intrinsic secrecy,’’
IEEE/ACM Trans. Netw., vol. 23, no. 1,
pp. 56–69, Feb. 2015.
[78] J. Lee, A. Conti, A. Rabbachin, and
M. Z. Win, ‘‘Distributed network
secrecy,’’ IEEE J. Sel. Areas Commun.,
vol. 31, no. 9, pp. 1889–1900,
Sep. 2013.
[79] A. Mukherjee, S. A. A. Fakoorian, J. Huang,
and A. L. Swindlehurst, ‘‘Principles of
physical-layer security in multiuser wireless
networks: A survey,’’ IEEE Commun.
Surv. Tuts., vol. 16, no. 3, pp. 1550–1573,
3rd Quart., 2014.
[80] M. Bloch and J. Barros, Physical-Layer
Security: From Information Theory to Security
Engineering. Cambridge, U.K.: Cambridge
Univ. Press, 2011.
[81] X. Zhou, L. Song, and Y. Zhang, Eds.,
Physical Layer Security in Wireless
Communications. Boca Raton, FL,
USA: CRC Press, 2013.
[82] A. Agrawal, Z. Rezki, A. Khisti, and
M.-S. Alouini, ‘‘Noncoherent capacity of
secret-key agreement with public discussion,’’
IEEE Trans. Inf. Forensics Security, vol. 6,
no. 3, pp. 565–574, Sep. 2011.
[83] V. Tarokh, N. Seshadri, and A. R. Calderbank,
‘‘Space-time codes for high data rate
wireless communication: Performance
criterion and code construction,’’
IEEE Trans. Inf. Theory, vol. 44, no. 2,
pp. 744–765, Mar. 1998.
[84] A. Fakoorian, H. Jafarkhani, and
A. L. Swindlehurst, ‘‘Secure space-time
block coding via artificial noise alignment,’’
in Proc. 45th Asilomar Conf. Signals Syst.
Comput., Pacific Grove, CA, USA, Nov. 2011,
pp. 651–655.
[85] S. Yan, N. Yang, R. Malaney, and
J. Yuan, ‘‘Transmit antenna selection
with Alamouti coding and power allocation
in MIMO wiretap channels,’’ IEEE
Trans. Wireless Commun., vol. 13, no. 3,
pp. 1656–1667, Mar. 2014.
[86] N. S. Ferdinand, D. Benevides da Costa, and
M. Latva-aho, ‘‘Physical layer security in
MIMO OSTBC line-of-sight wiretap channels
with arbitrary transmit/receive antenna
correlation,’’ IEEE Wireless Commun. Lett.,
vol. 2, no. 5, pp. 467–470, Oct. 2013.
[87] J. E. Savage, ‘‘The complexity of
decodersVPart II: Computational
work and decoding time,’’ IEEE Trans.
Inf. Theory, vol. 17, no. 1, pp. 77–85,
Jan. 1971.
[88] D. Ng, E. S. Lo, and R. Schober,
‘‘Energy-efficient resource allocation
for secure OFDMA systems,’’ IEEE Trans.
Veh. Technol., vol. 61, no. 6, pp. 2572–2584,
Jul. 2012.
[89] M. C. Gursoy, ‘‘Secure communication
in the low-SNR regime,’’ IEEE Trans.
Commun., vol. 60, no. 4, pp. 1114–1123,
Apr. 2012.
[90] C. Comaniciu and H. V. Poor, ‘‘On
energy-secrecy tradeoffs for Gaussian
wire-tap channels,’’ IEEE Trans.
Inf. Forensics Security, vol. 8, no. 2,
pp. 314–323, Feb. 2013.
[91] T.-Y. Chen, X. Chen, and D. Guo,
‘‘Many-broadcast channels: Definition
and capacity in the degraded case,’’ in
Proc. IEEE Int. Symp. Inf. Theory,
2014, pp. 2569–2573.
[92] X. Chen and D. Guo, ‘‘Gaussian many-access
channels with random transmitter activities,’’
in Proc. IEEE Int. Symp. Inf. Theory, 2014,
pp. 3127–3131.
[93] W. Saad, X. Zhou, Z. Han, and H. V. Poor,
‘‘On the physical layer security of backscatter
wireless systems,’’ IEEE Trans. Wireless
Commun., vol. 13, no. 6, pp. 3442–3451,
Jun. 2014.
[94] A. Mostafa and L. Lampe, ‘‘Physical-layer
security for MISO visible light
communication channels,’’ IEEE J. Sel.
Areas Commun., vol. 33, no. 9, pp. 1806–1818,
Sep. 2015.
[95] V. Saligrama, M. Alanyali, and O. Savas,
‘‘Distributed detection in sensor networks
with packet losses and finite capacity links,’’
IEEE Trans. Signal Process., vol. 54, no. 11,
pp. 4118–4132, Nov. 2006.
 Mukherjee: Physical-Layer Security in the Internet of Things

ABOUT THE AUTHOR

Amitav Mukherjee (Member, IEEE) received the
B.S. degree from the University of Kansas,
Lawrence, KS, USA, in 2005, the M.S. degree
from Wichita State University, Wichita, KS, USA,
in 2007, both in electrical engineering, and the
Ph.D. degree in electrical and computer engineer-
ing from the University of California Irvine, Irvine,
CA, USA, in 2012.
He is currently a Research Engineer at Ericsson
Research, San Jose, CA, USA, where he is involved
in 3GPP LTE-A RAN1 standardization. From 2012 to 2014, he was a Re-
searcher with the Wireless Systems Research Laboratory, Hitachi America
Ltd., Santa Clara, CA, USA. From 2010 to 2012, he held internships with
Qualcomm Inc., San Diego, CA, USA; Mitsubishi Electric Research Labs
(MERL), Cambridge, MA, USA; Nokia Research Center, Helsinki, Finland;
Intel Corporation, Santa Clara, CA, USA; and Nokia Research Center,
Berkeley, CA, USA. His research interests encompass statistical signal
processing and wireless communications, with over 60 publications and
24 pending/issued patents in these areas.
Dr. Mukherjee received the Best Student Paper Award at the 2010 IEEE
International Workshop on Signal Processing Advances for Wireless
Communications (SPAWC), Marrakech, and has served as a Technical
Program Committee (TPC) Co-Chair/member for several conferences
such as IEEE International Conference on Communication (ICC), IEEE
Global Communication Conference (GLOBECOM), and IEEE Vehicular
Technology Conference (VT).

Vol. 103, No. 10, October 2015 | Proceedings of the IEEE 1761