You are on page 1of 15

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

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
1

Multi-band Spectrum Sensing and Resource


Allocation for IoT in Cognitive 5G Networks
Waleed Ejaz∗ and Mohamed Ibnkahla

Abstract—The proliferation of the Internet of Things (IoT) in total energy consumption and greenhouse gas emissions.
demands a diverse and wide range of requirements in terms Fourthly, IoT systems have to be economical. Last, but not
of latency, reliability, energy efficiency, etc. Future IoT systems least, IoT systems have to support a vast range of applications.
must have the ability to deal with the challenging requirements of
both users and applications. Cognitive 5G network is envisioned The unique characteristics of IoT applications make them
to play a key role in leveraging the performance of IoT systems. entirely different from traditional human user communication.
IoT systems in cognitive 5G network are expected to provide Predominantly, future IoT applications have to consider a wide
flexible delivery of broad services and robust operations under range of quality-of-service (QoS) requirements and conditions
highly dynamic conditions. In this paper, we present multi-band such as delay, reliability, energy consumption, etc. In addition,
cooperative spectrum sensing and resource allocation framework
for IoT in cognitive 5G networks. Multi-band approach can IoT nodes can have different computation and communication
significantly reduce energy consumption for spectrum sensing capabilities, memory, and power consumption. Thus, IoT sys-
compared to the traditional single-band scheme. We formulate tems have to adapt to the evolving nature of applications and
an optimization problem to determine a minimum number of services. There are many technologies available for IoT ap-
channels to be sensed by each IoT node in multi-band approach plications including Zigbee, Z-Wave, Bluetooth Low Energy,
to minimize the energy consumption for spectrum sensing while
satisfying probabilities of detection and false alarm requirements. and WiFi HaLow, etc. However, these technologies need to
We then propose a cross-layer reconfiguration scheme (CLRS) be connected to the Internet via 3rd Generation Partnership
for dynamic resource allocation in IoT applications with different Project (3GPP) long term evolution (LTE) or WiFi. The
quality of service (QoS) requirements including data rate, latency, worldwide deployment of LTE and LTE-advanced (LTE-A)
reliability, economic price, and environment cost. The potential makes it a suitable candidate because of ubiquitous coverage,
game is employed for cross-layer reconfiguration, in which
IoT nodes are considered as the players. The proposed CLRS mobility support, and plug and play features [2]. However,
efficiently allocate resources to satisfy QoS requirements through LTE is inherently designed for broadband human-to-human
opportunistic spectrum access. Finally, extensive simulation re- applications which makes it power hungry. The shortcomings
sults are presented to demonstrate the benefits offered by the of LTE and LTE-A are addressed in [3]–[5] with the focus on
proposed framework for IoT systems. architectural enhancements, QoS requirements, and applica-
Index Terms—Cognitive 5G networks, Internet of Things, tions. With the 3GPP release 12, LTE now supports Category
multi-band spectrum sensing, resource allocation. 0 (Cat 0) for low-power IoT nodes. Furthermore, release 13
specified enhanced machine-type communication (MTC) and
I. I NTRODUCTION narrowband IoT (NB-IoT) to address issues such as device
Internet of Things (IoT) is emerging as an important cost, battery life, coverage, support of a massive number of
paradigm that enables communication among physical objects IoT nodes, etc. [6]. Many leading industries including Ericsson
by empowering them with the ability to sense, communicate, and Nokia are working on eMTC and NB-IoT in order to offer
and process. IoT will enable real-world assets and objects to sophisticated IoT solutions.
exchange information, interact with people, support business According to authors in [7], the design of fifth generation
processes, and co-create knowledge. Effective deployment (5G) cellular networks will lead by the five disruptive technol-
of IoT systems will lead to significant cost savings, new ogy directions: device-centric architecture, millimeter Wave,
revenues, and employee productivity enhancements. It is pro- massive-MIMO, smarter devices, and MTC/ IoT communica-
jected that there will be approximately 11.6 billion mobile tions. 5G cellular networks should satisfy some fundamental
devices and connections by 2020, among which 7.4% are requirements for IoT systems including: 1) support of massive
low-power devices [1]. Further, the applications of IoT are IoT nodes, 2) minimum data rate should be ensured in all given
getting increasingly versatile, sophisticated, and ubiquitous. conditions, and 3) low latency. According to LTE-unlicensed
Thus, they are expected to face several challenges. Firstly, (LTE-U), the unlicensed spectrum can be used by cellular
IoT nodes need reliable links with low latency. Secondly, the devices, which can significantly enhance the capability of
large number of IoT nodes expected in near future will cause 5G to accommodate large number of IoT nodes. A resource
spectrum scarcity. Thirdly, IoT systems have to be energy- management scheme based on software defined network is
efficient as a large number of devices leads to an increase proposed for heterogeneous networks in [8]. Nevertheless,
disruptive changes at node and architecture level are required
W. Ejaz and M. Ibnkahla are with the Department of Systems and in 5G cellular networks to meet the requirements of new
Computer Engineering, Carleton University, Ottawa, ON, Canada (email:
waleed.ejaz@ieee.org; ibnkahla@sce.carleton.ca). technologies. Recently, two distinct architectures for MTC
∗ (Corresponding author) in 5G have been developed, i.e., European Telecommunica-

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
2

tions Standards Institute (ETSI) SmartM2M and oneM2M


Applications
[9]. The target of both is to resolve vertical fragmentation
of the IoT market. In [10], authors proposed a green IoT User interface User requirements Context information
architecture for IoT applications. A power-saving mechanism
is developed to optimize the data processing policies based Transport
on the data received by IoT nodes. A cognitive radio (CR)
based architecture for IoT is presented in [11]. A 5G cellular Number of retransmissions Congestion control

architecture is presented in [12] with key technologies like


massive MIMO, energy efficient communications, CR, visible Network
light communication, etc. Particularly, CR can play an im-
Routing Admission control QoS management
portant role to enhance the performance of IoT systems by
accessing spectrum opportunistically. An extensive research MAC
has been done on spectrum sharing in cognitive radios [13]–
Transmission and Congestion/ sensing
[18]. However, revolutionary steps may be required to take sleep scheduling window
Transmission rate
full advantage of IoT over cognitive 5G networks. Further, it
is important to investigate resource allocation to enhance the Physical
connectivity, communication, and collaboration of IoT nodes
Transceiver and antenna reconfiguration
in 5G networks [19].
In this paper, we propose a multi-band cooperative spectrum
sensing and resource allocation framework for IoT in cognitive
5G networks. It is assumed that each IoT node is equipped Fig. 1. Possible reconfiguration of hardware and software components for
with cognitive capability, i.e., ability to perform spectrum IoT systems in cognitive 5G networks.
sensing and reconfiguration. Spectrum sensing in cognitive
IoT communication must consider the hardware capabilities of
systems. For multi-band spectrum sensing, the objective is to
IoT nodes and the radio environment. In single-band spectrum
minimize energy consumption by assigning optimal number
sensing, each IoT node may have to sense all channels when
of channels to be sensed by each IoT node. Whereas, for
there is no coordination. This approach has some limitations,
cross-layer reconfiguration, the aim is to efficiently allocate
e.g., 1) IoT node may sense a channel which is always busy
resources to satisfy QoS requirements for different application
and 2) sensing all channels by an IoT node is computationally
in IoT systems.
expensive and complex. The limited capabilities of IoT nodes
have certain challenges for spectrum sensing. For example, it
A. Contributions
is very demanding to sense all channels by each IoT node
because IoT nodes are compact and low power. Moreover, The main contributions of this paper are as follows:
high sampling rates are computationally very expensive and • We propose a framework for multi-band spectrum sensing
difficult to achieve in IoT nodes. To address these challenges, and cross-layer reconfiguration for resource allocation for
we consider cooperative multi-band spectrum sensing where IoT in cognitive 5G networks.
compressive sensing is used as a local sensing scheme. In • We formulate an optimization problem to find minimum
multi-band approach, spectrum sensing task can be divided number of channels to be sensed by each IoT node in
among large number of IoT nodes. When compared to single- multi-band approach while satisfying the requirements for
band spectrum sensing, multi-band approach offers less energy the probability of detection and false alarms.
consumption for spectrum sensing, high throughput, and min- • We further propose a cross-layer reconfiguration scheme
imized hand-off frequency. (CLRS) for channel allocation in IoT systems based on
On the other hand, reconfiguration across multiple layers a game-theoretic approach.
(i.e., cross-layer reconfiguration) will be key for the success • The Potential Game is employed as game-theoretic
of IoT systems and services. The cross-layer reconfiguration framework for IoT systems 1) to satisfy QoS requirements
will adjust parameters at each layer to meet different QoS through opportunistic spectrum access and 2) for dynamic
requirements of IoT applications. We consider a layered resource allocation based on opportunistic capacity, reli-
model (application, transport, network, medium access control ability, latency, energy efficiency, environment impact.
(MAC), and physical (PHY) layers) presented in Fig. 1 for • Finally, simulations are conducted to verify the perfor-
illustration purpose, which shows possible reconfiguration at mance of the proposed framework and compared with
each layer of the protocol stack for IoT systems in cognitive existing work in literature.
5G networks. The parameters in each layer are linked directly The rest of this paper is organized as follows. In Section
or indirectly. Cross-layer designs can use the information and II, a brief overview of related works is presented. The sys-
objectives from multiple layers which allow various possible tem model for multi-band cooperative spectrum sensing and
reconfiguration levels. This result in significant improvements CLRS is given in Section III. In Section IV, we describe the
in the performance of IoT systems [20]–[22]. Therefore, it is proposed multi-band spectrum sensing and CLRS for channel
important to study multi-band cooperative spectrum sensing allocation. Simulation results are presented in Section V and
and cross-layer reconfiguration for resource allocation for IoT the conclusions are drawn in Section VI.

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
3

II. R ELATED W ORK cross-layer approach is proposed in [22] to deal with the
There have been research efforts in the past few years in heterogeneity of the IoT systems and impact of Internet as
the area of spectrum sensing and cross-layer designs for IoT an element of network architecture. The cross-layer approach
systems to address the challenges in cognitive networks. This is used to find optimal routing paths and communication
section gives a short review of the state-of-the-art in this field. parameters among IoT nodes for efficient and reliable end-
to-end communication. A cross-layer approach enabling com-
munication of a large number of IoT nodes in 5G networks
A. Spectrum sensing in IoT is proposed in [30]. The objective is to minimize overhead
Spectrum sensing must consider hardware limitations of and improve transmission efficiency while considering the
low power IoT nodes. Thus, traditional cooperative spectrum buffering of data and clustering of nearby users. The solu-
sensing cannot be adopted as is for IoT systems. Several tion uses information about the devices such as source and
spectrum sensing schemes are proposed in the literature for destination address, which are known by serving base station.
IoT systems. In [23], the authors proposed a hybrid framework In addition, information about the delivered content to the
for the dynamic spectrum management in IoT to reduce the buffer payloads and sending multiple payloads at once is used.
data acquisition and improve detection performance. A non- The clustering process allows all nearby users to exchange
iterative re-weighted compressive spectrum sensing algorithm payload information directly using device-to-device (D2D)
is proposed with geo-location database assistance for IoT sys- communication. In [31], a cross-layer optimization is proposed
tems. Authors in [24] proposed an energy-efficient reliable de- for reliability improvement of data delivery in IoT, where the
cision transmission scheme for cooperative spectrum sensing link-layer reliability is used on specific IoT nodes (having
in IoT. The proposed scheme improves detection probability frequent packet loss that can be detected at edge routers) and
and reduces energy consumption for low-power IoT nodes. the number of IoT nodes can be increased gradually to improve
In [25], authors proposed an efficient cooperative spectrum data delivery and network resources. A survey on adaptive
sensing (CSS) scheme which select sensor nodes to contribute communication techniques for IoT is presented in [32]. A
in CSS to minimize energy consumption in spectrum sensing. conceptual model is introduced to identify the adaptation in
An efficient two-stage compressive spectrum sensing algo- each layer of the protocol stack. Cross-layer communication is
rithm for MTC devices is proposed in [26], where 1) MTC validated in IoT architecture with pattern recognition services
devices sense only a part of the spectrum using compressive in [33]. Here, cross-layer communication enables application
sensing scheme and 2) a denoising algorithm is proposed to layer to reconfigure the lower layers to improve the decision-
enhance the spectrum sensing performance. In [27], a sub- making process. In [34], a cross-layer framework is proposed
Nyquist wideband sensing algorithm is proposed for MTC. for trustful communication. The framework combines MAC,
The proposed algorithm achieves high detection performance location, timing, and routing information in a reconfigurable
with reduced run-time and implementation complexity when manner. Trustfulness is handled by extending MAC with
compared to traditional wideband spectrum sensing. How- advanced encryption standard (AES), which can be used to
ever, aforementioned work did not consider the multi-band encrypt and authenticate packets.
approach. In the summary given in Table II, existing literature on
A basic multi-band spectrum sensing scheme is proposed in cross-layer implementations do not address dynamic resource
[16], [18] to reduce the expense and complexity of spectrum allocation in IoT applications with different QoS requirements.
sensing process. Each device senses multiple channels (a Hence, we employ potential game for cross-layer reconfigu-
subset of a total number of channels to be sensed) and reports ration to satisfy QoS requirements of IoT applications and
its result to fusion center which then fuses the sensing data dynamic resource allocation.
from all devices for cooperative decisions. In [28], an optimal III. S YSTEM M ODEL
cooperative multi-band spectrum sensing scheme is proposed.
A problem is formulated with an objective to maximize A. Architecture
throughput with constraints on energy and signal processing The IoT architecture adopted in this paper is presented in
resources consumed on spectrum sensing. The sensing dura- Fig. 2, where all devices are assumed to use CR technol-
tion for multi-band spectrum sensing is reduced by minimizing ogy [11]. Thus, the users can use any available spectrum
the reactive handoff latency in [29]. However, how to assign opportunistically. Primary users (PUs) have exclusive rights
channels to be sensed to each participating device is not to operate in licensed bands. The cognitive 5G architecture
considered in the existing literature summarized in Table I. for IoT is heterogeneous in nature and ensures the coverage
In this paper, we formulate an optimization problem for the of CR-enabled 5G cellular system everywhere and all the time.
assignment of a minimum number of channels for spectrum It will support a massive number of connected devices, and
sensing to IoT nodes and then implement compressive sensing a wide range of applications. Further, it is spectrum-efficient
to determine channels’ occupancy. as it can use all non-contiguous spectrum, can achieve better
system capacity, increase spectrum efficiency, reduce energy
consumption, and increase cell average throughput.
B. Cross-layer implementations for IoT Recent studies show that 80% of IoT communication takes
Several cross-layer implementations have been proposed place indoors and 20% outdoors [12], [35]. For indoor com-
for IoT systems that enable reconfiguration. For example, a munication, many technologies can be used for short range

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
4

TABLE I
E XAMPLES OF S PECTRUM S ENSING FOR I OT S YSTEMS .

a
Ref. Objective M Remarks
[23] A hybrid framework is proposed to reduce the A non-iterative re-weighted compressive spec-
data acquisition and improve detection perfor- trum sensing algorithm is proposed with geo-
mance location database assistance for IoT systems
[24] An energy-efficient reliable decision transmis- Detection probability is improved and energy
sion scheme is proposed for cooperative spec- consumption is reduced for low-power IoT
trum sensing nodes
[25] Minimize energy consumption for spectrum An efficient scheme is proposed to select sensor
sensing nodes for cooperative spectrum sensing
[26] Maximize spectrum sensing performance An efficient two-stage compressive spectrum
sensing algorithm is proposed for MTC devices
[27] Achieve high detection performance with re- A sub-Nyquist wideband sensing algorithm is
duced run-time and implementation complexity proposed for MTC
[16], [18] Reducing the expense and complexity of spec- 3 Each device senses multiple channels (ran-
trum sensing process domly assigned) for cooperative decisions
[28] Maximize throughput 3 An optimal multi-band spectrum sensing
scheme is proposed with constraints on energy
and signal processing resources
[29] Minimize the reactive handoff latency 3 The proposed scheme reduces the sensing du-
ration for multi-band spectrum sensing

a (M)ultiband

TABLE II
E XAMPLES OF C ROSS - LAYER I MPLEMENTATIONS FOR I OT.

Ref. Objective Layers involved Remarks


[22] Optimize end-to-end delay Network, MAC, Reconfiguring optimal routing paths and their corre-
and energy consumption PHY sponding communication parameters
for given QoS requirements
[30] Optimize overhead and Transport, MAC, Adjusting buffering of data and clustering nearby users
improve transmission PHY
efficiency
[31] Improve reliability of data MAC, Transport Adjusting number of critical nodes to use link layer
delivery acknowledgment requests
[32] Survey to maximize PHY, MAC, Discussed many possible adaptive and reconfigurable
performance and minimize Network, Transport measures at each layer
the adverse effects
[33] Validate cross-layer All layers Reconfiguring lower layers according to application
communication in IoT layer requirements to improve the decision making pro-
architecture with pattern cess
recognition services
[34] Enrich authentication Network, MAC, Reconfiguring communication parameters to adjust ac-
mechanism with temporal PHY cording to the requirements of application layer
information

communications such as WiFi, ultrawideband, millimeter-wave then communicate with the outdoor macrocell base stations
communications, radio-frequency identification (RFID), fem- [36].
tocell, Bluetooth, ZigBee, visible light communication, etc.
[12], [18], [35]. For outdoor communication, indoor BSs for B. System Description
each access network can directly or indirectly communicate
with antennas installed outside, on the top of buildings, or in We consider NA number of access networks for up-link
streets in the case of residential areas. These antennas will transmission in IoT systems and NP number of IoT nodes
equipped with cognitive capability. It is important to note that

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
5

Unlicensed Band Licensed Band sub-nyquist rates. In compressive sensing, compressed mea-
Dense CR small
cells
surements of signal are collected and then signal is recovered
CR ad hoc sensor from these measurements at IoT node. After obtaining the re-
network
constructed signal, energy detection is performed to determine
the presence or absence of PU.
The energy TEk of the received signal rk,c (n) by the k-th
CR BS IoT node is given by
N
1 X 2
IoT node TEk = |rk,c (n)| . (3)
N n=1
CR ad hoc sensor network in
licensed band
The local probability of detection Pd,ck and the probability
D2D communication
of false alarm Pf,ck over the additive white Gaussian noise
Geo-location Stand-alone
database sensing device PU CR BS Cloud server (AWGN) channel are expressed as [37]
CR IoT Cognitive IoT Core network Client
PU BS
gateway node p 
p
Pd,ck = QN 2γk , λEk , (4)
Fig. 2. Cognitive 5G architecture for IoT [11].

λ
each IoT node can act as a gateway and many other sensor Γ(N, E2 k )
Pf,ck = , (5)
nodes can report their data to it. For example, an IoT node in Γ(N )
this paper can represent a gateway of smart home as shown in where Γ(., .), Γ(.), and QN (., .) are the incomplete gamma
Fig. 2. A smart home can have a single IoT node to collect data function, the gamma function, and the Generalized Marcum Q-
from a large number of sensor nodes deployed in the home. function, respectively. γk is the instantaneous signal-to-noise
However, data collection process is not focus of this paper. ratio (SNR) of the k-th IoT node within the correspondent
Each access network a has αa channels in each cell, where observation time and λEk is the threshold of k-th IoT node
the total number of cells in each access network is γa . This for the energy detector.
model is similar to the one presented in our previous work A list of symbols used in the model and their description
[18]. The system capacity for the combination of all access is provided in Table III.
networks can be written as

γa X
NA X αa
D. QoS Requirements for IoT Systems
X
a IoT applications have diverse QoS requirements including
C= Bc log2 (1 + ζb,c ), (1)
a=1 b=1 c=1 data rate, latency, reliability, etc. Moreover, due to massive
where Bc is the bandwidth of channel c and ζb,ca
is the signal- deployment of IoT nodes, the environment and economic
to-interference-plus-noise ratio (SINR) at the IoT node on impact must be considered in the IoT system design. This is
channel c belonging to cell b in access network a. For multi- because an increase in service price will increase the overall
band spectrum sensing, we consider a cluster based network cost of IoT system and an increase in energy consumption
with C channels and a control channel. It is assumed that the is considered as a major threat to the environment as it is
C channels consist of channels for NA access networks. The indirectly linked to greenhouse gas emissions.
control channel is used for communication between IoT nodes Let SC be the set of service classes in the IoT system. Each
and cluster head/ central entity. class sc ∈ SC has the following QoS parameters: data rate
The total energy efficiency for the combination of all access (RD ), latency (L), reliability (RE ), economic impact (EC ),
networks is defined as and environment impact (EN ). Without loss of generality
γa X
and for illustration purposes, we consider a limited set of
NA X αa a
X Bc log2 (1 + ζb,c ) service classes with QoS parameter attributes as indicated in
E= , (2)
a=1
P (d ) + PRF ab,c
c=1 A b,c
a v Table IV. We defined these parameters over levels 0 to 3
b=1
according to their significance for illustration purposes, with
where PA is the amplifier energy required at the transmitter 0 as unimportant, 1 as low, 2 as important, and 3 as most
and v is the path loss exponent. dab,c and PRF ab,c are the important. We consider the following service classes:
distance between the IoT node using channel c and energy
consumed by radio frequency circuit of IoT node using chan- High data rate/ Real time (HRRt): This service class has
nel c belonging to cell b in access network a, respectively. stringent requirements on both data rate and latency. Therefore,
both are most important parameters in Table IV, i.e., RD =3
and L=3. The performance of cognitive IoT communication
C. Compressive Spectrum Sensing degrades if the requirement of any one of them is not sat-
We consider compressive sensing algorithm proposed in isfied. Thus, other parameters (RE , EC , and EN ) are set to
[26] as local spectrum sensing scheme for IoT nodes. Com- lower levels of importance. The applications this service class
pressive sensing can make spectrum sensing fast by taking include video transmission, surveillance, and smart grid.

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
6

TABLE III TABLE IV


D ESCRIPTION OF THE SYMBOLS USED IN THE MODEL . QOS SERVICE CLASSES FOR I OT APPLICATIONS .

Symbol Description No. Service class RD L RE EC EN


NA Number of access networks available 1 High data rate/ Real 3 3 2 0 1
NP Number of IoT nodes time (HRRt)
αa Number of channels in each cell of a-th access 2 Low data rate/ Real 1 3 2 2 1
network time (LRRt)
γa Total number of cells in a-th access network 3 High data rate/ Non 3 1 1 1 2
a
ζb,c SINR at IoT using channel c belongs to cell b in Real time (HRNRt)
a-th access network 4 Low data rate/ Non real 1 1 1 3 3
Bc Bandwidth of channel c time (LRNRt)
C System capacity
PRF Energy consumed by radio frequency circuit
PA Amplifier energy required at the transmitter
dab,c Distance between the IoT node using channel c Compressive sensing
and cell b in a-th access network I1 subset of C
IoT node 1
E Energy efficiency of system Compressive sensing Cooperative
TE k Energy of the received signal by k-th IoT node Optimization
I2 subset of C Sensing
Decisions and
of channels to IoT node 2 Adaptive
Pd,ck Local probability of detection be sensed
Energy
Sensing
Efficient
Pf,ck Local probability of false alarm using (12)
Resource
b,a Allocation
RD k
Data rate of k-th IoT node
b,a Compressive sensing
Lk Latency of k-th IoT node IM subset of C
b,a IoT node NP
RE k
Reliability of k-th IoT node
b,a
EC k
Economic price for bandwidth Bkb,a Fig. 3. Multi-band spectrum sharing framework for IoT in cognitive 5G
b,a
EN k
Environment cost of power Pkb,a networks.
hb,a
k,c Channel gain between k-th IoT node and BS of
cell a high importance to economic and environment impact. The
Pkb,a Power consumed by k-th IoT node application such as supervisory control and data acquisition
σcn ,cm Interference function between k-th and m-th IoT (SCADA), distribution automation, distributed management
nodes and control, etc. are fit in this service class.

IV. S PECTRUM S ENSING AND R ESOURCE A LLOCATION


Low data rate/ Real time (LRRt): This service class consists F RAMEWORK
of applications which require low data rate and are delay
In this section, we present a framework for IoT in cognitive
sensitive, i.e., RD =1 and L=3. Thus, the cognitive IoT nodes
5G networks which consists of multi-band spectrum sensing
can not tolerate delay-bound. Other parameters are set to lower
and cross-layer based resource allocation as shown in Fig. 3.
levels of importance because we need real time communica-
tion. Applications in this service class include smart metering,
price signaling in smart grid, sensing in smart buildings, etc. A. Multi-band Cooperative Spectrum Sensing
High data rate/ Non Real time (HRNRt): This service The cognitive system employs multi-band spectrum sensing
class has high data rate requirement and can be delay tol- where compressive sensing is used as a local sensing scheme.
erant, i.e., RD =3 and L=1. The performance of cognitive The system can help to minimize energy consumption for
IoT nodes deteriorates when the required data rate is higher spectrum sensing and maximize throughput while minimiz-
than achievable. In this service class, we set parameters for ing the handoff frequency [16]. In multi-band cooperative
reliability and economic cost as low whereas the environment spectrum sensing, each cognitive IoT node senses multiple
impact as important. This is because the resources with less channels to determine PU presence or absence, i.e., each IoT
economic cost and reliability can be a suitable option for users nodes to solve multiple binary hypothesis model:
of this service class such as WiFi channels. Smart homes for
multimedia distribution and cloud-based file sharing are the H0k,c : yk,c = wk,c
(6)
common applications in this service class. H1k,c : yk,c = xc + wk,c ,
Low data rate/ Non real time (LRNRt): This service class where c = 1, 2, ....Ik ≤ αa is the channel index, such that each
requires low data rate and has no strict requirements on IoT node sense Ik channels, yk,c is the sensed signal at c-th
latency, i.e., level for both RD and L are 1. However, the channel, xc is the transmitted PU signal, and wk,c is a zero-
performance is still bounded by the given tolerable limits mean AWGN with unit variance. Each k-th IoT node decides
for both data rate and delay. In this service class, we give H0k,c and H1k,c if the c-th channel is unoccupied and occupied,

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
7

τs T- τs (
1, if k-th IoT node is sensing c-th channel
φkc = (10)
1 2 Ik Data transmission 0, otherwise.
The energy consumption level of the k-th IoT node cor-
T
responds to four modes: sensing, transmitting, receiving, and
Fig. 4. Frame structure for multi-band spectrum sensing. silent. Let Ps,k be the power consumed in sensing by k-th
IoT node. Thus, the energy consumed in spectrum sensing
is Ps,k × τs , where τs is the sensing time. Our objective
is to minimize the total energy consumption for spectrum
respectively. The frame structure for multi-band distributed
sensing by minimizing the number of channels sensed by
spectrum sensing is shown in Fig. 4.
each IoT node while satisfying requirements of the probability
Each IoT node performs spectrum sensing for multiple of detection and the probability of false alarm. Thus, an
bands and employs threshold rule to find the local decision. All optimization procedure shall be invoked according to the
IoT nodes report their local decisions (LDk,c ) for the sensed required probability of detection and false alarm. The utility
channels to the central entity in the form of 0 or 1 using control function for the energy minimization for spectrum sensing can
channel. At the central entity, the global decision (GDc ) for be written as:
all channels is obtained by combining local decisions. In this
paper, we consider m-out-of-K rule for deciding presence Uk,c = Ps τs φk,c . (11)
or absence of PU. According to this rule, the central entity
Consequently, the energy minimization problem can be
declares the PU’s presence if at least m IoT nodes among
formulated as:
K report its presence, and declares its absence in the opposite
case. The mathematical expression for global detection in case
NP X
αc
of m-out-of-K rule can be written as: X
min : Uk,c
φk,c ∀k,c
 k=1 c=1
K
P
LDk,c ≥ m Subject to : Qd,c ≥ Q̄d , ∀c (12)
1,


GDc = k=1 ∀c (7) Qf,c ≤ Q¯f , ∀c
PK
0,

 LDk,c < m. NP
X
k=1 φk,c ≥ 1, ∀c
k=1
The global probability of detection Qd,c and the global
probability of false alarm Qf,c for c-th channel are calculated where Q̄d and Q¯f are the targeted global probability of
at the central entity by combining the results of cooperative detection and global probability of false alarm, respectively.
IoT nodes and is given by [38]: Constraints 1 and 2 satisfy the requirement of the global prob-
ability of detection and false alarm, respectively. Constraint 3
NP   ensures that each channel should be sensed by at least one IoT
X NP node.
Qd,c = k
φk,c Pd,c (1 − Pd,ck )NP −k , (8)
k k
The optimization problem in (12) is integer programming
k=NP /2
since all the decision variables are integer. One possibility
NP   is to compute all possible combinations of φk,c (i.e., ex-
X NP haustive search) in order to find optimal solution, however,
Qf,c = k
φkc Pf,c (1 − Pf,ck )NP −k . (9)
k k
this is computationally expensive and impractical for a large
k=NP /2
number of IoT nodes. Thus, we apply branch and bound
The number of channels sensed by each IoT node Ik algorithm given in Algorithm 1 which also enumerates similar
depends on the total number of channels to be sensed and to exhaustive search, however, cuts down enumeration when
total number of IoT nodes participating in the cooperation. possible. This can significantly reduce the average complexity
It is assumed that the central entity assigns channels to IoT of branch and bound [39]. In branch and bound algorithm, in
nodes in such a way that each channel will be sensed at least each iteration, φ∗k,c is the incumbent solution and U ∗ is its
once. The spectrum sensing time depends on the size of Ik objective. The iterative process continues until there are some
as shown in Fig. 4. Moreover, one channel can be sensed active nodes. For each active node Θ, algorithm solve linear
by multiple cognitive IoT nodes depending on several factors programming (LP) relaxation and denote the solution as φk,c
including priority, interference mitigation, detection reliability, and LP relaxation of Problem(a) as ULP (Θ). Now there are
etc. [16]. The scheme allows channels to be re-assigned for three possible actions: 1) if ULP (Θ) ≥ U ∗ then it will prune
spectrum sensing to cognitive IoT nodes and to re-optimize node k, 2) ULP (Θ) < U ∗ and φ(Θ) is feasible for integer
sensing parameters whenever IoT nodes notice changes in the program then update the incumbent solution, i.e., φ∗ = φ(Θ)
performance or radio environment. We define a binary variable and prune node Θ, and 3) ULP (Θ) < U ∗ and ψ(Θ) is
φk,c to denote whether the k-th IoT node is actually sensing infeasible for integer program then the algorithm will mark the
the c-th channel or not as follows: children of node Θ as inactive. This will result in -suboptimal

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
8

Algorithm 1 Multi-band Spectrum Sensing using branch and be employed to guarantee convergence to Nash equilibrium.
bound algorithm. Nash equilibrium is considered as steady state point after
1: Set φ∗ ∗
k,c = 0 ∀k, c and U = 0 which player cannot change their strategies to maximize
2: while there are some active nodes do their utilities. Therefore, each player maximizes its utility by
3: Select an active node Θ and mark it as inactive choosing appropriate strategies while considering the current
4: Solve linear programing (LP) relaxation: denote so- strategies of other players [41]. The best response update
lution as φk,c and LP relaxation of Problem(Θ) as strategy of a player while considering strategy profile s−1 at
ULP (Θ) t + 1 must satisfy
5: if ULP (Θ) ≥ U ∗ then
6: Prune node a 0

7: else if ULP (Θ) < U ∗ and φ(Θ) is feasible for integer Skt+1 ∈ argmax Uk (sk , st−k ), (14)
0
sk ∈Sk
program then
8: φ∗ = φ(Θ) where the strategy profile of k-th player at time t is represented
9: Prune node Θ by (stk , st−k ).
10: else if ULP (Θ) < U ∗ and ψ(Θ) is infeasible for integer
Definition 1: Nash equilibrium is a set of pure strategy profiles
program then
of all players, s∗ = (s∗k , s∗−k ) if and only if
11: Mark the children of node Θ as active
12: end if
0 0 0
13: end while Uk (s∗k , s∗−k ) ≥ Uk (sk , s∗−k ), ∀sk 6= s∗k , ∀Sk ∈ Sk , ∀k ∈ NP .
14: return the best solution and it’s minimum value (15)
Definition 2: A normal strategy game is considered as exact
solution [40]. The worst case complexity of branch and bound potential game [42] if there exists potential function Φ : S →
algorithm is equal to exhaustive search. However, the average R, which satisfies
complexity of branch and bound algorithm is extremely low
0 0
and this can be considered as reasonable performance indicator Φ(sk , s−k ) − Φ(sk , s−k ) = Uk (sk , s−k ) − Uk (sk , s−k ),
[39]. (16)
0

B. CLRS for Resource Allocation based on Game Theory where sk ∈ Sk and Φ is exact potential function of the game.
It is important to note that Φ is not the function of any player
We propose a CLRS for IoT nodes reconfiguration and
involved in the game and indicates any unilateral deviation of
resource allocation based on QoS requirements. The scheme
players. Further, if the strategy profile s∗ is selected according
is based on the potential game where the IoT nodes act as
to argmax Φ(S) then the Nash equilibrium is guaranteed. In
players. Each player (or IoT node) is capable of reconfiguring s∗ ∈S
its operating parameters according to the best available access the potential game, the strategies of users improved to increase
network. The PHY layer is responsible for the CR module, potential function. Thus, if a player deviates from its strategy
operating frequency, and transmit power. The MAC layer is after Nash equilibrium then its utility will not increase.
responsible for access network selection and resource alloca- In CLRS, reconfiguration is performed according to the
tion according to the requirement of service classes given in diverse QoS requirements for service classes given in Ta-
Table IV. ble IV based on weighted instantaneous date rate, latency,
reliability, economic impact, environment impact, and cross
Basic concepts: The potential game is a type of strategic non- co-tier interference. A utility function is defined in order to
cooperative game in which all players autonomously learn express the utility of k-th device for a particular service to
their strategies. There are three fundamental components of meet QoS requirements for HRRt, LRRt, LRNRt, and HRNRt
non-cooperative game which includes players, strategies, and service classes (details on forming utility functions from game
utilities. A game can be represented as theory prospective can be found in [43]). The utility function
for CLRS for each IoT node can be formulated as [18]:
P = {NP , {Sk }k∈NP , {Uk (.)}k∈NP }, (13)
b,a
where NP is the number Uk = ωkRD RD − ωkL Lb,a RE b,a
k + ωk REk
Q of players, Sk is set of strategies k
of k-th player S = Sk , and Uk is the utility of k-th + ωkRP hb,a b,a EC b,a b,a
k∈NP k,c Pk − ωk ECk Bk
player, Uk :S → R which maps strategy profiles S into a NP
X
b,a b,a
real value R. The utility function Uk shows the satisfaction − ωkEN EN k
Pk − hb,a
m,c σck ,cm
of k-th player while considering its own strategy Sk and m=1,m6=k
strategies of other players which are denoted by S−k = NP
X
{S1 , S2 , ..., Sk−1 , Sk+1 , ..., SNP }. − hb,a
k,c σcm ,ck (17)
m=1,m6=k
Best Response Update strategy and Nash equilibrium: In
b,a
the potential game, best response dynamics update strategy can where RD k
is the data rate, Lb,a
k
b,a
is the latency, RE k
is the

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
9

• In the decision step, each player decides to update its


strategy according to the best response update strategy.
Radio Enviornment

Application
If the device updates its strategy, then it will share this

layer
 QoS requirements strategy with its neighboring players.

Decision V. P ERFORMANCE E VALUATION

MAC
layer
 Network Selection
 Channel allocation In this section, we evaluate the performance of proposed
multi-band spectrum sensing and CLRS for IoT in cognitive
 CR module 5G networks. We consider different scenarios for IoT appli-

PHY
layer
Analysis &  Operating frequency cations with the number of IoT nodes NP varying from 300
Sensing  Transmit power
Reasoning to 425. In our case, 300 to 425 is a reasonable number for
IoT nodes as many sensor nodes can connect with each one
Fig. 5. Steps for reconfiguration in IoT nodes with reconfigurable parameters
at each layer.
of them. For example, in case of smart homes, on average if
we have 20 sensors nodes per house then each cell represents
8000 sensor nodes. These IoT nodes are randomly distributed
b,a
reliability, EC is the economic price for bandwidth Bkb,a , and in a rectangular field of size 1000m × 1000m.
k
b,a
EN k
is the environment cost of power Pkb,a corresponding to
a particular application class of the k-th player in cell b of A. Multi-band Spectrum Sensing
access network a. ωkRD , ωkL , ωkRE , ωkEC , and ωkEN are the The energy consumed for spectrum sensing is investigated
coefficients which intensify the QoS requirements according while allocating minimum number of channels to be sensed
to the service classes mentioned in Table IV. ωkRP , the reward/ by each IoT node. We consider different scenarios for the
penalty coefficient is set to 1 when the access network is cost number of IoT nodes NP and number of channels CP(varies
effective and has less impact on the environment (e.g., ZigBee C
from 30 to 50). The k-th IoT node senses Ik = c φk,c
and WLAN) and is set to -1 otherwise. hb,a m,c is the channel channels and transmits data on one of the available channels.
gain between m-th IoT node and cell b which serves the k-th The average SNR of IoT nodes is assumed to be −10dB. For
IoT node. σck ,cm is the interference function which is one if energy consumption and sensing time for spectrum sensing,
channel c is same for the k-th and m-th IoT node and zero we adopt simulation parameters given in [17].
otherwise. Fig. 6 shows the energy consumption for spectrum sensing
If we model potential function for CLRS as: versus number of channels C and targeted global probability
of detection Qd for NP = 400 and targeted gloabal probability
P N of false alarm Qf = 0.05. It is clear that the higher the
1X
Φ(S) = Uk , (18) number of channels and requirement of Qd , the higher the
2 energy consumed for spectrum sensing. For instance, energy
k=1
consumption increases rapidly as the number of channels
then (16) will hold [14]. The proof to show Φ(S) is exact
increases, whereas the energy consumption increases slowly
potential function is given in Appendix A.
with the increase in Qd . This is due to the fact that increasing
By employing the best response update strategy (in which
number of channels allocate more channels to IoT nodes
the IoT nodes (players) maximize their utility while consid-
to participate in the collaborative sensing process. On the
ering current strategies of other IoT nodes (other players)),
contrary, increasing requirement of Qd will only require few
the convergence of proposed potential game is guaranteed to
more IoT nodes to participate in collaborative sensing. It is
unique Nash equilibrium. The potential function in (18) is now
important to assign channels for sensing for each IoT node
exploited to show the uniqueness of equilibrium.
while considering its sensing capabilities. Otherwise, more IoT
In this game theoretical framework, each player (IoT node) nodes may be required to sense the channel to achieve required
executes the three-step cycle shown in Fig. 5. Qd which results in high energy consumption. Thus, optimal
• In the sensing step, each player interacts with the central allocation of users for multi-band sensing is required to reduce
entity which has spectrum sensing results for available the energy consumption for a given number of channels and
channels and acquires their link gains. Qd .
• For the analysis and reasoning step, each player maxi- Fig. 7 shows energy consumption versus number of IoT
mizes its utility by choosing appropriate strategies while nodes for different number of channels to be sensed. We
considering the current strategies of other players. Nash compared the proposed scheme with basic multi-band [16]
Equilibrium is considered as a steady state point after and sensor selection for cooperative spectrum sensing (CSS)
which the players cannot change their strategies to max- [25]. As mentioned in Section II, basic multi-band approach
imize their utilities. It has been demonstrated that for randomly select channels to be sensed for each IoT node
the potential function in (18), the players reach Nash subject to the constraint on probability of detection only.
Equilibrium after a certain number of iterations. For an in- Sensor selection for CSS in which sensors are selected to
depth study of the Nash Equilibrium and the convergence participate in CSS with an objective to minimize energy
of this game, the reader is referred to [42], [44]. consumption in spectrum sensing. We consider three scenarios

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
10

the resources are allocated based on the interference factor


and the channel gain only [48] and a scheme that consider
0.46
a variant of (17), i.e., without economic and environment
0.5
0.44
impact (CLRS∗ ). In case of JRANSS, we consider 25% of total
0.42
IoT nodes in each service class because it allocate resources
Energy consumption (J)

0.45
0.4 without considering QoS service classes given in Table IV.
0.4
0.38 It is worth mentioning that all the simulation parameters
0.35
0.36 including channel gains, path loss coefficient, and other factors
0.3
0.34 are chosen randomly for each channel. The performance
0.25
0.32
matrices considered in the simulations are throughput, energy
0.2 efficiency, delay, reliability, economic cost, and environment
0.3
50 impact.
45 0.96 0.28
Fig. 8 (a) compares aggregate throughput versus the number
40 0.94 0.26
35 0.92
of IoT nodes for four service classes in CLRS, JRANSS,
Number of channels 30 0.9 Targetted Q d
and CLRS∗ . It should be noted that the higher throughput
in the cases of service class 1 and 3 is due the fact that, in
these classes, there is a stringent requirement of high data
Fig. 6. Energy consumed in spectrum sensing versus number of channels C rate for all IoT nodes. However, other classes achieve less
and targeted global probability of detection Qd .
aggregate throughput since the high data rate attribute is low.
When compared to JRANSS which allocate resources based
for IoT nodes, i.e., 300, 350, and 400 with varying number of on the interference factor and the channel gains only, the
channels (i.e., 30, 40, and 50) to be sensed in each scenario. It throughput of proposed CLRS for service class 2 is compa-
is evident that the energy consumed for the spectrum sensing rable. When compared to the scheme that exclude economic
increases with the increase in number of IoT nodes and and environment impact, there is only a minute difference of
number of channels. However, energy consumption for the aggregate throughput in case of service class 1 and 3. However,
proposed multi-band spectrum sensing is lower than both the aggregate throughput is high for service class 4 in CLRS∗
sensor selection CSS and the basic multi-band scheme. This since CLRS can compromise on throughput and give more
is because the former assigned multiple bands for spectrum weight to the economic and environment impact. This clearly
sensing to IoT nodes based on their capabilities to ensure shows the effectiveness of the proposed scheme to meet data
required Qd and Qf , whereas in later schemes sensor nodes rate requirements for IoT applications.
are selected for CSS to satisfy spectrum sensing requirements Fig. 8 (b) shows energy-efficiency comparison for the
regardless of device capabilities. This illustrates the superiority aforementioned schemes. Clearly, service class 1 and 3 has
of multi-band spectrum sensing approach proposed in this the highest energy-efficiency for both CLRS and CLRS∗ .
work. However, in case of class 2 and 4, when there is a stringent
requirements of reliability and low attribute for environment
cost, the system will choose channels without considering
B. CLRS for Resource Allocation energy footprint in case of CLRS. Moreover, energy-efficiency
Here, we present the performance of the proposed CLRS for JRANSS is also comparable with the service class 2 for
through computer simulations. We consider four service CLRS. When compared with CLRS∗ , the CLRS offers high
classes given in Table IV, i.e., HRRt, LRRt, LRNRt, and energy-efficiency for all classes except service class 4. This is
HRNRt. Let SC be the set of service classes in the IoT due to the significant importance of EC and EN as compared
system. It is important to note that the performance of CLRS to other attributes in CLRS for the service class 4 (as given
is not dependent on the size of SC . Each class sc ∈ SC in Table IV).
has the following QoS parameters: data rate (RD ), latency Similarly, Fig. 8 (c) and (d) show delay and reliability
(L), reliability (RE ), economic impact (EC ), and environment (packet delivery ratio) versus number of IoT nodes. It is
impact (EN ). For illustration purposes, it is assumed that evident that the delay of class 1 and 3 is smaller than the
cellular, WLAN, and ZiBee channels are available for data other service classes in case of CLRS and CLRS∗ . However,
transmission. The simulation parameters for cellular bands delay for CLRS is equal to or less than the CLRS∗ for both
such as maximum transmit power of base station is 46 dBm, service classes 1 and 3. For service classes 2 and 4, the delay
system bandwidth of 10MHz with 50 resource blocks of is higher in case of proposed CLRS both service classes have
180kHz are chosen according to 3GPP specifications given strict requirements on EC and EN which force them to choose
in [45]. The simulation parameters for WLAN are according channel/ network with less economic price and environment
to WiFi such as maximum transmit power is 30dBm, access impact and less importance of latency attribute. Delay in the
point cell coverage is 10m, etc. (detailed parameters can be case of JRANSS is higher than all service classes except
found in [46]). Similarly, simulation parameters for ZigBee are service class 4 because JRANSS allocate resources based on
chosen according to specifications given in [47]. The CLRS is the interference factor and the channel gain only (regardless of
compared with user optimized scheme called joint resource the service classes). On the other hand, reliability for service
allocation and network selection scheme (JRANSS), where classes 1 and 2 is the highest for both CLRS and CLRS∗

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
11

0.9
0.8
Energy consumption (J)

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
300 350 400
Number of IoT nodes
Proposed multiband, C=30 Basic multiband, C=30 Sensor selection CSS, C=30
Fig. 7. Energy consumed versus number of IoT nodes for different number of channels to be sensed.
Proposed multiband, C=40 Basic multiband, C=40 Sensor selection CSS, C=40
Proposed multiband, C=50 Basic multiband, C=50 Sensor selection CSS, C=50
as the reliability attribute is important. Reliability for service VI. C ONCLUSIONS
classes 3 and 4 follows the specifications of QoS given in
Table IV. Once again, the proposed CLRS performs better (it
allocates resources according to the priority of service classes) In this paper, we have proposed a spectrum sharing frame-
than JRANSS in which the reliability is similar to the one for work for IoT in cognitive networks. The framework consists
service class 3 in case of CLRS. of multi-band spectrum access and cross-layer reconfiguration
for resource allocation. In multi-band spectrum access, we
formulated a problem for optimal number of channels to
Figs. 9 and 10 compare the economic price and environment
be assigned to each IoT node for spectrum sensing while
cost of proposed CLRS, JRANSS, and CLRS∗ , respectively.
satisfying the constraints on all channels must be sensed by
The economic price is NP · EC k and the environmental
at least one IoT node, targeted global probability of detection
impact is NP · EN k for providing services to NP IoT nodes,
(Qd ), and targeted global probability of false alarm (Qf ). We
where EC k and EN k are considered as service usage fee
have studied the effect of the number of channels, Qd , and the
in dollars and power consumed in watts corresponding to
number of IoT nodes on energy consumed in spectrum sensing
particular service class of the k-th IoT node, respectively. It
process. In addition, we proposed a cross-layer reconfigura-
can be seen that the economic price and environment impact
tion scheme (CLRS) for resource allocation in cognitive 5G
is high when there are stringent requirements of data rate
networks. We formulated a utility function which involves dif-
and reliability. Moreover, it can be seen that CLRS has less
ferent QoS parameters including data rate, latency, reliability,
economic price and environment impact when compared to
economic cost, and environment impact. The potential game
JRANS (for service classes 3 and 4) and CLRS∗ (for all
approach is used in CLRS scheme, which allocates resources
service classes) as these schemes do not consider economic
based on the QoS requirements of IoT nodes. Simulation
and environment impact for resource allocation.
results exhibit the performance of proposed framework for IoT
system. It is emphasized that allocating an optimal number
The complexity of CLRS can be analyzed for the given of channels to be sensed for multi-band spectrum sensing
number of players (NP ), each with possible actions (A). The can significantly reduce energy consumption in comparison
average number of iterations for the best response strategy runs with existing spectrum sensing approaches. Similarly, results
before reaching Nash equilibrium is less than log(NP ) + e$ , for CLRS show that resources are allocated while satisfying
where $ is the Euler constant [49]. For the communication QoS requirements for each service class. Further, CLRS has
overhead (due to information exchange required by the game- less economic price and environment impact for particular
theoretical approach), it can be significantly reduced if IoT service class when compared with the existing schemes. The
nodes use existing common control channels. This complexity relatively high complexity cost of CLRS is distributed among
cost is fully justified given the significant performance im- all IoT nodes and advocated by the performance improvements
provement that CLRS offers in terms of throughput, energy- offered by CLRS. Future work can involve mobile IoT nodes
efficiency, delay, reliability as well as economic and environ- in the proposed framework, where channel fading changes over
ment impact. the time.

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
12

×10 9
80 6
CLRS SC 1 CLRS SC 1
* *
CLRS SC 1 CLRS SC 1
70
CLRS SC 2 5 CLRS SC 2
* *
CLRS SC 2 CLRS SC 2
Aggregated throughput (Gbps)

Energy Efficiency (bits/joule)


60 CLRS SC 3 CLRS SC 3
CLRS * SC 3 4 CLRS * SC 3
50 CLRS SC 4 CLRS SC 4
CLRS * SC 4 CLRS * SC 4
JRANSS JRANSS
40 3

30
2

20

1
10

0 0
50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400
Number of IoT nodes Number of IoT nodes

(a) (b)

3000 0.9
CLRS SC 1 CLRS SC 1
*
CLRS SC 1
*
CLRS SC 1 0.44
200 0.8
CLRS SC 2 CLRS SC 2
2500 *
CLRS * SC 2 0.43

Reliability (packet delivery ratio)


CLRS SC 2
0.7
CLRS SC 3 100 CLRS SC 3
* *
CLRS SC 3 CLRS SC 3 0.42
2000 370 380 390 0.6
CLRS SC 4 CLRS SC 4
240 245 250
Delay (ms)

* *
CLRS SC 4 CLRS SC 4
0.5 JRANSS
JRANSS
1500
0.4

1000 0.3

0.2
500
0.1

0 0
50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400
Number of IoT nodes Number of IoT nodes

(c) (d)
Fig. 8. (a) Aggregated throughput, (b) energy efficiency, (c) delay, and (d) reliability (packet delivery ratio) versus varying number of IoT nodes for different
service classes.

1500 1100 776


CLRS SC 1 CLRS SC 1
1118 774
CLRS SC 1 * 1000 CLRS SC 1 *

CLRS SC 21116 CLRS SC 2 772


* 900
CLRS SC 2 CLRS * SC 2 770
Economic cost (Normalized)

CLRS SC 31114 CLRS SC 3 768


Environment cost (Watts)

800
*
1000 CLRS SC 3 290 CLRS * SC 3 244245246
CLRS SC 4 700 CLRS SC 4
*
CLRS SC 4 CLRS * SC 4
JRANSS 600
JRANSS

500

400
500
300

200

100

0 0
50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400
Number of IoT nodes Number of IoT nodes

Fig. 9. Economic cost versus the number of IoT nodes. Fig. 10. Environment impact versus the number of IoT nodes.

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
13

R EFERENCES [24] R. Zhu, X. Zhang, X. Liu, W. Shu, T. Mao, and B. Jalaian, “ERDT:
Energy-efficient reliable decision transmission for intelligent cooperative
[1] “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast spectrum sensing in industrial IoT,” IEEE Access, vol. 3, pp. 2366–2378,
Update, 2015–2020,” tech. rep., Cisco, San Jose, CA, USA, Feb. 2015. Nov. 2015.
[25] A. Ebrahimzadeh, M. Najimi, S. M. H. Andargoli, and A. Fallahi,
[2] A. Lo, Y. W. Law, and M. Jacobsson, “A cellular-centric service
“Sensor selection and optimal energy detection threshold for efficient
architecture for machine-to-machine (M2M) communications,” IEEE
cooperative spectrum sensing,” IEEE Transactions on Vehicular Tech-
wireless communications, vol. 20, no. 5, pp. 143–151, Oct. 2013.
nology, vol. 64, no. 4, pp. 1565–1577, Apr. 2015.
[3] T. Taleb and A. Kunz, “Machine type communications in 3GPP net- [26] Z. Qin, Y. Gao, M. D. Plumbley, C. G. Parini, and L. G. Cuthbert,
works: potential, challenges, and solutions,” IEEE Communications “Efficient compressive spectrum sensing algorithm for M2M devices,”
Magazine, vol. 50, no. 3, Mar. 2012. in IEEE Global Conference on Signal and Information Processing
[4] A. Rico-Alvarino, M. Vajapeyam, H. Xu, X. Wang, Y. Blankenship, (GlobalSIP), pp. 1170–1174, Dec. 2014.
J. Bergman, T. Tirronen, and E. Yavuz, “An overview of 3GPP enhance- [27] Y. Ma, Y. Gao, and C. G. Parini, “Sub-Nyquist rate wideband spectrum
ments on machine to machine communications,” IEEE Communications sensing over TV white space for M2M communications,” in IEEE 16th
Magazine, vol. 54, no. 6, pp. 14–21, Jun. 2016. International Symposium on a World of Wireless, Mobile and Multimedia
[5] 3GPP, “Cellular system support for ultra-low complexity and low Networks (WoWMoM), pp. 1–6, Jul. 2015.
throughput Internet of Things (CIoT),” tech. rep., TR45.820, Dec. 2015. [28] Y. Abdi and T. Ristaniemi, “Optimal Multiband Cooperative Spectrum
[6] G. Americas, “LTE and 5G technologies enabling the Internet of Sensing with Random Interruptions,” in IEEE Global Communications
Things,” tech. rep., Dec. 2016. Conference (GLOBECOM), pp. 1–7, Dec. 2016.
[7] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, [29] M. Soltani, T. Baykas, and H. Arslan, “On Reducing Multiband Spec-
“Five disruptive technology directions for 5G,” IEEE Communications trum Sensing Duration for Cognitive Radio Networks,” in IEEE 84th
Magazine, vol. 52, no. 2, pp. 74–80, Feb. 2014. Vehicular Technology Conference (VTC-Fall), pp. 1–5, Sep. 2016.
[8] T. Maksymyuk, M. Kyryk, and M. Jo, “Comprehensive spectrum [30] J. Plachy, Z. Becvar, and E. C. Strinati, “Cross-layer approach enabling
management for heterogeneous networks in LTE-U,” IEEE Wireless communication of high number of devices in 5G mobile networks,” in
Communications, vol. 23, no. 6, pp. 8–15, Dec. 2016. IEEE 11th International Conference on Wireless and Mobile Computing,
[9] M. R. Palattella, M. Dohler, A. Grieco, G. Rizzo, J. Torsner, T. Engel, Networking and Communications (WiMob), (Abu Dhabi), pp. 809–816,
and L. Ladid, “Internet of Things in the 5G era: Enablers, architecture, Oct. 2015.
and business models,” IEEE Journal on Selected Areas in Communica- [31] N. Hasan, M. Ali, A. Barradas, and N. Correia, “Cross-layer optimiza-
tions, vol. 34, no. 3, pp. 510–527, Feb. 2016. tion for reliability improvement of data delivery in 6LoWPAN-based
[10] Y.-W. Ma, J.-L. Chen, Y.-S. Lee, and H.-Y. Chang, “GreenIoT Architec- networks,” in 14th Annual Mediterranean Ad Hoc Networking Workshop
ture for Internet of Things Applications,” KSII Transactions on Internet (MED-HOC-NET), (Vilamoura), pp. 1–7, Jun. 2015.
and Information Systems (TIIS), vol. 10, no. 2, pp. 444–461, Feb. 2016. [32] P. Du and G. Roussos, “Adaptive communication techniques for the
[11] W. Ejaz and M. Ibnkahla, “Machine-to-machine communications in cog- Internet of Things,” Journal of Sensor and Actuator Networks, vol. 2,
nitive cellular systems,” in IEEE International Conference on Ubiquitous no. 1, pp. 122–155, Mar. 2013.
Wireless Broadband (ICUWB), pp. 1–5, Oct. 2015. [33] A. M. da Costa Souza and J. R. de Almeida Amazonas, “ A New
[12] C.-X. Wang, F. Haider, X. Gao, X.-H. You, Y. Yang, D. Yuan, H. M. Internet of Things Architecture with Cross-Layer Communication,”
Aggoune, H. Haas, S. Fletcher, and E. Hepsaydir, “Cellular architecture in The Seventh International Conference on Emerging Networks and
and key technologies for 5G wireless communication networks,” IEEE Systems Intelligence (EMERGING), (Nice, France), pp. 1–6, IARIA, Jul.
Communications Magazine, vol. 52, no. 2, pp. 122–130, Feb. 2014. 2014.
[13] W. Lu, Y. Zhu, M. Wang, H. Peng, and X. Liu, “Sum Transmission [34] A. A. Frohlich, A. Massayuki Okazaki, R. Vieira Steiner, P. Oliveira,
Rate Maximization Based Cooperative Spectrum Sharing with Both and J. Everson Martina, “ Cross-layer Approach to Trustfulness in
Primary and Secondary QoS-Guarantee,” KSII Transactions on Internet the Internet of Things,” in IEEE 16th International Symposium on
and Information Systems (TIIS), vol. 10, no. 5, pp. 2015–2028, 2016. Object/Component/Service-Oriented Real-Time Distributed Computing
[14] Q. D. La, Y. H. Chew, and B. H. Soong, “An interference-minimization (ISORC), (Paderborn), pp. 1–8, Jun. 2013.
potential game for OFDMA-based distributed spectrum sharing sys- [35] V. Chandrasekhar, J. G. Andrews, and A. Gatherer, “Femtocell networks:
tems,” IEEE Transactions on vehicular technology, vol. 60, no. 7, a survey,” IEEE Communications magazine, vol. 46, no. 9, pp. 59–67,
pp. 3374–3385, Sep. 2011. Sep. 2008.
[15] N. U. Hasan, W. Ejaz, N. Ejaz, H. S. Kim, A. Anpalagan, and M. Jo, [36] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,
“Network selection and channel allocation for spectrum sharing in 5G O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities and
heterogeneous networks,” IEEE Access, vol. 4, pp. 980–992, Feb. 2016. challenges with very large arrays,” IEEE Signal Processing Magazine,
vol. 30, no. 1, pp. 40–60, Jan. 2013.
[16] G. Hattab and M. Ibnkahla, “Multiband spectrum access: great promises
[37] F. F. Digham, M.-S. Alouini, and M. K. Simon, “On the energy
for future cognitive radio networks,” Proceedings of the IEEE, vol. 102,
detection of unknown signals over fading channels,” IEEE Transactions
no. 3, pp. 282–306, Mar. 2014.
on communications, vol. 55, no. 1, pp. 21–24, Jan. 2007.
[17] M. Zheng, L. Chen, W. Liang, H. Yu, and J. Wu, “Energy-efficiency [38] I. F. Akyildiz, B. F. Lo, and R. Balakrishnan, “Cooperative spectrum
Maximization for Cooperative Spectrum Sensing in Cognitive Sensor sensing in cognitive radio networks: A survey,” Physical communication,
Networks,” IEEE Transactions on Green Communications and Network- vol. 4, no. 1, pp. 40–62, Mar. 2011.
ing, vol. 1, no. 1, pp. 29–39, Mar. 2017. [39] N. Thakoor and J. Gao, “Branch-and-bound for model selection and its
[18] A. El-Mougy, M. Ibnkahla, G. Hattab, and W. Ejaz, “Reconfigurable computational complexity,” IEEE Transactions on Knowledge and Data
wireless networks,” Proceedings of the IEEE, vol. 103, no. 7, pp. 1125– Engineering, vol. 23, no. 5, pp. 655–668, May 2011.
1158, Jul. 2015. [40] F. Mansourkiaie, L. S. Ismail, T. M. Elfouly, and M. H. Ahmed,
[19] C. Ju and Q. Shao, “Global Optimization for Energy Efficient Re- “Maximizing lifetime in wireless sensor network for structural health
source Management by Game Based Distributed Learning in Internet of monitoring with and without energy harvesting,” IEEE Access, vol. 5,
Things,” KSII Transactions on Internet and Information Systems (TIIS), pp. 2383–2395, Feb. 2017.
vol. 9, no. 10, pp. 3771–3788, Oct. 2015. [41] M. Voorneveld, “Best-response potential games,” Economics letters,
[20] M. Ibnkahla, Adaptation and Cross Layer Design in Wireless Networks. vol. 66, no. 3, pp. 289–295, Mar. 2000.
CRC press, 2008. [42] D. Monderer and L. S. Shapley, “Potential games,” Games and economic
[21] S. Okdem, “A cross-layer adaptive mechanism for low-power wireless behavior, vol. 14, no. 1, pp. 124–143, 1996.
personal area networks,” Computer Communications, vol. 78, pp. 16–27, [43] Y. Zhao, S. Mao, J. O. Neel, and J. H. Reed, “Performance evalua-
Mar. 2016. tion of cognitive radios: Metrics, utility functions, and methodology,”
[22] C. Han, J. M. Jornet, E. Fadel, and I. F. Akyildiz, “A cross-layer Proceedings of the IEEE, vol. 97, no. 4, pp. 642–659, Apr. 2009.
communication module for the Internet of Things,” Computer Networks, [44] W. H. Sandholm, “Potential games with continuous player sets,” Journal
vol. 57, no. 3, pp. 622–633, Feb. 2013. of Economic Theory, vol. 97, no. 1, pp. 81–108, Mar. 2001.
[23] Y. Gao, Z. Qin, Z. Feng, Q. Zhang, O. Holland, and M. Dohler, “Scalable [45] E. U. T. R. Access, “Further advancements for E-UTRA physical layer
& reliable IoT enabled by dynamic spectrum management for M2M in aspects,” 3GPP TR 36.814, Tech. Rep., 2010.
LTE-A,” IEEE Internet of Things Journal, vol. 3, no. 6, pp. 1135–1145, [46] K. M. Metwaly, K. G. Seddik, and M. Y. ElNainay, “A Cooperative
Dec. 2016. Scheme for the Coexistence of the LTE and WiFi Systems,” in IEEE

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
14

Wireless Communications and Networking Conference (WCNC), pp. 1–


6, IEEE, May 2017.
[47] Z. Zhao, X. Wu, X. Zhang, J. Zhao, and X.-Y. Li, “ZigBee vs WiFi: Φ(sk , s−k ) = U (sk , s−k ) + X(s−k ), (21)
Understanding issues and measuring performances of their coexistence,”
in IEEE International Performance Computing and Communications where X(s−k ) = ∆1 (s−k )+∆2 (s−k )+∆3 (s−k )+∆4 (s−k )+
Conference (IPCCC), pp. 1–8, Dec. 2014.
∆5 (s−k ) + ∆6 (s−k ) + ∆7 (s−k ). This function is independent
[48] I. W. Mustika, K. Yamamoto, H. Murata, and S. Yoshida, “Potential
game approach for self-organization scheme in open access heteroge- of k-th IoT node. Further, if the strategy of k-th IoT node
0
neous networks,” in Sixth International ICST Conference on Cognitive changes from sk to sk then (21) can be written as:
Radio Oriented Wireless Networks and Communications (CROWN-
COM), pp. 216–220, Jun. 2011.
0 0
[49] S. Durand and B. Gaujal, “Average complexity of the best response
algorithm in potential games,” in 17éme conférence Roadef, 2016.
Φ(sk , s−k ) = U (sk , s−k ) + X(s−k ). (22)
From (21) and (22),
A PPENDIX A
0 0
P OTENTIAL FUNCTION Φ(S) IS E XACT P OTENTIAL Φ(sk , s−k ) − Φ(sk , s−k ) = U (sk , s−k ) − U (sk , s−k ), (23)
F UNCTION
which holds (16).
This proof shows that the potential function in (18) for
CLRS is exact potential game. The potential function for
CLRS in (18) can be decomposed into seven parts as follows:

Φ(S) = ∆1 (S) + ∆2 (S) + ∆3 (S) + ∆4 (S)+ (19)


∆5 (S) + ∆6 (S) + ∆7 (S),
where
P N
1X b,a
∆1 (S) = ωkRD RD
2 k
k=1
NP
1 X
∆2 (S) = − ωkL Lb,a
k
2
k=1
NP
1X b,a
∆3 (S) = ωkRE RE
2 k
k=1
NP
1 X
∆4 (S) = ωkRP hb,a b,a
k,c Pk
2
k=1
NP
1 X b,a b,a
∆5 (S) = − ωkEC EC Bk
2 k
k=1
P N
1X b,a b,a
∆6 (S) = − ωkEN EN Pk
2 k
k=1
NP NP
1 X X
∆7 (S) = − hb,a
m,c σck ,cm
2
k=1 m=1,m6=k Waleed Ejaz (S’12-M’14-SM’16) received the
NP
X B.Sc. and M.Sc. degrees in Computer Engineering
− hb,a
k,c σcm ,ck .
from the University of Engineering and Technol-
ogy, Taxila, Pakistan and the National University of
m=1,m6=k Sciences and Technology, Islamabad, Pakistan, and
the Ph.D. degree in Information and Communication
The term ∆1 (sk , s−k ) can be decomposed as Engineering from Sejong University, Republic of
Korea, in 2014. Since 2014, he held academic and
  research positions at Carleton University, Ryerson
NP
1  RD b,a X  University, and Queen’s University in Canada. His
b,a
∆1 (sk , s−k ) = ω RDk + ωqRD RD . (20) current research interests include Internet of Things
2 k q
 (IoT), energy harvesting, 5G cellular networks, and mobile cloud computing.
q=1,q6=k
His professional activities include an Associate Editor of the IEEE Com-
PNP RD b,a munications Magazine, IEEE Canadian Journal of Electrical and Computer
It is evident that ∆1 (s−k ) = 12 q=1,q6 =k ωq RDq is Engineering, and the IEEE ACCESS. In addition, he is handling several
independent of the strategy of k-th player. Similarly, ∆2 (s−k ), special issues in prestigious journals as well as in organizing committees of
∆3 (s−k ), ∆4 (s−k ), ∆5 (s−k ), ∆6 (s−k ), and ∆7 (s−k ) are several major IEEE conferences and workshops. Dr. Ejaz completed certificate
courses on Teaching and Learning in Higher Education from the Chang School
not depending on the corresponding k-th IoT node strategy. at Ryerson University. He is a registered Professional Engineer (P.Eng.) of the
Specifically, this can be written as: province of Ontario, Canada.

2327-4662 (c) 2017 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2775959, IEEE Internet of
Things Journal
15

Mohamed Ibnkahla is a Full Professor at the


Department of Systems and Computer Engineering,
Carleton University, Ottawa, Canada. He holds the
Cisco Research Chair in Sensor Technology for the
Internet of Things (IoT) and NSERC/Cisco Indus-
trial Research Chair in Sensor Networks for the
Internet of Things. He obtained the Ph.D. degree
and the Habilitation a Diriger des Recherches degree
(HDR) from the National Polytechnic Institute of
Toulouse (INPT), Toulouse, France, in 1996 and
1998, respectively. He obtained an Engineering de-
gree in Electronics (1992) and a Diplome dâEtudes Approfondies degree
(equivalent to MSc) in Signal and Image Processing (1992) from INPT. Prior
to joining Carleton University, he has been a Professor at the Department of
Electrical and Computer Engineering, Queenâs University, Kingston, Canada,
from 2000 to 2015. Over the past 10 years, he has been conducting multi-
disciplinary research projects designing, developing and deploying advanced
wireless sensor networks (WSN) for real-world Internet of Things applications
including: environment monitoring, smart cities, intelligent transportation
systems, food traceability, water quality monitoring, smart homes, health
care, smart grid and public safety. He published 6 books and more than 70
peer-reviewed journal papers and book chapters, 20 technical reports, 110
conference papers, and 4 invention disclosures. He is the author of Wireless
Sensor networks: A Cognitive perspective, CRC Press - Taylor and Francis,
2012 and Cooperative Cognitive Radio Networks: The Complete Spectrum
Cycle, CRC Press - Taylor and Francis, 2015. In the past 5 years he gave
more than 30 keynote talks and invited seminars. He received the Leopold
Escande Medal, 1997, France, and the Premierâs Research Excellence Award,
Canada, 2001. He is the joint holder of 5 Best Paper Awards.

2327-4662 (c) 2017 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.

You might also like