Spectrum Assignment in Cognitive Radio Networks for ... - IEEE Xplore

0 downloads 0 Views 2MB Size Report
Abstract—The expected widespread usage of Internet-of-. Things (IoT) devices and the associated diversity of applica- tions will place extra pressure on network ...
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.2018.2817339, IEEE Internet of Things Journal 1

Spectrum Assignment in Cognitive Radio Networks for Internet-of-Things Delay-sensitive Applications under Jamming Attacks Haythem Bany Salameh, Senior Member, IEEE,, Sufyan Almajali, Moussa Ayyash, Senior Member, IEEE, and Hany Elgala

Abstract—The expected widespread usage of Internet-ofThings (IoT) devices and the associated diversity of applications will place extra pressure on network resources including bandwidth availability. In networking within IoT, spectrum scarcity requires the consideration of the adaptive cognitive radio (CR) technology to achieve interference-free and on-demand IoT solutions for a number of applications. However, CR networks share the same security weaknesses of legacy wireless networks. Jamming is considered a common attack, where jammers can attack networks in a proactive or reactive approach. In this paper, we study the channel assignment problem under both proactive and reactive jamming attacks. Specifically, we propose a novel probabilistic-based channel assignment mechanism that aims at minimizing the invalidity ratio of CR packet transmissions subject to delay constraints. The proposed scheme exploits the statistical information of licensed primary users activities, fading conditions and jamming attacks over idle channels to provide communicating CR IoT devices with the most secured channels of lowest invalidity ratios. Simulation results show that our proposed security, availability, and quality-aware channel assignment algorithm can significantly improve network performance.

I. I NTRODUCTION In current cellular and wireless networks, limited types of devices such as smart-phones, tablets, and laptops are connected. However, the Internet-of-Things (IoT) is a network of interconnected objects including environmental sensors, health monitoring devices, smart meters, home appliances, autonomous cars and many others [1], [2]. In essence, IoT made it possible to enable machine-to-machine and humanto-machine communications [3]. In the IoT paradigm, new applications areas are possible including precision agriculture, and smart manufacturing. The different hardware, software, and networking components must be designed, and refined to work together in supporting the development of IoT solutions. Future IoT deployments introduce new requirements and one H. Bany Salameh is with the Department of Telecommunications Engineering, Yarmouk University, Jordan, e-mail:[email protected]. S. Almajali is with the Computer Science Department, Princess Sumaya University for Technology, Jordan. M. Ayyash is with the Department of Information Studies, Chicago State University, USA. H. Elgala is with the Computer Engineering Department, University at Albany-SUNY, USA. A limited subset of initial results was preliminary presented at the Fourth International Conference on Software Defined Systems (SDS) 2017, Spain. Copyright (c) 2012 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to [email protected].

primary requirement is security. Many IoT applications directly interact with daily human tasks and in different domains such as health-care applications. Lacking a high-security level could expose users/devices to new threats that are designed to target IoT applications [4]. The envisioned IoT devices are mainly interconnected through wireless communication technologies, battery operated, and located in remote and scattered locations. The high bandwidth demand to serve massive numbers of IoT devices may lead to spectrum scarcity for IoT applications. Cognitive radio networks (CRNs) is a wireless technology that allows bandwidth utilization in a more efficient way in comparison with traditional wireless networks. CRNs use the concept of dynamic spectrum management for allocating the radiofrequency bands. CRN users who are unlicensed share the spectrum with the licensed primary radio (PR) users. CRNs allow wireless devices to have dynamic access to the entire available spectrum based on certain rules and conditions. For large-scale IoT deployments, CRNs could help alleviating the issue of spectrum scarcity. The use of CRNs has been advocated as a typical communication solution for many IoT applications [5]–[8]. In such case, an IoT device works as a secondary user (SU) and opportunistically utilizes same frequency bands owned by PR users. This provides the means and capabilities for IoT devices to access higher bandwidth. Several IoT applications require bandwidth access in a periodic way, where an IoT device sends a small number of packets to update a remote server or a cloud system. Examples of such IoT applications are remote health monitoring applications, smart city systems, and remote farm control systems. CRNs offer ease of spectrum access and a cost-effective solution for bandwidth access for such IoT applications and services. A. Motivation Confidentiality, integrity, and availability (known as the security triad CIA) represent the backbone of information security [9]. Many research projects focus their attention on the first two factors of the security triad and overlook availability. The availability factor represents the ability of a network, a device, a server or an application to deliver the requested services to the authorized legitimate users. Examples of such services include providing access to network bandwidth (channel assignment), providing access to a resource such as a file or a database record, or reporting data from

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 2

sensors. Unavailable services imply the denial of a legitimate user from accessing the requested services and information. In some IoT applications, this could impose a human life threat such as in health-care applications that remotely track patients’ health status. In other IoT applications, unavailable services could cause financial loss such as IoT applications for smart cities designed to reduce power consumption at the city level. In IoT-based wireless CRNs, the unavailability threat is defined by the denial of a CR IoT device from successfully delivering/receiving data packets by disrupting the CR transmissions over the selected channels and preventing secondary IoT devices from effectively utilizing spectrum resources (i.e., frequency channels). Several security attacks can threaten CRNs and impact CR service availability. Attacks include primary user emulation (PUE) attack, denial of service attack (DoS), spectrum sensing data falsification attack (SSDF), spoofing attack, and jamming attack [10], [11]. The Jamming attack is one of the most common and dangerous types of attacks in CRNs and legacy wireless networks [10]–[12]. Jammers corrupt the communication by sending jamming signals proactively or reactively, resulting in either damaging ongoing transmitted packets or CR users denial from sending data packets during the jamming period [13]. Jamming severely reduces the bandwidth availability in CRNs [12]. Several defense approaches are available to protect against jamming attacks. Approaches include detection, prevention, and mitigation. In general, these approaches require the use of additional network resources. Various spectrum assignment schemes for CRNs have been proposed in literature (e.g., [14]–[18]). Most of them were designed without considering the security attacks or assuming that defense approaches (i.e., detection, prevention, or mitigation) are in place. In practice, employing defense approaches require additional resources such as power, complexity and spectrum. On the other hand, when security-unaware channel assignment schemes are employed in a CRN that are vulnerable to jamming attacks, the number of forced-terminated CR packets will increase due to jamming attacks, resulting in severe degradation in network availability, poor network throughput, and wasted spectrum opportunities. Hence, the solution for this problem is to design new security-aware spectrum sharing algorithms that mitigate jamming attacks without consuming additional new network resources or power. B. Contribution In this paper, we present a new algorithm for dealing with jamming attacks in IoT-based CRNs. We consider both proactive and reactive jamming attacks. Our algorithm assigns channels to CR IoT devices while taking into consideration the security requirements of IoT-based CRNs. The jamming attacks are mitigated without consuming additional resources or power. For each CR communicating pair, we develop a mathematical model to represent the packet-invalidity ratio, which is defined as the probability that the packet transmission delay is greater than a predefined delay requirement. Specifically, we derive expressions for the packet-invalidity ratio for a given delay requirement based on probabilistic models of

jammer activities, PR activities and CR link quality conditions. The invalidity ratio is calculated using three factors: CRN linkquality, jamming level, and PR channel availability durations. This new probabilistic channel assignment mechanism seeks to enhance the CRN throughput by selecting the channel with the minimum packet-invalidity ratio for each CR transmission. The proposed channel assignment allows for selecting the most secure channel for each CR communicating pair while considering the regular PR activities and CRN fading conditions. To the best of our knowledge, our proposed scheme is the first adaptive spectrum allocation algorithm that aims at improving the CRIoT performance while jointly accounting for jamming attacks, PR activities and CR link quality conditions. The most relevant work to this paper is the MaXPoS scheme [19], where both PR activities and CR channel quality are considered when assigning channels. We conduct simulation experiments for a dynamic IoT network based on CRN with delay requirements. The results show that the proposed scheme that considers security, link-quality, and network availability can significantly improve network performance in terms of throughput. C. Organization The rest of this paper is structured as follows. In Section II, we describe our network model. The mathematical analysis of transmission invalidity is presented in Section III. Section IV presents our new security-aware channel assignment scheme after defining the spectrum assignment problem. Section V presents the performance evaluation of our proposed approach. Section VI presents the related work. Finally, we conclude the paper in Section VII. II. R ELATED W ORK Different research projects presented different approaches for handling jamming attacks in CRNs. Methods included detecting, preventing, and counter-measuring attacks. Prevention methods rely on duplicating network resources and adding redundancy at the level of links, channels, devices, and servers. One main approach for detecting jamming attacks in CRNs is to use Intrusion Detection Systems (IDSs) which are either signature-based or anomaly-based [10]. Signature-based produces good results for jamming attacks of specific behaviors but suffers when CRNs are attacked by a slightly modified jamming behavior. The anomaly-based approach works better for dealing with jamming attacks of unknown behavior. However, anomaly-based detection suffers from the issue of inaccurate classification (i.e., false positive and false negative decisions). In [10], the authors presented an anomaly-based solution for detecting jamming attacks. This solution attempts at identifying the abnormal behavior of the attacks. To achieve this, the proposed solution in [10] performs learning and detection phases. A variety of CRN features are used to study and profile the behavior of normal network operations. Examples of features include carrier sense thresholds, protocol behavior, traffic flow, primary user access time, error bit rate, number of unsuccessful attempts, ambient noise levels, signal strength (SS), signal-to-noise ratio (SNR), and packet delivery ratio (PDR). In [10], the authors relied on two main features:

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 3

PDR and SS. During the learning phase, the authors studied the normal behavior of the CRNs under no Jamming attacks in terms of PDR and SS. During detection phase, jamming attacks are present and a comparison against the baseline profile is used to indicate a jamming attack. The author presented only a detection solution with simple simulation results. No details about the practical implementations. In [11], the authors presented another anomaly-based IDS system for dealing with general CRN security attacks including jamming attacks. The solution is more comprehensive and focuses on detection only. In [11], the proposed solution includes several modules such as online/offline monitoring module, feature selection, anomaly behavior analysis, and the action module. The work in [11] significantly outperforms the one in [10] in terms of the effectiveness of the feature-selection process. The solution is resources intense, as it requires the nodes to participate in the monitoring and data collection process. This makes it unsuitable for IoT networks due to their energyconstrained nature. In [20], the authors proposed a Q-learning algorithm for learning the behavior of jamming attacks. It requires an excessive involvement of the nodes in the learning process. In general, centralized IDS-based solutions place a burden on CRNs as CRs must be notified on time. Also, the anomaly based solutions require a learning period that might be interrupted and impacted by the jamming attacks. In addition, the learning phase and the continuous update of the discovered jamming behavior place a power pressure on the CR nodes. IoT devices are resource-constrained in terms of processing, storage, or power. This makes the massive participation of IoT nodes in the monitoring and data gathering processes infeasible in many IoT deployment scenarios [21]–[23]. In [24], two jamming mitigation techniques inside CRNs are provided. The two techniques targeted mobile users and assumed no access to multiple channels when jamming occurs. This single-channel access assumption results in avoiding the switching when jamming happens. This results in power saving. These techniques rely on time-based hopping, where each user is assigned a channel for a certain time. The first timehopping technique uses pre-shared keys along with random number generators to guarantee secure communication among users when generating the time-hopping patterns. The second technique considered the channel quality when generating the hopping patterns. The two time-hopping techniques rely on time evasion when jamming happens. This impacts IoT delay sensitive applications [25]. In [26], a similar technique of [24] that uses a time-hopping scheme was presented. The solution is dedicated for sensor CRNs with cluster-head structure. The work in [26] has addressed random jamming only and assumed that channel switching is not allowed when the assigned channel is jammed or occupied by legacy PR users. A collaborative anti-jamming technique is employed in [27] to defend against collaborative jamming attack in CRNs. This technique divides the nodes of the network into proxies and followers and allows the proxies to act as intermediate nodes before getting to the base station. Proxies relay information between followers and base stations. This creates another path which mitigates the impact of jamming attacks on the original

path of communication, nodes to base stations. Channel hopping and surfing are used to avoid attacked channels. Channel surfing moves the transmission to another channel when a jamming attack is detected over a particular channel [28]. Once jamming is detected, the solution suggests activating the logic of changing the channel [29]. The solution selects a new target channel while taking into consideration the channel status. Channel switching requires notifying network users about it. Channel hopping [30] is the most commonly used approach to deal with jamming attacks. The key difference between channel hopping and channel surfing is that channel hopping does not wait for attack detection in order to react. Note that both channel hopping and channel surfing require tight synchronization. In addition to anti-jamming solutions for CRNs, several solutions have been proposed for anti-jamming in IoT environment. In [31], the solution involves the use of a specialized gaming algorithm, where IoT nodes have built-in interference sensors. Nodes report their interference results to a fusion center, which detects jamming attacks according to certain threshold measures. Such solution does not consider the type of underlying network nor able to benefit from already provided information such as PDR and SNR. Also, it requires the implementation of fusion centers and IoT nodes with extra sensing capabilities. In [32], the authors provided another IoT anti-jamming solution that uses a gaming algorithm. Their solution includes monitoring bit error rate (BER) by the access points (APs). APs distribute the power over the different available channels by minimizing the access to transmission power for jammers and areas of potentially detected attacks. In general, we can categorize existing jamming detection and countermeasures techniques into two groups. The first group requires sophisticated network protocols to countermeasure an attack and consumes additional time and computing resources to detect and countermeasure the attack. This leads to wasting network resources and power drain, which can lead to more dangerous attack consequences and making it unsuitable for wireless-based IoT applications [21]–[23]. Worse yet, many IoT devices are heterogeneous. This heterogeneity complicates implementing solutions that require equal participation from IoT nodes [31], [32]. The second group requires additional capabilities at the level of IoT nodes and controllers (e.g., sensing capabilities or APs with additional security logic). This limits the applicability of such solutions. In this paper, we aim at addressing the jamming attack problem by allocating channels at the device level. Very few attempts have been made to allocate channels to CR IoT devices while considering the PR activities, CR channel-quality conditions and jamming behaviour. Our proposed solution focuses on resolving the jamming issue at the level of CRN provider and relives IoT devices from implementing new capabilities whether at the level of IoT nodes or IoT controllers. Also, our solution does not require any detection stage or heavy participation of IoT nodes, which preserves their energy resources. III. N ETWORK M ODEL We consider a CSMA/CA-based CRN for IoT time-critical applications that coexists with a number of different licensed

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 4

PR networks (PRNs), each with its own licensed spectrum and carrier frequency. Let M represent the set of all PR channels. CR IoT devices can dynamically access any frequency channel of the licensed spectrum. The status of each PR channel i ∈ M is modeled using a 2-state BUSY/IDLE alternating renewal (i) process, where the BUSY period (TB ) indicates that channel i is not available for CR communications, while the IDLE (i) period (TI ) indicates that channel i can be utilized by the CR IoT devices using the federal communications commission (i) (FCC) maximum allowable power limit (Pmax ). The BUSY and IDLE periods are generally distributed random variables with probability density function (pdf) fT (i) (t) and fT (i) (t), B I respectively. It has been shown that the process of PR users occupying channel i usually obeys a Poisson process, where (i) the IDLE period has rate 1/T I , and the BUSY period has (i) rate 1/T B [15], [17]. By using CR technology, each CR user can sense the channel usage of PRs. We assume the channelusage pattern over each channel i slowly changes with time, and CR IoT devices can obtain the PR spectrum usage pattern by conducting cooperative spectrum sensing with neighboring CR devices. At any given time, the available channels for a given transmission j is Mj ⊆ M. In general, there are two main types of jamming attacks that are widely used in the literature: non-reactive and reactive jamming [33], [34]. For each channel i, the latter attack refers to those attacks that keep non-active when the channel is idle, but start sending jamming pulses in order to corrupt the ongoing CR communications once a CR activity is sensed. On the other hand, the non-reactive attack is not aware of any activity of CR IoT devices and sends RF jamming pulses based on a predefined strategy. The proactive (reactive) jammer is mathematically represented by a predefined strategy G(I (i) ) (i) (i) (G(pJ )), where I (i) and pJ are the jamming interval and jamming probability associated with channel i; respectively (i) [35]. The terms I (i) and pJ are defined as the time interval between two adjacent jamming signals transmitted over channel i by the jammer and the probability that a given transmission over channel i is successfully jammed; respectively. The estimation of the jamming strategy is an important topic, which has been investigated in [36]. In [36], the authors proposed a learning process based on the maximum likelihood estimation (MLE) to develop a statistical model that allows the SUs in defining the jamming strategy and estimating the needed parameters using previous known observations. For our purposes, we assume that a similar MLE-based technique for determining the jamming strategy/parameters is in place. In a time-critical application, it is essential to provide network availability guarantees in terms of packet delay performance rather than the traditional performance metrics (e.g., throughput, packet loss, etc.). Thus, we consider that a given data packet transmission over a given channel i becomes invalid whenever the transmission delay D(i) is greater than a predefined delay threshold Dth , determined by the application layer. The application layer stops retransmissions when the packet is successfully received, or the transmission delay becomes > Dth as the packet is considered obsolete.

IV. T RANSMISSION I NVALIDITY R ATIO A NALYSIS A given packet transmission j over a given channel i ∈ M (i) is considered a failure whenever the transmission delay Dj (i) exceeds Dth 1 . Let pf (j) denote the failure probability at the MAC layer. We note that depending on the type of network, wireless MAC generally has its own re-transmission strategy based on the employed CSMA/CA mechanism (e.g., the default short and long retransmission limits in IEEE 802.11g are 3 and 7; respectively). The transmission j is said to be a failure over the selected channel(s) only when all transmission (i) attempts fail. Let p(i) (j) and psucc (j) respectively denote the failure and success probabilities of a given attempt. For a packet of length L and transmission rate R(i) , the required transmission time over each channel i can be determined as (i) tx (j) = RL(i) . Note that a packet is lost if jamming (or PR transmission) and packet transmission occur simultaneously. This is due to the fact that CR devices are expected to operate using low transmission power (i.e., the FCC enforces a power mask on CR transmissions to protect PR users) [18]. Because we focus on computing the invalidity ratio for a given (i) transmission j over a given channel i (rj ) at a given time, the subscripts j and i are omitted in the rest of this section to simplify the notation. A. Proactive Jamming For the non-reactive jamming, a packet is successfully delivered if both the idle period of the selected CR channel and the jamming interval over that channel are no smaller than the packet transmission time. In this case, the failure probability p can be computed as:   p = 1 − psucc = 1 − Pr {min{TI , TJ } ≥ tx } = 1 − Pr({TI ≥ tx , TJ ≥ tx }) = 1 − Pr({TI ≥ tx }Pr({TJ ≥ tx })    = 1 − 1 − FTI (tx) 1 − FTJ (tx)

(1)

where FTI and FTJ are the cumulative distribution function (CDF) of the IDLE and jamming intervals; respectively. Thus, given the number of MAC-layer retransmission attempts Nx , pf can be computed as: !    Nx Nx (2) pf = p = 1 − 1 − FTI (tx) 1 − FTJ (tx) By using (2), an upper bound of the packet-invalidity ratio (r) can be computed using the same methodology in [34], [35] as shown in (3), given at the top of the next page. To make our analysis tractable, the status of each PR channel i, ∀i ∈ M is described by a Markov renewal process alternating between BUSY and IDLE periods. This model was previously considered in (e.g., [14]–[17]), which captures well the temporal characteristics of PR spectrum opportunities. Based on this model, BUSY/IDLE periods for the different 1 Note that a generic PR channel i is considered. The provided expressions can be applied to any of the M PR channels by using the PR’s related parameters (PR activity, jamming parameters, achieved transmission rate, etc.)

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 5

r ≤

=

pf E[dk ] (1 − pf )(Dth − E[dk ]) + pf E[dk ]    Nx 1 − 1 − FTI (tx) 1 − FTJ (tx) E[dk ] ! (3)     Nx   Nx 1 − 1 − 1 − FTI (tx) 1 − FTJ (tx) (Dth − E[dk ]) + 1 − 1 − FTI (tx) 1 − FTJ (tx) E[dk ]

where E[dk ] is the mean of the i.i.d. MAC layer delay and dk is the delay during the kth retransmission. PR channels follow exponentially distributed random variables that are statistically independent2 . For a non-reactive jamming, we consider the memoryless jamming model [35], where TJ is exponentially distributed. Furthermore, for the ith channel, the IDLE and Jamming periods are statistically independent of each other. Given these models and using (3), it can be shown that r is upper-bounded by (4), given at the top of next page. B. Reactive Jamming For the reactive jamming, a packet is successfully delivered if the idle period of the assigned channel is greater than the CR packet transmission duration and the packet is not jammed during its transmission time. In this case, the failure probability p can be computed as:   p = 1 − psucc = 1 − Pr {TI ≥ tx } (1 − pJ )   (5) = 1 − 1 − FTI (tx) (1 − pJ ) where pJ denotes the jamming probability. Given Nx , pf is given by: pf =



  1 − 1 − FTI (tx) 1 − pj

s.t. !Nx (6)

.

A. Problem Statement and Formulation Our main objective is to optimize the network performance by minimizing the transmission invalidity ratio for each CR transmission, where each CR is equipped with one transceiver (each CR can utilize one channel only). Specifically, for a given CR transmission j ∈ N , the set of available channels for transmission j is (Mj ), the communicating pair need to select the appropriate channel i∗ (conduct the channel assignment) that minimizes the packet-invalidity ratio (i.e., this ratio is computed such that the packet is considered invalid whenever

(i) TJ

that the provided expression in (3) is applicable to any F

distributions.

(i)

(i)

V. T HE PROPOSED S ECURITY- AWARE C HANNEL A SSIGNMENT

F

(i)

Given µ∗ , Mj , rj and SINRj , ∀i ∈ Mj , the channel assignment optimization problem can be formulated as: X (i) (i) min xj rj xj ∈{0,1}

By using (6), an upper bound of r can be computed using the same methodology used for the non-reactive jamming as shown in (7). Based on the BUSY/IDLE Markov channelavailability model, r is given by (8). Note that the invalidity ratio is mainly a function of channel availability, channel quality, and jamming behavior.

2 Note

the transmission delay D ≥ Dth ) subject to the following design constraints: (1) the received signal-to-interference-noise ratio (SINR) over an assigned channel i should be greater than a given threshold µ∗ , (2) the single-transceiver per CR user (each CR user can utilize only one channel at a time), and (3) the exclusive channel occupancy constraint, in which each idle channel can be utilized by at most one CR transmission at any given time. Note that the last constraint can be ensured by updating the list of idle channels Mj by removing the channels that are reserved for ongoing CR transmissions. To proceed in our analysis, we introduce a binary decision (i) variable xj that is defined as:  1, if channel i is selected by transmission j (i) xj = (9) 0, otherwise.

(i)

TI

and

(i) SINRj

i (i)

− µ∗ ≥ Υ(xj − 1) X (i) xj ≤ 1

∀i ∈ Mj (10)

i∈Mj

where Υ is very large positive constant. B. The Proposed Solution An observation of our optimization problem in (10) indicates that this formulation constitutes a binary linear programming problem, where the optimal solution to such a problem is, in general, NP-hard. To make our analysis more amenable for further investigating, we observe that 1) the number of idle PR channels for each user j (|Mj |) at any given time is finite, and 2) each user can be assigned only one channel at most, and 3) the SINR constraint can be guaranteed by removing all channels with SINR < µ∗ from Mj . Let Mfj represent the feasible channel set. Based on these facts, the channel assignment optimization in (10) can be transferred into a sorting problem with the objective of assigning the idle channel that has the lowest invalidity ratio (denoted by i∗ ) to the CR transmission j. This spectrum assignment can be computed in polynomial-time according to the following algorithm: (i) 1) Given µ∗ , Mj and SINRj , ∀i ∈ Mj , the idle PR channels with received SINR ≥ µ∗ is identified as feasible channels. Then, the algorithm removes all channels with

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 6

 r≤

 1− 1−e



T I +T J L TITJ R

1−e N !



T I +T J L TITJ R

x

Nx E[dk ] 

(Dth − E[dk ]) + 1 − e



T I +T J L TITJ R

(4)

Nx E[dk ]

  Nx E[dk ] 1 − 1 − FTI (tx) 1 − pJ ! .     Nx   Nx E[dk ] 1 − 1 − 1 − FTI (tx) 1 − pJ (Dth − E[dk ]) + 1 − 1 − FTI (tx) 1 − pJ 

r≤

 r≤



L − RT



Nx

I 1− 1−e E[dk ] 1 − pJ ! .    Nx  Nx L  L  − RT − RT I I 1− 1− 1−e 1 − pJ (Dth − E[dk ]) + 1 − 1 − e 1 − pJ E[dk ]

SINR < µ∗ from Mj (identifying Mfj ). After that, the algorithm determines the achieved rates and the needed (i) transmission periods over all channels (tx = RL(i) ) in f Mj . (i) 2) Given the distribution of the idle durations TI , ∀i ∈ (i) f (i) Mj , the jamming strategy G(I ) or G(pJ ), Dth , Nx , (i) (i) E[dk ], tx , ∀i ∈ Mfj , the algorithm computes rj , ∀i ∈ f Mj using (3) or (7). (i) 3) Given rj , ∀i ∈ Mfj and noting that the number of feasible idle channels is finite for a given CR transmission, the algorithm selects the channel i∗ with the minimum invalidity ratio as follows: (i)

i∗ = arg min rj i∈Mfj

(11)

Note that i∗ can be selected in polynomial-time by sorting the idle channels in Mfj in an increasing order according to their invalidity ratio and after that choosing the first channel in the sorted list. Algorithm 1 shows the pseudocode of our proposed channel assignment.

(7)

(8)

Algorithm 1 Channel Assignment (i)

(i)

Input: Mj , SINRj , µ∗ , TI , ∀i ∈ Mj , G(I (i) ) or (i) (i) G(pJ ), Dth , Nx , E[dk ], tx , ∀i ∈ Mj Output: A feasible channel assignment i∗ or return infeasible assignment Let Mfj = Mj for all i ∈ Mj (i) if SINRj < µ∗ f Mj = Mfj − {i} (i) else Compute the invalidity ratio of channel i (rj ) end-of-if end-of-for for all i ∈ Mfj (i) Sort the channels in a increasing order of rj end-of-for Let U be the sorted channel list if U 6= φ Identify the channel that is on the top of U, say k Return i∗ = k else Return “no feasible assignment found” end-of-if

VI. P ERFORMANCE E VALUATION In this section, simulations are conducted to investigate the performance of the proposed security-aware channel assignment scheme. We develop our simulations using CSIM programs (CSIM is a C-based, discrete-event, process-oriented simulation package [37]). A. Simulation Setup We consider a CRN with 20 CR transmissions that coexist with ten PRNs in a field of 150 × 150 m2 . The first five PRNs are licensed in the 600 MHz band with five orthogonal channels, each with a bandwidth of 0.5 MHz. The other

five PRNs are licensed to operate in the 900 MHz band with five orthogonal 0.5 MHz frequency channels. The status of each channel follows the two-state BUSY/IDLE alternating renewal process presented in Section III. We fix the busy probabilities of the different PR channels to PB . The (i) TI average availability durations for the 10 channels are 5, 100, 30, 5, 45, 50, 100, 5, 45, 30 ms; respectively. We use the Rayleigh-fading channel model with a path-loss exponent of n = 4 to determine the link-quality between communicating CR pairs. The CR SINR threshold, the control-packet size, and the data packet length are set to µ∗ = 5 dB, 120 bits and L = 2

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 7

4

6 4

0.5

Security-aware MAC, BW = 0.5 MHz PoS MAC, BW = 0.5 MHz Security-aware MAC, BW = 1 MHz PoS MAC, BW = 1 MHz

Throughput (Mbps)

8

5 Security-aware MAC, BW = 0.5 MHz PoS MAC, BW = 0.5 MHz Security-aware MAC, BW = 1 MHz PoS MAC, BW = 1 MHz

Throughput (Mbps)

Throughput (Mbps)

10

3 2 1

2 0 0

5

10 15 Jamming Interval Factor x (in ms)

20

0.3 0.2 0.1

0 0

25

Security-aware MAC, BW = 0.5 MHz PoS MAC, BW = 0.5 MHz Security-aware MAC, BW = 1 MHz PoS MAC, BW = 1 MHz

0.4

5

(a) Low PR activity

10 15 Jamming Interval Factor x (in ms)

20

0 0

25

2

(b) Moderate PR activity

4 6 8 10 Jamming Interval Factor x (in ms)

12

14

(c) High PR activity

Fig. 1. Throughput performance vs. jamming activity for different PB and BW .

Throughput (Mbps)

2 1.5 1 0.5 0 0.1

Security-aware MAC, x = 11 ms, BW = 0.5 MHz MaXPoS MAC, x = 11 ms, BW = 0.5 MHz Security-aware MAC, x = 11 ms, BW = 1 MHz MaXPoS MAC, x = 11 ms, BW = 1 MHz

7 6 5 4 3 2

10

Throughput (Mbps)

Security-aware MAC, x = 0.5 ms, BW = 0.5 MHz MaXPoS MAC, x = 0.5 ms, BW = 0.5 MHz Security-aware MAC, x = 0.5 ms, BW = 1 MHz MaXPoS MAC, x = 0.5 ms, BW = 1 MHz

Throughput (Mbps)

3 2.5

Security-aware MAC, x = 25 ms, BW = 0.5 MHz MaXPoS MAC, x = 25 ms, BW = 0.5 MHz Security-aware MAC, x = 25 ms, BW = 1 MHz MaXPoS MAC, x = 25 ms, BW = 1 MHz

8 6 4 2

1 0.2

0.3

0.4

0.5 PB

0.6

0.7

(a) High jamming activity

0.8

0.9

0.1

0.2

0.3

0.4

0.5 PB

0.6

0.7

(b) Moderate jamming activity

0.8

0.9

0 0.1

0.2

0.3

0.4

0.5 PB

0.6

0.7

0.8

0.9

(c) Low jamming activity

Fig. 2. Throughput performance vs. PB for N = 10, L = 2 KByte, Dth = 20 ms under different jamming activities for BW = 0.5 and 1 MHz.

KByte, respectively. We also set the thermal noise power (i) density to Pth = 10−21 W/Hz and the maximum transmission (i) power to Pmax = 1 W, ∀i. We use Shannon’s formula to compute the transmission rates of each CR transmission over the idle channel. The locations of CR communicating pairs are randomly allocated within the 150 × 150 m2 region. We consider a time-critical application, in which a given data packet is considered invalid whenever the transmission delay D(i) is > a predefined delay requirement Dth = 20 ms [35]. We set E[dk ] to 1 ms and the maximum number of MAC layer retransmissions Nx = 4 [34], [35]. Our main performance metric is network throughput. Specifically, we investigate the network throughput performance under a reference security-unaware CRN assignment scheme (i.e., the MaXPoS scheme [19]) as a function of x and PB under both proactive and reactive jammers. The MaXPoS assignment selects the available channel with the largest probability of success for each CR link (i.e., MaXPoS is channel qualityand PR activity-aware scheme), irrespective of the jamming behavior. To resolve spectrum contention between CR IoT devices, we use the multi-channel CSMA/CA-based channel access MAC protocol described in [19]. When used with MaXPoS and our proposed security-aware channel assignment schemes, the considered MAC protocol is referred to as MaXPoS-MAC and security-aware MAC, respectively. The reported results are averaged over 200 runs, each lasts for 1000 seconds. B. Simulation Results 1) Proactive Jamming: To simulate the jamming attacks, we consider a non-reactive jammer with memoryless jamming

strategy over each channel i, where the jamming interval as(i) sociated with the ith channel TJ is exponentially distributed (i) (i) with mean T J . The average jamming interval T J over the ten channels are 5, 0.2, 10, 2, 20, 5, 0.1, 2.9, 20, 0.2 × x ms, where x represents the jamming interval factor (jamming (i) attack level). Recall that the term TJ is defined as the time interval between two consecutive jamming signals sent over channel i by the jammer. We first study the impact of proactive jamming attacks on network performance. Figure 1 illustrates the throughput performance for a time-critical IoT application with Nx = 4, E[dk ] = 2 ms, and Dth = 20 ms under the attack of (i) memoryless jammers with 1 ≤ T J x ≤ 14 ms for low, moderate and high PR activities. Two different channel bandwidth BW are also considered. This figure shows that our securityaware assignment significantly outperforms MaXPoS-MAC, irrespective of the various network conditions. Specifically, Figure 1 reveals that under low-to-moderate PR activities, the proposed security-aware scheme enhances throughput performance by up to 140% compared to MaXPoS-MAC. This enhancement is mainly attributed to the appropriate securityaware spectrum assignment that minimizes the packet invalidity ratio of CR transmissions, and hence, reduces the number of dropped (invalid) packets. This figure also shows that throughput enhancement is larger at larger BW . This is expected because larger values of BW indicate higher achieved rate and shorter transmission time, which reduces the invalidity rate by increases the successful delivery of data packets. It can be noticed that under high PR activities, network throughput of the proposed security-aware scheme gracefully degrades to that of MaXPoS-MAC due to the limited number of available

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 8

Throughput (Mbps)

5

Security-aware MAC, Low PJmax MaXPoS MAC, Low PJmax

4 3 2 1 0 0.1

0.2

0.3

0.4

0.5 PB

0.6

0.7

0.8

0.9

(a) Low reactive jamming probability

Throughput (Mbps)

3

Security-aware MAC, High PJmax MaXPoS MAC, High PJmax Security-aware MAC, Moderate PJmax MaXPoS MAC, Moderate PJmax

2

1

0 0.1

0.2

0.3

0.4

0.5 PB

0.6

0.7

0.8

0.9

(b) Moderate and high reactive jamming probability Fig. 3. Throughput performance vs. PB for different jamming probability.

2) Reactive Jamming: In this section, we investigate the performance of our proposed protocol under reactive jamming strategy. We consider the simulation setup presented in Section

Throughput (Mbps)

5

Security-aware MAC, PB = 0.1 MaXPoS MAC, PB = 0.1

4 3 2 1 0 0

0.2

0.4 P max J

0.6

0.8

1

(a) Low PR activity

2.5 Throughput (Mbps)

idle channels. Figure 1 indicates that, for both protocols, as x increases the achieved throughput increases. This figure also indicates that throughput enhancement is limited at smaller x. This is expected because larger values of x indicate less jamming attack level. We now investigate the performance under different PR activities under a low, moderate and high level of jamming for BW = 0.5 and 1 MHz. Figures 2(a)-(c) reveal that the throughput improvement is smaller at larger PB . This is expected as the larger PB , the smaller is the number of available PR channels, which will decrease the chances of finding a proper idle channel, over which a given data packet can correctly be delivered within the delay threshold. This results in increasing the number of invalid packets, and hence reducing the achieved throughput improvement (e.g., for PB = 0.9, network throughput of the proposed scheme degrades to that of MaXPoS-MAC as the number of idle channels becomes very small). Figures 2 also indicates that throughput enhancement is smaller at lower jamming level. This is because as the jamming attacks become less severe, the dominant network performance factors become the PR activities and channel quality, which are already taken into consideration in the design of MaXPoS-MAC. It can be noticed that throughput enhancement is larger at larger values of BW . This is because as BW increases, our securityaware spectrum assignment can further minimize the packet invalidity ratio of CR transmissions, and hence improving the achieved throughput.

Security-aware MAC, PB = 0.9 MaXPoS MAC, PB = 0.9 Security-aware MAC, PB = 0.5 MaXPoS MAC, PB = 0.5

2 1.5 1 0.5 0 0

0.2

0.4 PJmax

0.6

0.8

1

(b) Moderate and High PR activity Fig. 4. Throughput performance vs. reactive jamming activity for different PB .

VI-A, but we change the jamming strategy. To simulate the reactive jamming attacks, we consider a reactive jammer with an independent jamming strategy over each channel i, where each channel i is associated with the jamming probability (i) pJ . The jamming probability over the ten PR channels are {0.06, 0.75, 0.03, 0.15, 0.015, 0.06, 1, 0.105, 0.015, 0.75} × PJmax , where 0 ≤ PJmax ≤ 1 represents the jamming probability factor such that the maximum probability over any channel does not exceed 1. By varying PJmax , we can study the performance of the CRN under different jamming conditions. We investigate network performance as a function of PB under low, moderate and high levels of jamming. Figure 3 shows that the throughput improvement is smaller at larger PB due to the less availability of PR channels. Figures 3 also indicates that throughput enhancement is smaller at lower probability of jamming. This is because as the jamming attacks become less severe, the dominant network performance factors become the PR activities and channel quality, which are already considered in the design of MaXPoS-MAC. Figure 4 shows that, under low-to-moderate PR activities, our security-aware assignment significantly outperforms MaXPoS-MAC, irrespective of the various network conditions. It can be noticed that under high PR activities, network throughput of the proposed securityaware scheme gracefully degrades to that of MaXPoS-MAC due to the limited number of idle channels. This figure also shows that as the jamming probability factor PJmax increases, the achieved throughput decreases. This is expected because larger values of PJmax indicate more sever jamming attacks.

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 9

VII. C ONCLUSIONS This paper presented a new channel assignment algorithm for IoT-based CRNs with time-sensitive traffic under jamming attacks. The proposed algorithm represents a new solution to countermeasure jamming attacks, where it does not require jamming detection stage nor redundancy in network resources. The algorithm uses a set of factors when selecting channels: security, link-quality, and PR activity. A closedform expression for the packet invalidity ratio is derived and takes into consideration the unique characteristics of the CRN environment as well as the effect of jamming attacks. The ratio is used to design a probabilistic security, linkquality, and PR activity-aware channel assignment mechanism that provides the communicating pair with the most secured channel with the lowest invalidity probability, while leading to improved system throughput. Simulation results showed that a remarkable throughput improvement can be achieved over conventional channel assignment mechanisms by being security-aware while jointly considering the PR activities and CR link-quality over different channels. R EFERENCES [1] A. Al-Fuqaha, M. Guizani, M. Mohammadi, M. Aledhari, and M. Ayyash, “Internet of things: A survey on enabling technologies, protocols, and applications,” IEEE Communications Surveys Tutorials, Vol. 17, No.4, pp. 2347–2376, 2015. [2] P. Rawat, K. Singh, and J. Bonnin, “Cognitive radio for M2M and internet of things: A survey,” Computer Communications, Vol. 94, pp. 1-29, 2016. [3] V. Petrov, S. Edelev, M. Komar, and Y. Koucheryavy, “Towards the era of wireless keys: How the IoT can change authentication paradigm,” in Proc. of the IEEE World Forum on Internet of Things (WF-IoT), 2014. [4] A. Ranjan and G. Somani, “Access control and authentication in the internet of things environment,” in Computer Communications and Networks, pp. 283–305, 2016. [5] A. Khan, M. Rehmani, and A. Rachedi, ”When cognitive radio meets the internet of things?”, in Proc. of the International Wireless Communications and Mobile Computing Conference (IWCMC), Sep. 2016. [6] A. Khan, M. Rehmani, and A. Rachedi, ”Cognitive-radio-based internet of things: Applications, architectures, spectrum related functionalities, and future research directions”, IEEE Wireless Communications, Vol. 24, No.3, pp. 17–25, 2017. [7] Q. Wu et al., ”Cognitive Internet of Things: A New Paradigm Beyond Connection,” IEEE Internet of Things Journal, vol. 1, no. 2, pp. 129-143, April 2014. [8] J. Ploennigs, A. Ba and M. Barry, “Materializing the Promises of Cognitive IoT: How Cognitive Buildings are Shaping the Way,” IEEE Internet of Things Journal, 2017. [9] K. Apampa, G. Wills, and D. Argles, “Towards security goals in summative e-assessment security,” in Proc. of the International Conference for Internet Technology and Secured Transactions, (ICITST), London, Nov. 2009, pp. 1-5. [10] C. Manogna and K. Naik, “Detection of Jamming Attack in Cognitive Radio Networks”, International Journal of Recent Advances in Engineering & Technology (IJRAET), Vol. 2, Issue 6, pp. 69-72, 2014. [11] Y. Jararweh, H. Bany Salameh, A. Alturani, L. Tawalbeh, and H Song, “Anomaly-based framework for detecting dynamic spectrum access attacks in cognitive radio networks,” Telecommunication Systems, 2017. [12] V. Balogun and A. Krings, “On the impact of jamming attacks on cooperative spectrum sensing in cognitive radio networks”, in Proc. of the Eighth Annual Cyber Security and Information Intelligence Research Workshop (CSIIRW’13), NY, USA, 2013. [13] K. Grover, A. Lim, and Q. Yang, “Jamming and anti-jamming techniques in wireless networks: a survey,” International Journal of Ad Hoc and Ubiquitous Computing, Vol. 17, No. 4, 2014. [14] H. Bany Salameh, M. Krunz, and D. Manzi, ”Spectrum Bonding and Aggregation with Guard-band Awareness in Cognitive Radio Networks,” IEEE Trans. on Mobile Computing, vol. 13, no. 3, pp. 569-581, 2014.

[15] G. Uyanik, M. Abdel-Rahman, and M. Krunz, “Optimal channel assignment with aggregation in multi-channel systems: A resilient approach to adjacent-channel interference, Ad Hoc Netw., vol. 20, pp. 64-76, 2014. [16] H. Bany Salameh, “Efficient Resource Allocation for Multi-cell Heterogeneous Cognitive Networks with Varying Spectrum Availability,”IEEE Transactions on Vehicular Technology, Vol. 65, No. 8, 2016. [17] M. Abdel-Rahman and M. Krunz, “Stochastic guard-band-aware channel assignment with bonding and aggregation for DSA networks,” IEEE Trans. on Wireless Communications, vol. 14, no. 7, pp. 3888-3898, 2015. [18] H. Bany Salameh, H. Kasasbeh, and B. Harb, “A batch-based MAC design with simultaneous assignment decisions for improved throughput in guard-band-constrained cognitive networks,” IEEE Transactions on Communications, Vol. 64, No. 3, pp. 1143–1152, March 2016. [19] H. Bany Salameh and O. Badarneh, “Opportunistic Medium Access Control for Maximizing Packet Delivery Rate in Dynamic Access Networks,” Journal of Network and Computer Applications, Vol. 36, Issue 1, pp. 523-532, 2013. [20] F. Slimeni, B. Scheers, Z. Chtourou, and V. Nir, ”Jamming mitigation in cognitive radio networks using a modified q-learning algorithm”, in Proc. of the International Conference on Military Communications and Information Systems (ICMCIS), May 2015. [21] I. Yaqoob, E. Ahmed, I. Abaker, T. Hashem, A. Ibrahim, A. Ahmed, A. Gani, M. Imran, and M. Guizani, ”Internet of things architecture: Recent advances, taxonomy, requirements, and open challenges”, IEEE Wireless Communications, vol. 24, no. 3, pp. 10–16, 2017. [22] S. Thakare, A. Shriyan, V. Thale, P. Yasarp, and K. Unni, ”Implementation of an energy monitoring and control device based on IoT”, in Proc. of the IEEE Annual India Conference (INDICON), Dec. 2016. [23] H. Jayakumar, A. Raha, Y. Kim, S. Sutar, W,S. Lee, and V. Raghunathan, ”Energy-efficient system design for IoT devices”, in Proc. of the 21st Asia and South Pacific Design Automation Conference, 2016. [24] N. Adem, B. Hamdaoui, and A. Yavuz, “Mitigating jamming attacks in mobile cognitive networks through time hopping”, Wireless Communications and Mobile Computing, Vol. 16, No. 17, pp. 3004–3014, 2016. [25] I. Awan and M. Younas, ”Towards QoS in Internet of Things for Delay Sensitive Information”, in Proc. of theTrends in Mobile Web Information Systems (MobiWIS), pp. 86–94, 2013. [26] N. Adem and B. Hamdaoui, “Jamming resiliency and mobility management in cognitive communication networks”, in Proc. of the IEEE International Conference on Communications (ICC), May 2017. [27] W. Wang, S. Bhattacharjee, M. Chatterjee, and K. Kwiat, “Collaborative jamming and collaborative defense in cognitive radio networks”, Pervasive and Mobile Computing, Vol. 9, No. 4, pp. 572–587, 2013. [28] W. Xu, T. Wood, and Y. Zhang, “Channel surfing and spatial retreats: defenses against wireless denial of service,” in Proc. of the ACM workshop on Wireless security, 2004, pp. 80–89. [29] S. Khattab, D. Mosse, and R. Melhem,“Modeling of the channel-hopping anti-jamming defense in multi-radio wireless networks,” in Proc. of the 5th International ICST Conference on Mobile and Ubiquitous Systems: Computing, Networking and Services, Ireland, 2008, pp. 1-10. [30] L. Lazos, S. Liu, and M. Krunz, “Mitigating control-channel jamming attacks in multi-channel ad hoc networks”, in Proc. of the Second ACM Conf. on Wireless Network Security (WiSec), 2009, pp. 169–180. [31] M. Labib, S. Ha, W. Saad, and J. H. Reed, ”A colonel blotto game for anti-jamming in the internet of things”, in Proc. of the IEEE Global Communications Conference (GLOBECOM), Dec. 2015. [32] N. Namvar, W. Saad, N. Bahadori, and B. Kelley, ”Jamming in the internet of things: A game-theoretic perspective”, in Proc. of the IEEE Global Communications Conference (GLOBECOM’16), 2016. [33] W. Xu, W. Trappe, Y. Zhang, and T. Wood, “The feasibility of launching and detecting jamming attacks in wireless networks,” in Proc. of the 6th ACM International Symposium on Mobile Ad Hoc Networking and Computing, NY, USA, 2005, pp. 46–57. [34] Z. Lu, W. Wang, and C. Wang, “From jammer to gambler: Modeling and detection of jamming attacks against time-critical traffic,” in Proc. of the IEEE INFOCOM, April 2011, pp. 1871–1879. [35] Z. Lu, W. Wang, and C. Wang, “Modeling, evaluation and detection of jamming attacks in time-critical wireless applications,”IEEE Transactions on Mobile Computing, Vol. 13, No. 8, pp. 1746–1759, 2014. [36] Y. Wu, B. Wang, and K. Liu, “Optimal defense against jamming attacks in cognitive radio networks using the Markov decision process approach”, in Proc. of the IEEE Globecom Conference, 2010. [37] “Mesquite Software Incorporation,” www.mesquite.com.

2327-4662 (c) 2018 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.2018.2817339, IEEE Internet of Things Journal 10

Haythem A. Bany Salameh (M’06-SM’16) received the Ph.D. degree in electrical and computer engineering from the University of Arizona, USA, in 2009. He is currently a Professor of telecommunication engineering with Yarmouk University (YU), Jordan. He was a recipient of the Jordan Science Research Support Foundation (SRSF) Prestigious Award for Distinguished Research in ICT in 2015, the Best Researcher Award for Scientific Colleges in YU in 2016, and the SRSF Award for Creativity and Technological Scientific Innovation in 2017. His research interests include wireless networking, with emphasis on dynamic spectrum access, cognitive radio networking, Internet-of-Things, and distributed protocol design. He has served and continues to serve on the Technical Program Committee of many international conferences.

Sufyan Almajali obtained his Ph.D. in Computer Science from Illinois Institute of Technology, Chicago, IL. He has 18 years of academic and industrial experience. He is currently the dean of King Hussein School of Computing Sciences at Princess Sumaya University for Technology. His research interest is in IoT Security, Mobile Edge Computing, and Network Security. He served as chief technology officer at Secure Data Replicator in Chicago, where he supervised the development of an online real-time data replication system.

Moussa Ayyash (M’98-SM’12) received his B.S., M.S. and Ph.D. degrees in Electrical and Computer Engineering. He is currently a Professor at the Department of Information Studies, Chicago State University, Chicago. He is the Director of the Center of Information and National Security Education and Research. His current research interests span digital and data communication areas, wireless networking, visible light communications, network security, Internet of Things, and interference mitigation. Dr. Ayyash is a member of the IEEE Computer and Communications Societies and the Association for Computing Machinery.

Hany Elgala is an assistant professor in the Computer Engineering Department, at the University of Albany-State University of New York (SUNY). Before moving to SUNY he was a research professor at Boston University and the Communications Testbed leader at the National Science Foundation Smart Lighting Engineering Research Center. His research focuses on visible light communications (VLC) or LiFi, wireless networking, and embedded systems. He is a member of the IEEE and IEEE Communications Society.

2327-4662 (c) 2018 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.