Priority- and Delay-Aware Medium Access for Wireless ... - IEEE Xplore

IEEE SYSTEMS JOURNAL

1

Priority- and Delay-Aware Medium Access for Wireless Sensor Networks in the Smart Grid Irfan Al-Anbagi, Member, IEEE, Melike Erol-Kantarci, Member, IEEE, and Hussein T. Mouftah, Fellow, IEEE

Abstract—Monitoring smart-grid assets in a timely manner is highly desired for emerging smart-grid applications such as transformer monitoring, capacitor bank control, plug-in hybridelectric-vehicle load management, and power quality assessment. Wireless sensor and actor networks (WSANs) are anticipated to be widely utilized in a wide range of smart-grid applications due to their numerous advantages along with their successful adoption in various critical areas including military and health. For resource-constrained WSANs, transmitting delay-critical data from smart-grid assets calls for data prioritization and delay responsiveness. In this paper, we introduce two medium-access approaches, namely, delay-responsive cross-layer (DRX) data transmission and fair and delay-aware cross-layer (FDRX) data transmission, which aim to address the delay and service requirements of smart grids. DRX is based on delay-estimation and data-prioritization steps that are performed by the application layer, in addition to the MAC layer parameters responding to the delay requirements of the smart-grid application and the network condition. On the other hand, FDRX incorporates fairness into DRX by preventing a few nodes from dominating the communication channel. We provide a comprehensive performance evaluation of those approaches. We show that DRX reduces the end-to-end delay while FDRX has lower collision rate compared with DRX. We outline the tradeoffs regarding these approaches and draw future research directions for robust communication protocols for smart-grid monitoring applications. Index Terms—Delay-sensitive, medium-access control, smart grid, wireless sensor and actor networks (WSANs).

I. I NTRODUCTION

I

NTEGRATING distributed renewable-energy-generation techniques effectively, fine-grained demand management, and enhanced monitoring of the smart-grid assets are among the top priority tasks on the smart-grid agenda. In this context, accurate and near-real-time information collected from generators, transmission equipment, transformers, capacitor banks, and substations becomes essential for many smart-grid applications. Wireless sensor and actor networks (WSANs) are considered potential tools for monitoring and controlling smart grids. A WSAN is composed of a large number of low-cost, low-power, small, and multifunctional sensor and actor nodes. Sensor and actor nodes communicate wirelessly over short distances.

Manuscript received April 15, 2012; revised September 24, 2012; accepted January 23, 2013. The authors are with the School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, ON K1N 6N5, Canada (e-mail: ialan055@ uottawa.ca; [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSYST.2013.2260939

Sensor nodes can collect various kinds of data, e.g., voltage, current, frequency, etc., while actors perform tasks such as closing/opening circuit breakers, turning on/off loads, etc. WSANs are preferred due to their ability to work in extreme environmental conditions, in addition to having enhanced fault tolerance, low power consumption, self-configuration, rapid deployment, and low cost. In environments where high voltages are in use, WSAN can also provide necessary insulation. Despite the advantages of WSANs, they have not been utilized extensively for monitoring critical smart-grid assets. This is mostly due to the inherent limitations of WSANs in real-time data delivery. WSANs use low-power communication links in dense deployments, hence introducing low data rates and delays in channel access. The aforementioned challenges raise reliability concerns in the smart grid. In fact, reliable data delivery has been widely studied in the wireless sensor network (WSN) literature where the term “reliable” generally refers to ensuring data to be delivered from source to destination or sink. In the context of smart grids, reliability includes timeliness as well, since obsolete data or control signals may be even worse than having no data or signals. For instance, in a scenario where plug-in hybrid-electric-vehicle load management is coupled with the status of the electricity distribution system, delayed information regarding the status of the transformers in the substation may result in unnecessary load control, causing inconvenience for consumers or, worse, risk the stability of the grid. Meanwhile, it is also apparent that not all of the collected data from the substations are significant in control actions. Some data are collected for general monitoring purposes and can be processed in a nonreal-time manner. The significance of predictable reliability, timeliness, and quality of service (QoS) in smart-grid communications has been also outlined in the recent studies [1]. In addition, it is well known that protocols designed in an application-specific manner improve the performance of the WSN [2], [3]. For this reason, we focus on the use of WSNs in the smart-grid domain and aim to improve their performance in terms of delay and QoS. In this paper, we present two protocols that aim to address data-prioritization and delay-sensitive data transmission for WSANs in the smart grid. The first approach, delay-responsive cross-layer (DRX) data transmission, has been proposed in [4]. DRX uses application-layer data prioritization to control the medium access of sensor and actor nodes. DRX first performs delay estimation. If the estimated delay cannot meet the delay requirements of the smart-grid application, then the channel access of the node is fast-tracked by reducing the

1932-8184 © 2013 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2

clear channel assessment (CCA) duration. The second approach, namely, fair and delay-aware cross-layer (FDRX) data transmission, incorporates fairness into delay-sensitive data transmission [5]. Similar to DRX, FDRX initially executes delay assessment. If the estimated delay is higher than the delay requirements of the application, then the node is given higher priority to access the channel. To provide fairness, the node periodically yields to the other nodes in the WSAN. Hence, FDRX provides fairness by periodically allowing other nodes in the personal area network (PAN) to contend fairly to access the channel. In this paper, we present exhaustive simulation results of DRX and FDRX. We include comparisons with previously proposed QoS supporting mechanisms [6]–[8]. We further evaluate the yielding intensity of FDRX and different CCA durations of DRX. DRX has the lowest average end-to-end delay while FDRX has lower packet collision rate compared with DRX. We also evaluate the performance of DRX and FDRX in various smart-grid environments with varying channel properties. Furthermore, we discuss the smart-grid applications with strict delay requirements and present that DRX and FDRX both satisfy the delay requirements of those applications.

A. Main Contributions Smart grids call for low-latency and priority-aware monitoring solutions. Thus, to make WSNs operationally ready for the smart grid, enhancing the performance of WSNs is highly desired. The main contribution of this paper is to provide an exhaustive performance evaluation of the DRX and FDRX schemes in the smart-grid environment and compare their performance to existing schemes. We evaluate the performance of DRX and FDRX schemes with different CCA detection methods, namely, carrier sensing with energy detection and energy-detection methods. We also consider the effects of the yielding intensity on the performance of the studied schemes. The impact of priority and delay awareness in medium-access techniques on the end-to-end delay, delivery ratio, and energy consumption of the WSANs are presented. We further evaluated the performance of DRX and FDRX schemes with deterministic and empirical path loss models. Additionally, we discuss the applicability of priority- and delay-aware medium-access schemes in various smart-grid applications including transformer monitoring, capacitor bank control, and fault current indicator which have tight delay requirements.

B. Paper Organization The rest of this paper is organized as follows. In Section II, we present the related work. In Section III, we give a brief overview of the IEEE 802.15.4 medium-access protocol and in Section IV, we describe the problem. In Section V, we present the analytical model for delay estimation that is utilized by DRX and FDRX. In Section VI, we introduce DRX and FDRX in detail and discuss the results in Section VII. Finally, Section VIII concludes this paper and gives future research directions.

IEEE SYSTEMS JOURNAL

II. R ELATED W ORK In the literature, several studies have proposed the use of WSNs for monitoring utility assets and power-grid equipment [8]–[13]. In [9], a WSN system has been proposed to monitor partial-discharge (PD) activity in high-voltage transformers where PD data were collected from individual sensors on the transformers and transmitted to the base station. In [10], the authors discuss the use of wireless multimedia sensor and actor networks in various smart-grid settings, including electricity production facilities, transmission and distribution systems, and customer premises. Employing WSNs in customer premises for the purpose of pervasive demand response actions has been evaluated in [11]. The performance of WSNs in smart-grid assets has been elaborated in [12] and [13]. The study in [12] presents an evaluation of the performance of WSNs in substations, underground transformer vaults, and power rooms. The authors particularly focus on the impact of noise on the low-power wireless links that are being used by IEEE 802.15.4-based WSNs. Meanwhile, the impact of delay on smart-grid applications has been initially investigated in [13], considering a WSN that is used for the condition monitoring of a wind turbine. Cross-layer protocols have been studied in the general context of WSNs. In [14] and [15], the authors propose a crosslayer protocol to combine the functionalities of medium-access, routing, and congestion control and address receiver-based contention, congestion control, and duty cycling in WSNs. Reducing the end-to-end delay of a WSN has been also studied for more general applications. In [16], the authors have proposed an adaptive back-off (BO) exponent (BE) management scheme for carrier-sense multiple access with collision avoidance (CSMA/CA) of 802.15.4 and investigated its effects on the power consumption of the node. Besides generic delay reduction, QoS has also been studied in the literature where high-priority sensor data are aimed to be forwarded with less delay or higher reliability. In [6], the authors propose an adaptive mechanism by implementing BE management to reduce packet collision. DRX [4] and FDRX [5] also use an adaptive mechanism but with different crosslayer techniques, as will be explained later in this paper. In [7], the authors present a QoS support mechanism in beaconenabled mode using CSMA/CA BO time. Furthermore, in [17], the authors have proposed a distributed algorithm that meets the application-specific reliability and energy consumption requirements. In [8], priority-based schemes to guarantee the timebounded delivery of high-priority packets in event-monitoring networks have been proposed. In [8], the authors propose to reduce the number of CCAs performed in high-priority nodes from two to one and perform frame tailoring to avoid collision. Different from [8], the DRX [4] and the FDRX [5] schemes implement an adaptive process in modifying the CCA duration. Note that they do not modify the number of CCAs. In addition, the impact CCA methods such as energy detection and preamble detection have been thoroughly investigated in [18] and [19]. However, the impact of adaptive CCA duration has not been explored. DRX and FDRX basically aim to reduce the end-to-end delay by adaptively changing the duration of the

AL-ANBAGI et al.: PRIORITY- AND DELAY-AWARE MEDIUM ACCESS FOR WIRELESS SENSOR NETWORKS

3

CCA of certain nodes and setting this parameter to default when prioritization is not required. III. BACKGROUND ON IEEE 802.15.4 MAC P ROTOCOL A. IEEE 802.15.4 MAC Operation The IEEE 802.15.4 standard defines the MAC and physical layers including the CSMA/CA process [20]. CSMA/CA is used with a slotted binary exponential BO scheme to reduce collisions. Two channel-access techniques are defined in the IEEE 802.15.4 standard; these are the beacon-enabled mode, which employs a slotted CSMA/CA and exponential BO, and a basic unslotted CSMA/CA without beacons. The MAC sublayer uses four variables to regulate channel access; these variables are the number of BOs (N BO), contention window (CW ), back-off exponent (BE), and retransmission times (RT ). Prior to a particular transmission in the slotted CSMA/CA, the MAC sublayer initializes the four variables as follows: N BO = 0, CW = 2, BE = min BE, and RT = 0. In the next step, the MAC sublayer delays for a random N BO period ranging from 0 to (2BE − 1). When the BO period becomes zero, the node can perform the first CCA for a certain amount of time. If two successive CCAs are idle, then the node is allowed to start packet transmission. On the other hand, if either of the CCA fails due to a busy channel, the MAC sublayer will increase the value of both N BO and BE by one. This process is repeated until either the maximum value of BOs (M axBackof f s) or the maximum value of BE (M axBE) is reached, and at this point, the packet is dropped and channel access failure is declared. On the other hand, if the channel access is successful, the node initiates the transmission of the packet. If the acknowledgement (ACK) mechanism is activated, the node waits for an ACK which indicates successful packet transmission. If the transmitting node does not receive the ACK within a specified duration, the RT is increased by one up to a value equal to M axF rameRetries. If RT is less than M axF rameRetries, the MAC sublayer initializes two variables CW = 0 and BE = M inBE and repeats the aforementioned process. Otherwise, the packet is discarded due to the retry limit. The default MAC parameters of the IEEE 802.15.4 standard are M inBE = 3, M axBE = 5, M axBackof f s = 4, and M axF rameRetries = 3. Other values such as interframe spacing (IF S) and the ACK wait duration are specified in [20]. Fig. 1 shows a simplified flowchart of the slotted CSMA-CA algorithm [21]. B. CCA in the IEEE 802.15.4 Protocol Three bands for operation are defined in the IEEE 802.15.4 standard: 868 MHz, 902 MHz, and 2.4 GHz. A data rate of 250 kbps can be provided by the 2.4-GHz band by utilizing one of 16 pseudo-orthogonal PN codes with a length of 32 chips to characterize 4 b of information. As specified in the standard, CCA can be performed using three different methods, namely, energy detection, carrier sensing, or a combination of the two. The standard also defines the CCA detection time as eight symbol periods; this means that the PHY layer should finish the

Fig. 1. Slotted CSMA-CA algorithm.

CCA and report the results to MAC within eight symbol periods which is equivalent to 128 μs (each symbol period is 16 s). In CCA, there is a possibility of a false alarm due to noise and interference which prevents transmission. False channel detection could cause collisions and affect the overall system performance. Therefore, there is considerable freedom in choosing an appropriate CCA method and its parameters, depending on the requirements of the application and the environment. Each CCA method differs in its ability to sense signal existence and in its power consumption. Hence, the choice of the CCA method and parameters has a considerable impact on the performance of MAC sublayer metrics such as delay and energy efficiency. These metrics are conflicting and require critical optimization of CCA parameters to achieve a practical adjustment [18].

4

IEEE SYSTEMS JOURNAL

In the carrier-sensing method, the node must accomplish time synchronization with the ongoing transmission. Achieving time synchronization in packet-based wireless systems is done by transmitting a preamble in front of each packet. This preamble consists of repetitions of a sequence of predefined symbols. The receiver performs a correlation of the known sequence with the received signal with varying time offsets. The correlation becomes high due to the repetition of the known symbols which corresponds to time synchronization. This high correlation is indicative of signal presence and provides an estimate of time offset [18], where this method is also known as preamble detection. Due to the use of signal spreading and the known symbols in the preamble, carrier-sensing method has much higher SNR than the energy-detection method and hence operates with higher reliability. The biggest disadvantage of carrier sensing is that it is required to be constantly running. This is because when the physical layer is requested to do a CCA within eight symbol durations, it may not be able to perform preamble detection because the channel could contain data packets with elapsed preamble when CCA is requested [19]. This makes the carriersensing method a power-hungry scheme. The third method combines the advantages of the carriersensing and energy-detection schemes, enabling carrier detection anywhere within the packet duration. This technique provides reliable detection with less energy consumption compared with the carrier-sensing technique. The details of this technique can be found in [20]. IV. P ROBLEM D EFINITION In this paper, we aim to address QoS issues of WSANs in delay-critical smart-grid applications. WSANs are favored in many applications for their low-cost, ubiquity, and flexibility. On the other hand, WSANs may incur high latency when sensor nodes try to access the communication medium simultaneously. Furthermore, data from several smart-grid applications will naturally have different delay requirements calling for prioritization. For instance, metering may tolerate delay while transformer monitoring will have low latency requirement particularly during peak load hours. The medium-access scheme of IEEE 802.15.4 is designed to regulate medium access in WSANs by a random access mechanism, i.e., CSMA/CA. However, CSMA/CA is not tailored for delay-critical smart-grid applications, since it does not have prioritized access nor delay-responsiveness properties. In the smart grid, transformer monitoring, capacitor bank control, and fault current indicator applications call for priority- and delayaware medium-access solutions. In this paper, we evaluate the performance of priority- and delay-aware medium-access solutions and discuss their applicability to the aforementioned smart-grid applications. V. D ELAY E STIMATION M ODEL We adapt the general analytical model for the slotted CSMA/ CA mechanism of the beacon-enabled mode of the IEEE 802.15.4 presented in [22]. We consider a star topology where all nodes in the PAN contend to acquire the channel and send

data to the PAN coordinator. Accurate and approximate models were proposed in [22], and both models solve a set of highly nonlinear equations using numerical methods. We follow several mathematical and algebraic strategies to derive the following formulas, but we do not report all the computations due to limited space. We found that accurate analysis is computationally demanding and not suitable for use in sensor devices with limited resources. In this paper, we use the approximate model. It is essential to mention here that, in this paper, we are using the analytical model of [22] to estimate the end-to-end delay which will trigger the operation of the prioritization and delay-reduction schemes in MAC sublayers. The model is based on the idea that sensor nodes can estimate the probabilities of a busy channel ρ, σ, and ν, where ρ is the probability that the first CCA CCA1 is busy, σ is the probability that CCA2 is busy, and ν is the probability that a node attempts first carrier sensing CCA1 in a randomly selected time slot. The probability ν depends on ρ and σ and on the probability that a transmitted packet encounters a collision Pco . Pco is the probability that at least one of the N − 1 remaining nodes transmits in the same time slot. If all nodes transmit with probability ν, Pco is given by Pco = 1 − (1 − v)N −1

(1)

where N is the number of nodes. The probability of having CCA1 busy ρ is given by the summation of the probability of finding a channel busy during CCA1 due to data transmission (ρ1 ) and the probability of finding a channel busy during CCA2 due to ACK transmission (ρ2 ) ρ = ρ1 + ρ2

(2)

where   ρ1 = L 1 − (1 − v)N −1 (1 − ρ)(1 − σ) ρ2 = Lack

(3)

 N v(1 − v)N −1  1 − (1 − v)N −1 (1 − ρ)(1 − σ) N 1 − (1 − v) (4)

with Lack as the length of the acknowledgement. The probability that CCA2 is busy is given by σ=

1 − (1 − v)N−1 + N v(1 − v)N−1 2 − (1 − v)N + N v(1 − v)N−1

(5)

where v = (1 + a)(1 + b)p0,0,0 a = ρ + (1 − ρ)σ

and

(6) b = Pco (1 − am+1 ).

(7)

p0,0,0 is the approximate stationary distribution of the Markov chain and given by [22]  Wo (1 + 2a)(1 + b) + Ls (1 − a2 )(1 + b) p0,0,0 = 2 −1      2 2 2 n−1 Pco (1 − a ) +1 +1 (8) + Υ0 ( Pco (1 − a )

AL-ANBAGI et al.: PRIORITY- AND DELAY-AWARE MEDIUM ACCESS FOR WIRELESS SENSOR NETWORKS

where m = M axBackof f s, Wo = 2M inBE , n = M axF rameRetries, and Ls is the time period of successful transmission and is given by the following relation: Ls = L + tack + Lack + IF S.

(9)

Here, L is the total length of the packet including overhead and payload, tack is the acknowledgment waiting duration, and IF S is the interframe spacing. Υ0 = Lo po /(1 − po ), where Lo is the idle state length and po is the probability of going back to the idle state. Equations (2), (5), and (6) can be solved to find the values of ρ, σ, and ω, respectively. The average estimated delay is given by E[D] = P T D

(10)

where P = [Pr(X0 |Xt ) . . . Pr(Xn |Xt )]T , D = [d0 . . . dn ]T , dj = Ts + jTc + (j + 1)E[T]   j j 1 − Pco (1 − am+1 ) Pco (1 − am+1 ) . (11) Pr(χj |χt ) = 1 − (Pco (1 − am+1 ))n+1 Ts and Tc are the time durations of successful and collided packet transmissions, respectively. χj is the occurrence of successful packet transmission at time j + 1 given that at time j, the transmission is unsuccessful. χt is the occurrence of successful packet transmission within n attempts  TT ) E[T ] = 2Ts (1 + P

(12)

 = [P˜ (Bo |Bt ) . . . P˜ (Bm |Bt )]T , T = [t0 . . . tm ]T where P

expected delay based on the model described in Section V. Thus, a node makes a decision based on the delay estimation algorithm by making the MAC sublayer respond to the specific delay requirement of the application. If a node finds out that the estimated delay is higher than a predefined threshold τTH , then the application layer places a flag in the application layer header, indicating that lower layers should treat the packet accordingly. Thus, upon the arrival of those packets to the MAC sublayer, it requests the physical layer to make changes in its parameters. In DRX, the MAC sublayer requests the physical layer to reduce the CCA duration from eight to four symbol periods (from 128 to 64 μs). In doing so, the physical layer senses the channel in half of the regular CCA duration and reports the results to the MAC sublayer. Thus, this node can acquire the channel and get to transmit its data before other contending nodes. If the node finds the channel busy, it invokes the BO algorithm as described in [20]. In this scheme, we assume that there are no devices transmitting at the same frequency band other than the IEEE 802.15.4 nodes to avoid any possible coexistence problems. Algorithm 1 describes the DRX scheme. Initially, the application layer evaluates the captured data and decides if the priority of the monitored parameter value Φ is beyond an acceptable threshold (i.e., higher or lower than normal limit values [23]). Then, the algorithm invokes the delay estimation process E[D]. If the estimated delay is found to be higher than the threshold τTH value (different delay thresholds for deferent smart-grid applications are obtained from [23] and used later in Section VII), then the CCA duration is divided by two; otherwise, the algorithm does not make any changes on the physical-layer parameters and transmits the data using a regular CCA duration process.

ti = [(2i+1 − 1)Wo + 3i − 1]/4 max (ρ, (1 − ρ)σ)i P˜ (Bi |Bt ) = m . k k=1 max (ρ, (1 − ρ)σ)

Algorithm 1: DRX Scheme (13)

Bi is the occurrence of a busy channel for the ith time and then an idle channel at the i + 1th time, and Bt is the successful sensing event in m attempts. The estimated delay from this analytical model that has been initially proposed in [22] is utilized by the medium-access schemes described in the following section. VI. P RIORITY- AND D ELAY-AWARE M EDIUM ACCESS IN WSANs We consider a WSAN that aims to monitor and control delaycritical data in a smart-grid environment. We assume that the data collected by certain sensors have high priority and should be delivered with minimum end-to-end delay. The presented schemes include an adaptation module which facilitates the interaction of the application layer with the MAC and physical layers. Those priority and delay-aware techniques aim to reduce the end-to-end delay by estimating the delay of critical data and then insuring that these data are delivered to the destination with minimum delay. Each node in the PAN initially implements the delay-estimation algorithm that estimates the

5

1 MEASURE THE VALUE OF Φ 2 if Φ > ΦTH then 3 // Invoke delay estimation algorithm 4 E[D] = P T D 5 if E[D] > τTH then 6 // Insert a flag in application layer header 7 APP Header = APP Header∗ 8 CCAduration = CAduration/2 9 MAC_CSMA-CA( ) 10 else 11 CCAduration = 8 symbol durations 12 MAC_CSMA-CA( ) 13 14 else 15 CCAduration = 8 symbol durations 16 MAC_CSMA-CA( ) 17 if CCA = successful then 18 Transmit Packet 19 else 20 if NB < MaxCSMABackoffs 21 go to 8 // repeat the CSMA-CA 22 else drop packet.

6

IEEE SYSTEMS JOURNAL

TABLE I I NITIAL S IMULATION PARAMETERS

The second presented scheme includes an improvement to the DRX scheme. The DRX scheme aims to reduce the endto-end delay without taking other nodes in the PAN into consideration. The proposed FDRX scheme can achieve the delay reduction and additionally allow other nodes to transmit fairly. Similar to the DRX scheme, the FDRX scheme initially implements the delay-estimation algorithm described in [22]. Based on the resulting values of the delay estimation, the MAC layer responds to the delay requirement of the application. The main difference between the DRX and the FDRX schemes is that the latter yields to other nodes in the PAN periodically to allow them to transmit. Thus, FDRX is fairer to other nodes. In the FDRX scheme, the MAC sublayer requests the physical layer to reduce the CCA duration from eight to four symbol periods (from 128 to 64 μs). This request is done based on a predefined yielding intensity α. The value α varies from zero to one; zero means the node is not yielding to other nodes (corresponds to DRX) and one means that node uses the default IEEE 802.15.4 MAC settings. The fairness property is added to ensure that a node only utilizes this scheme for a short period of time and then inverts back to default to allow other nodes to transmit. VII. P ERFORMANCE E VALUATION Fig. 2.

We implement the DRX and FDRX schemes in the QualNet [24] network simulator platform. The priority- and delay-aware medium-access schemes are tested with different numbers of nodes and traffic conditions. Furthermore, to investigate the performance of the DRX and the FDRX schemes in realistic smartgrid environments, we consider smart-grid-specific shadowing deviation and path-loss properties. In addition to that, the simulation parameters are selected similar to that of the analytical model described in Section V. We use a beacon-enabled star topology having N nodes and a coordinator, where N varies from 10 to 40 in each simulation scenario. We assume that the IEEE 802.15.4 MAC protocol is operating in the 2.4-GHz band with a maximum bit rate of 250 kbps. All nodes transmit constant bit rate (CBR) traffic. We assume that one node receives high-priority packets during the simulation time. The transmission range is set to 50 m, and all the nodes are in the same PAN. Each simulation is run for 300 s, and each result represents an average of ten runs. In the initial simulations, we set the delay threshold τTH to 0.400 s (following actual delay bound requirements presented in [23]). We assume that all nodes transmit with sufficient power, i.e., all nodes can reach the PAN coordinator. We also assume that noise factor is constant throughout the entire simulation. Table I shows the default parameters used in the simulations, and the remaining parameters are taken from [20]. We compare the performance of the presented schemes with an existing QoS supporting scheme [7] in terms of endto-end delay and packet delivery ratio. The scheme presented in [7] reduces the back-off time of a contending node to make it back off for a shorter period and then the rest of the nodes. The authors reduce the back-off time by reducing the value of the BE. We also compare our results to [8] where the authors reduce the number of CCAs performed in high-priority nodes from two to one and perform frame tailoring to avoid collision.

Average end-to-end delay.

Fig. 2 shows the relation between the average end-to-end delay and the number of nodes in the default IEEE 802.15.4 MAC settings, the modified back-off time (MBOT) scheme of [7], the single-CCA scheme [8], and the FDRX and DRX schemes. An obvious reduction in the end-to-end delay in the DRX scheme against the default IEEE 802.15.4 MAC settings and MBOT scheme is observed. Furthermore, there is a slight improvement in the delay when using DRX compared with the single-CCA scheme [8]. The significance of this delay reduction is illustrated more clearly in a smart-grid case study at the end of this paper. We also see that the DRX scheme has a higher impact on delay reduction compared with the FDRX (α = 0.5) scheme. α = 0.5 implies that FDRX is yielding 50% of the time. This higher delay reduction is obtained because the DRX scheme allows the node to utilize the channel more often and does not share the resources with other nodes in the PAN. On the other hand, FDRX is considered to be fair because it yields off to other nodes to allow them to transmit their data; hence, we observe that the delay reduction is less than the DRX scheme. It is also seen that the FDRX scheme performs slightly better than MBOT scheme. Fig. 3 shows the percentage of data packets received by the PAN coordinator (packet delivery ratio) from an individual node versus the number of nodes for the default IEEE 802.15.4 MAC settings, MBOT scheme, the single-CCA scheme [8], and the FDRX and DRX schemes. The packet delivery ratio drops as the number of nodes increase since the number of collisions is proportional to the number of nodes in the PAN. As seen in the figure, the DRX scheme performs better than the default IEEE 802.15.4 MAC settings and the single-CCA and MBOT schemes. We also observe that the FDRX scheme has a slightly higher percentage of delivery ratio than the default

AL-ANBAGI et al.: PRIORITY- AND DELAY-AWARE MEDIUM ACCESS FOR WIRELESS SENSOR NETWORKS

Fig. 3.

7

Packet delivery ratio. Fig. 5. Effect of yielding intensity α on the average end-to-end delay.

Fig. 4.

Effect of the CCA symbol duration on the average end-to-end delay.

IEEE 802.15.4 and MBOT scheme. Again, DRX performs better in terms of packet delivery ratio, because that node transmits at higher rate compared with other nodes in the PAN. The FDRX scheme comes next in terms of the packet delivery ratio because it is yielding to other nodes. To show the effect of different CCA durations and why we choose to divide by two in our schemes, we investigate the effect of reducing the CCA symbol duration on the average endto-end delay. Fig. 4 shows the effect of changing the CCA symbol duration from the default value (eight symbol durations) to the DRX value (four symbol durations). We can see that the average end-to-end delay starts to increase as the symbol duration increases. The results presented in Fig. 4 will assist in selecting an optimum value for the CCA symbol period that will minimize the end-to-end delay and maintain acceptable packet collision rate in the entire PAN. We further investigate the effects of the yielding intensity α of the FDRX scheme on the performance of the WSAN. This investigation will assist in optimizing the value of α to obtain certain delay bounds, packet delivery ratio, as well as packet collision rates. Fig. 5 shows the effect of yielding intensity α on the average end-to-end delay of a particular node implementing the FDRX scheme. As α increases, the average end-to-end delay also increases for all numbers of nodes. This is because when the yielding intensity approaches one, the scheme converges to the default setting, and as it approaches zero, it converges to DRX. Hence, based on the application and the delay bound requirements, we may select certain values of α which guarantee delay reduction and fairness at the same time.

Fig. 6. Effect of the yielding intensity α on the packet delivery ratio.

Fig. 7. Effect of the DRX and FDRX schemes on the energy consumption.

Fig. 6 shows the effect of the yielding intensity α on the packet delivery ratio of a particular node implementing the FDRX scheme. The results presented in this figure agree with the general behavior of the FDRX protocol, i.e., as α increases, the packet delivery ratio decreases. This is because the node implementing FDRX at lower α values will acquire the channel more often than the rest of the nodes and hence have a higher packet delivery ratio. We further investigate the effect of the DRX and FDRX schemes on the energy consumption of sensor nodes. We assume the use of the energy model of the MicaZ nodes [25]. In Fig. 7, the energy consumed in the transmit mode is slightly higher for DRX and FDRX schemes than the default settings since the nodes implementing these schemes will have the

8

IEEE SYSTEMS JOURNAL

Fig. 8. Effect of the yielding intensity α on the energy consumed. Fig. 10. Packets lost due to collision in the DRX scheme using energydetection method.

Fig. 9. Packets lost due to collision in the DRX scheme using carrier sensing with energy detection.

opportunity to transmit more often than their neighboring nodes. However, the increase in energy consumption is not significant (only 0.9%) compared with the increase in the packet delivery ratio and the reduction in the end-to-end delay. We also investigate the effect of different yielding intensities α on the energy consumed in a node implementing the FDRX scheme. The value of α can be adjusted according to the power requirements of individual nodes. Fig. 8 shows the effect of α on the energy consumed in the transmission mode. Again, as α approaches 1, we get a performance close to the default settings. In the previous set of results, we have investigated the effect of the DRX and the FDRX schemes on the performance of the node implementing these schemes. In the next set of results, we investigate the effect of implementing these two schemes on the overall WSAN performance in terms of the number of packets being lost at the sink due to collision. We use the same assumptions made in Section VII. Furthermore, to have a wider perspective, we compare the impact of two CCA methods, namely, CCA with energy detection and CCA with carrier sensing and energy detection, on the network performance. Fig. 9 shows the number of packets lost due to collision at the PAN coordinator in the entire WSAN. In this set of simulations, we use carrier sensing with energy-detection method. We observe that as the number of nodes increase, the number of packets lost due to collisions also increase, as expected. We also see that as the number of nodes increase, the packets lost in the DRX scheme becomes higher than the default IEEE 802.15.4 MAC settings. This slight increase of packet lost due to collisions is experienced by nodes that do not implement the DRX scheme since they

Fig. 11. Packets lost due to collision in the FDRX scheme using energydetection method.

fail to have their data transmitted to the PAN coordinator due to packet collisions. In the worst-case scenario, when the number of nodes is 40, the difference in the number of packets lost due to collision at the PAN coordinator is approximately 6%. However, for a lower number of nodes (10–20 nodes), the difference between the packet lost due to collision is negligible. Fig. 10 shows the number of packets lost due to collision at the PAN coordinator in the entire WSAN. In this simulation, we use the energy-detection method. We also see that as the number of nodes increase, the packet loss also increases and the difference is negligible at a lower number of nodes. However, we can see that the number of packets lost is very much higher than that of the carrier sensing with energy-detection method (Fig. 9). This agrees with the theory explained in Section III and presented in [19]. Figs. 11 and 12 show the number of packets lost due to collision at the PAN coordinator in the entire WSAN in the FDRX scheme for different yielding intensities with the energydetection and carrier-sensing methods, respectively. The trend of the results presented in these figures agrees with the general results presented previously. It is worthy to note that if the application requires certain bounds on the data delivery from the entire WSAN, we can choose certain values of α to maintain certain levels of packet collisions and end-to-end delay at the same time. We now investigate the performance of the DRX and FDRX schemes in a real smart-grid environment by taking the effect of the path loss models into consideration.

AL-ANBAGI et al.: PRIORITY- AND DELAY-AWARE MEDIUM ACCESS FOR WIRELESS SENSOR NETWORKS

9

TABLE II DATA PACKETS L OST D UE TO C OLLISION IN D IFFERENT E LECTRIC P OWER E NVIRONMENTS

Fig. 12. Packets lost due to collision in the FDRX scheme using carrier sensing with energy-detection method.

The path loss is defined as the difference (in decibels) between the transmitted and received powers, which represents the signal level attenuation caused by free space propagation, reflection, diffraction, and scattering. There are three types of path loss models, namely, empirical models which are based on data measurement, deterministic models which depend on the geometry of the site, and finally, semideterministic models which are based on empirical models in addition to deterministic models. In this paper, we investigate the performance of our two schemes in both empirical and deterministic path loss models. For the deterministic path loss model, we consider the two-ray path loss model for an outdoor environment (i.e., transformers in a substation) where there is normally two signal paths, one is direct from the sensor node to the sink and the other is reflected through a metal object or through the ground. The other deterministic path loss model that we consider is the TIA/ANSI Joint Technical Committee (JTC) path loss model for the personal communication service (PCS) bands for indoor areas, recommended by a technical working group for the 1900-MHz PCS bands [26]. We took the parameters for path loss calculations in the indoor as well as outdoor environments from [26]. For the empirical path loss model, we follow the work presented in [12], where the authors have conducted experiments with actual sensor nodes operating in the 2.4-GHz industrial, scientific, and medical bands with an effective data rate of 250 kbps. The work in [12] is performed to measure the link quality indicator and the received signal strength indicator with certain radio propagation parameters for different electric-power-system environments. Their experimental studies showed that their results provided more accurate multipath channel models than the Nakagami and Rayleigh models. Therefore, we simulate the DRX and FDRX schemes in similar environments to that of [12], namely, outdoor 500-kV substation environment, indoor main power room, and underground transformer vault. We use the following values for channel propagation parameters: outdoor substation (path loss = 3.51 and shadowing deviation = 2.95), indoor main power room (path loss = 2.38 and shadowing deviation = 2.25), and underground transformer vault (path loss = 3.15 and shadowing deviation = 3.19). We assume that the channel is having a lognormal shadowing model with a shadowing mean of 2.25 dB and that all sensor nodes are operating in non-line-of-sight mode.

TABLE III E ND - TO -E ND D ELAY VALUES FOR C RITICAL S MART-G RID A PPLICATIONS

Table II shows the number of data packets lost due to collision at the sink considering three different electrical power environments for the default settings and the DRX and FDRX schemes. In this scenario, we used 15 nodes and overloaded the nodes with CBR traffic to test the scheme in extreme traffic conditions. We observe that the DRX and FDRX schemes out perform the default settings. We also see that the results for the JTC path loss model is close to the indoor empirical path loss model and that that the two-ray model is somehow close to the outdoor 500-kV substation model for this simulation scenario. A. Case Study As a case study, we consider three critical smart-grid monitoring applications that have strict end-to-end delay requirements. These applications are capacitor bank control, fault current indicator, and transformer monitoring [23]. We obtained the functional requirements for these applications from [23]. We evaluate the performance of a WSAN with priority- and delay-aware medium-access schemes for those smart-grid applications. For this set of results, we consider a WSAN with 40 nodes in a star topology, where the sensor nodes monitor certain parameters such as current or voltage and transmit their data to a PAN coordinator. We assume that the PAN coordinator is connected to a high-speed network, e.g., ethernet passive optical network; hence, the delay from the PAN coordinator to the user is negligible. We simulate the WSANs using the default IEEE 802.15.4 MAC settings and the DRX and FDRX schemes. In Table III, we observe that the default IEEE 802.15.4 MAC setting has higher latency than the functional requirements of all applications, while both DRX and FDRX schemes succeed in reducing the latency below the functional requirements. The DRX and FDRX schemes are able to reduce the end-to-end delay by 60 and 35 ms, respectively.

10

IEEE SYSTEMS JOURNAL

VIII. C ONCLUSION In this paper, we performed a comprehensive performance evaluation of priority- and delay-aware medium-access techniques that respond to the delay requirements of smart-grid applications by predicting the end-to-end delay and creating cross-layer measures. These schemes achieve delay responsiveness by modifying the parameters of the physical layer of the IEEE 802.15.4 protocol. The first scheme, namely, DRX, first estimates the end-toend delay. If the packet is from a critical smart-grid application that has high priority and if the estimated delay cannot meet the delay requirements of this application, then DRX reduces the CCA duration in order to allow the high-priority packet to access the medium before other contending packets. The second approach, namely, FDRX, incorporates fairness into delaysensitive data transmission by yielding other nodes periodically. Our results show that DRX and FDRX schemes are able to reduce the delay for high-priority data while maintaining acceptable packet loss values. The DRX scheme has a higher effect on the end-to-end delay reduction compared with the FDRX. However, the DRX scheme does not take fairness into consideration; on the other hand, FDRX shows more flexible results and provides fairness to other nodes in the WSAN. The delay reduction achieved by DRX and FDRX will enhance the smart-grid operation in situations where sudden changes in loads or the generation cycle take place. The performance of DRX and FDRX schemes were further evaluated by taking the shadowing and path loss of actual electric power systems into consideration. The results of this evaluation showed that the DRX and FDRX outperformed IEEE 802.15.4 with the default settings in the considered smart-grid environment. We further performed a case study to show the effectiveness of the DRX and FDRX schemes in the reduction of the latency of actual smart-grid functional requirements. As a future work, we intend to study the performance of the proposed schemes in the existence of interfering nodes such as WiFi devices. Furthermore, we plan to implement those schemes in a smart-grid test bed. R EFERENCES [1] Y. Zhang, R. Yu, M. Nekovee, Y. Liu, S. Xie, and S. Gjessing, “Cognitive machine-to-machine communications: Visions and potentials for the smart grid,” IEEE Netw. J., vol. 26, no. 3, pp. 6–13, May/Jun. 2012. [2] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “An applicationspecific protocol architecture for wireless microsensor networks,” IEEE Trans. Wireless Commun., vol. 1, no. 4, pp. 660–670, Oct. 2002. [3] C. Wang, J. Liu, J. Kuang, A. S. Malik, and H. Xiang, “An improved LEACH protocol for application-specific wireless sensor networks,” in Proc. 5th Int. Conf. WiCom. Netw. Mobile Comput., Beijing, China, Sep. 24–26, 2009, pp. 1–5. [4] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, “A low latency data transmission scheme for smart grid condition monitoring applications,” in Proc. IEEE Annu. EPEC, London, ON, Canada, Oct. 10–12, 2012, pp. 20–25. [5] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, “Fairness in delayaware cross layer data transmission scheme for wireless sensor networks,” in Proc. 26th QBSC, Kingston, ON, Canada, May 28/29, 2012, pp. 146–149. [6] W. Sun, X. Yuan, J. Wang, D. Han, and C. Zhang, “Quality of service networking for smart grid distribution monitoring,” in Proc. 1st IEEE Int. Conf. Smart Grid Commun., Gaithersburg, MD, USA, Oct. 4–6, 2010, pp. 373–378.

[7] M. Youn, Y. Oh, J. Lee, and Y. Kim, “IEEE 802.15.4 based QoS support slotted CSMA/CA MAC for wireless sensor networks,” in Proc. Int. Conf. SENSORCOMM, Valencia, Spain, Oct. 14–20, 2007, pp. 113–117. [8] T. H. Kim and S. Choi, “Priority-based delay mitigation for eventmonitoring IEEE 802.15.4 LR-WPANs,” IEEE Commun. Lett., vol. 10, no. 3, pp. 213–215, Mar. 2006. [9] I. S. Hammoodi, B. G. Stewart, A. Kocian, and S. G. McMeekin, “Wireless sensor networks for partial discharge condition monitoring,” in Proc. 44th UPEC, Glasgow, U.K., Sep. 2009, pp. 1–5. [10] M. Erol-Kantarci and H. T. Mouftah, “Wireless multimedia sensor and actor networks for the next-generation power grid,” Ad Hoc Netw. J., vol. 9, no. 4, pp. 542–511, Jun. 2011. [11] M. Erol-Kantarci and H. T. Mouftah, “Wireless sensor networks for costefficient residential energy management in the smart grid,” IEEE Trans. Smart Grid, vol. 2, no. 2, pp. 314–325, Jun. 2011. [12] V. C. Gungor, B. Lu, and G. P. Hancke, “Opportunities and challenges of wireless sensor networks in smart grid,” IEEE Trans. Ind. Electron., vol. 57, no. 10, pp. 3557–3564, Oct. 2010. [13] I. Al-Anbagi, H. T. Mouftah, and M. Erol-Kantarci, “Design of a delaysensitive WSN for wind generation monitoring in the smart grid,” in Proc. IEEE CCECE, Niagara Falls, ON, Canada, May 8–11, 2011, pp. 001370–001373. [14] M. C. Vuran and I. F. Akyildiz, “XLP: A cross-layer protocol for efficient communication in wireless sensor networks,” IEEE Trans. Mobile Comput., vol. 9, no. 11, pp. 1578–1591, Nov. 2010. [15] I. F. Akyildiz, M. C. Vuran, and O. B. Akan, “A cross-layer protocol for wireless sensor networks,” in Proc. 40th Annu. Conf. Inf. Sci. Syst., Princeton, NJ, USA, Mar. 2006, pp. 1102–1107. [16] M. L. Sichitiu, “Cross-layer scheduling for power efficiency in wireless sensor networks,” in Proc. 23rd IEEE Annu. Joint Conf. Comput. Commun. Soc. INFOCOM, Hong Kong, Mar. 7–11, 2004, vol. 3, pp. 1740–1750. [17] V. Rao and D. Marandin, “Adaptive channel access mechanism for ZigBee (IEEE 802.15.4),” J. Commun. Softw. Syst., vol. 2, no. 4, pp. 283–293, Dec. 2006. [18] I. Ramachandran and S. Roy, “On the impact of clear channel assessment on MAC performance,” in Proc. IEEE GLOBECOM, San Francisco, CA, USA, Nov. 27/Dec. 1, 2006, pp. 1–5. [19] I. Ramachandran and S. Roy, “Clear channel assessment in energy constrained wideband wireless networks,” IEEE Wireless Commun., vol. 14, no. 3, pp. 70–78, Jun. 2007. [20] Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs), IEEE Std 802.15.4-2006, 2006. [21] H. Karl and A. Willig, Protocols and Architectures for Wireless Sensor Networks. Hoboken, NJ, USA: Wiley, 2005. [22] P. Park, P. Di Marco, P. Soldati, C. Fischione, and J. K. Henrik, “A generalized Markov chain model for effective analysis of slotted IEEE 802.15.4,” in Proc. 6th Int. Conf. MASS, Macau, China, Oct. 12–15, 2009, pp. 130–139. [23] M. Oldak and B. Kilbourne, “Communications Requirements Comments of Utilities Telecom Council,” Department of Energy, Washington, DC, USA, Jul. 2010. [24] QualNet Network Simulator. [Online]. Available: http://web. scalable-networks.com/content/qualnet [25] “Micaz Wireless Measurement System,” San Jose, CA, USA, 6020-006004 Rev A, 2004. [26] K. Pahlavan and A. H. Levesque, Wireless Information Networks, 1st ed. Hoboken, NJ, USA: Wiley, Jan. 1995.

Irfan Al-Anbagi (M’03) received the M.Sc. degree in electronic engineering from the University of Technology, Baghdad, Iraq. He is currently working toward the Ph.D. degree in electrical engineering at the School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, ON, Canada. His main research interest includes the design and development of ad hoc and sensor networks, the development of QoS schemes for WSNs, and smartgrid communication systems. From 2003 to 2010, he was a Senior Lecturer and a Program Leader with the College of Engineering, Caledonian University, Sultanate of Oman.

AL-ANBAGI et al.: PRIORITY- AND DELAY-AWARE MEDIUM ACCESS FOR WIRELESS SENSOR NETWORKS

Melike Erol-Kantarci (M’08) received the M.Sc. and Ph.D. degrees from the Department of Computer Engineering, Istanbul Technical University, Istanbul, Turkey, in 2004 and 2009, respectively. She is a Postdoctoral Fellow with the School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, ON, Canada. During her Ph.D. studies, she was a Fulbright Visiting Researcher with the Department of Computer Science, University of California, Los Angeles, CA, USA. She has over 40 referred journal articles and conference papers. Her main research interests are wireless sensor networks, smartgrid communications, cyber-physical systems, and underwater sensor networks. She is the Vice-Chair of the Women in Engineering Group of the IEEE Ottawa Section.

11

Hussein T. Mouftah (F’90) received the D.Sc. degree from Laval University, Quebec City, QC, Canada, in 1975. He has been with the School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, ON, Canada, since 2002, as a Senior Canada Research Chair and Distinguished University Professor. He was with the Department of Electrical and Computer Engineering, Queen’s University, Kingston, ON, in 1979–2002. He has three years of industrial experience mainly at the Bell-Northern Research of Ottawa (in 1977–1979). He is the author or coauthor of 8 books, 59 book chapters, and more than 1200 technical papers and 12 patents in this area. Dr. Mouftah is a Fellow of the Canadian Academy of Engineering, the Engineering Institute of Canada, and the Royal Society of Canada: The Academy of Science.