Cross-Layer QoS Optimization of Wireless Sensor Network for Smart Grid

Abstract

Wireless Sensor Network (WSN) has been considered as a promising solution for Smart Grid due to the significant advantages over traditional communication technologies. However, as a necessary function for monitoring, diagnosing, and controlling of remote electrical equipment, Quality of Service (QoS) is still a challenging problem in applying WSN in Smart Grid. This paper presents a comprehensive experimental study on optimization of multiobject QoS in cross-layers of WSN protocol. In PHY layer, a Link Quality Estimator is studied for the purpose of evaluating link quality under harsh Smart Grid environments. In MAC layer, a QoS Enhanced MAC algorithm is used to prevent disorderly contentions among nodes and to allocate more channel resources for critical traffic; in NWK layer, a routing algorithm is introduced for multiobject QoS optimization. All the algorithms are cooperated with each other in cross-layer manner to improve end-to-end reliability and latency. In comparison with other counterparts, the proposed algorithms achieve preferable performances. Finally, field experiments were conducted, the result of which showed that the proposed algorithms outperform the ZigBee protocol and the original Distribution Grid wireless communication technology in terms of the QoS performance of critical traffic.

1. Introduction

With the features of rapid self-organizing and superior coordinating ability, Wireless Sensor Network (WSN) has a wide application prospect in the field of remote supervisory control and fault diagnosis for Smart Grid. In 2012, WSN was listed as the future Home Area Network of Smart Grid by NIST [1]. And in the same year, IEEE published IEEE802.15.4g Low-Data-Rate Smart Metering Utility Networks standard in local and metropolitan area networks [2].

Recent advances of WSN in Smart Grid make it feasible to conduct the embedded electric utility monitoring and communicating. However, there are still some challenges in applying WSN in Smart Grid, such as harsh environmental conditions, reliability and latency requirements, packet errors and variable link capacity, and resource constraints [3]. Recent researches [4–8] mainly concentrated on the application of WSN in electrical power grid. However, an overlooked question is the analysis of varied quality of service (QoS) requirements for diversified communication traffics under complex environments. Since IEEE802.15.4 standard does not support QoS, several studies [9, 10] were conducted to provide QoS for WSN. However, some uncertain interference factors on wireless link in harsh environment were not taken into account.

As a matter of fact, with limited resources and lower boundary requirements, the study of WSN-based Smart Grid application should resolve the multiobjective (reliability and latency) end-to-end QoS optimization problems in multihop transmission with uncertain processes. To achieve the goal, a comprehensive experimental study on cross-layer based multiobjective QoS optimization was carried out in Smart Grid communication. This study sets two different classes of Smart Grid communication packets in cross-layer optimization by their reliability and latency QoS requirements. All the algorithms were implemented in three layers of WSN protocol, namely, physical (PHY) layer, media access control (MAC) layer and network (NWK) layer. First, a link quality estimator (LQE) was introduced to evaluate link quality under harsh environments. Then, a QoS enhanced MAC (QoS-MAC) algorithm was used to avoid disorderly contention among WSN nodes and to effectively allocate limited wireless channel resources to two classes of packets. Lastly, reliability and latency estimators were designed to provide estimation information for link quality routing (LQ-routing) algorithm, which optimized the multiobject end-to-end QoS performance in the multihop transmission. Figure 1 shows the cross-layer QoS optimization framework, in which IEEE802.15.4-based WSN is adopted.

Figure 1

Cross-layer QoS optimization framework of WSN.

The rest of this paper is organized as follows. Section 2 introduces the related works. Section 3 presents the analysis of the affecting factors on link quality and the LQE for evaluating link quality in harsh Smart Grid environments. Sections 4 and 5 discuss the enhancement of QoS in MAC layer and NWK layer by the proposed QoS-MAC algorithm and LQ-routing algorithm, respectively. Section 6 explains the field experiments conducted to verify the proposed cross-layer QoS optimization of WSN. Lastly, Section 7 is the conclusion of the paper.

2. Related Works

This section introduces relevant existing works on WSN from the perspectives of link quality estimator, QoS-based MAC protocols, QoS-based routing optimization and QoS-based optimization of WSNs for Smart Grid.

2.1. Link Quality Estimator

LQE, as a way of evaluating media channel of WSN, has triggered a comprehensive study [11]. In general, LQEs can be classified as either hardware-based or software-based.

Received signal strength indicator (RSSI) and link quality indicator (LQI) are common metrics of hardware-based LQEs, which can be directly read from hardware registers of transceiver. It takes them much less time to estimate link quality than software-based counterparts. However, RSSI and LQI are acquired based on part samples of a successfully received packet, which cannot reflect link quality status when the packet is lost [12–14]. Besides, they have no clear-cut relation with link quality status. Although packet reception ratio (PRR) can be theoretically computed by SNR [11], many popular IEEE802.15.4-compatibility chips (e.g., CC25XX, EM35X) do not provide SNR value. In addition, hardware-based LQEs are easily affected by the quality of transceiver hardware circuit, the directivity of antenna, and environmental factors.

PRR and RNP (required number of packets) are basic metrics of software-based LQEs. PRR is defined as ratio of the number of successfully received packets to all transmitted packets in receiver side, while RNP is defined as the ratio of the number of transmitted and retransmitted packets to the number of successfully received packets, minus 1, in sender side [15]. Expected transmission count (ETX) [16] is another software-based LQE, which is evaluated by both forward and backward PRR values. Those three software-based LQE metrics can better reflect the condition of wireless communication links. However, the estimation processes of software-based LQEs usually take additive latency to collect enough statistical information in improving estimation accuracy, which may be worse when the link quality fluctuates. To address this problem, window mean with exponentially weighted moving average (WMEWMA) [17], Kalman filter [18], Fuzzy Logic [19], and so forth, are adopted to estimate link quality metrics. However, in practice, Kalman filter and Fuzzy Logic are complex and require important storage resources. WMEWMA has the advantage of less complexity in computation, less storage spaces and faster response ability. It not only takes the past link quality status into account, but also reflects the rapid changes of estimation object, which means it is more suitable for WSN devices with limited computation capability.

The reliability requirements of Smart Grid communication are usually evaluated by packet success rate (which equals PRR), as a result of which, PRR is a more intuitional metric than its hardware-based and software-based counterparts for the algorithms of upper layer protocol. In this study, PRR is selected as the metric of LQE. Besides, WMEWMA is adopted to estimate link quality and to balance the performances of accuracy and the efficiency of estimation results in [17]. But the LQE in this study is executed in sender side; thus the overhead of sending back estimation result could be avoided.

2.2. QoS-Based MAC Algorithm

MAC protocol provides channel access control mechanisms, which makes it possible for nodes to communicate within WSN incorporating a shared medium. In general, QoS MAC protocols can be classified as schedule-based or contention-based protocols.

Time division multiple access (TDMA) algorithm [20–23] is mostly adopted for schedule-based MAC protocol. In TDMA algorithm, the access is divided into different time slots for nodes. Each node accesses media channel at its own time slot to avoid collision and to bring controllable transfer delay. Although TDMA protocols are advantageous in collision-free medium access, they suffer heavily from the following problems, such as clock synchronization, channel under-utilization, and fixed time-slot assignments [24]. Rate allocation algorithms [25–28] are another schedule-based MAC protocol, which assigns different rates to users on demand in optimizing the communication delay. However, it requires abundant information exchange, which visibly increases the complexity of protocol and additional overhead to achieve QoS in MAC.

Unlike schedule-based algorithm, contention-based QoS MAC protocols access media channel in a random contention manner. Several subsequent studies based on S-MAC [29, 30] adopt duty-cycling mechanism (synchronous sleep and wakeup for communication) with the aim of balancing the transfer delay and energy consumption in WSN. However, the periodical sleep-and-wakeup would result in significant latency. Furthermore, the nodes maintaining synchronous sleep and wakeup also increase additional overhead as TDMA algorithm. Sleepless random access MAC protocols allow the obtaining of significant QoS performance in comparison with duty-cycling [31]. The sleepless feature makes the protocol more energy-consuming, while, in Smart Grid, energy can be acquired from low-voltage equipment, which make it more appropriate compared with sleep-and-wakeup MAC. However, the random contention feature of CSMA/CA in sleepless mode of IEEE802.15.4 standard would lead to more or less influence in the performance of critical traffic by other regular traffics.

In this paper, a novel QoS-MAC is explored by introducing unfair contention among nodes. The QoS-MAC is aimed at maximizing the utilization of media channel and reducing the contention influence on critical traffic. Besides, the proposed MAC algorithm can be operated in the distribution way with no additional overhead.

2.3. QoS Based Routing Protocol

Routing is a key technology in achieving multihop transmission in WSN. Many routing solutions, including negotiation-based routing [32], query-based routing [33], location-based routing [34], multi-path-based routing, and ant-based routing [35] are proposed in WSN to balance the predetermined QoS metrics, such as delay, energy, reliability, and bandwidth, before delivering the data to the sink node [36]. Of those routing protocols, location-based routing protocols, which use localization information to select the next router from neighboring nodes geographically closer to the destination [37], are efficient in networks with stationary nodes. Therefore, location-based routing algorithm is suitable for Smart Grid application, as the geographical position of electrical equipment and utilities is known and almost immobile.

A lot of efforts, for example, SPEED [38] and MMSPEED [39] et al., are made to support QoS on location-based routing. In SPEED routing protocol, each node keeps the geographic information of its immediate neighbor to make routing decision. SPEED is designed to transmit packets through routes at a given speed to reduce delay. However, SPEED does not consider another QoS metric, reliability. MMSPEED attempts to improve SPEED by multispeed routing according to the delay requirement of packets when providing reliability QoS by multipath routing. MCMP [40] is another location-based routing protocol, which aims to provide multiconstrained QoS for different packets as path information is not readily available in wireless networks. MCMP utilizes multiple paths between sources and sink pairs for QoS provisioning. However, multipath routing may reduce the throughput of network. What is more; packets may be highly congested nodes near to sink. To prevent such congestion, LOCALMOR protocol [41] addresses reliability, namely, single-path multiple sink by a different approach, which increases sink nodes to reduce congestion. Moreover, link reliability and latency estimators are considered in the routing metrics. As a matter of fact, LOCALMOR reduces the range collected by a sink node.

All the location-based protocols mentioned above do not distinguish the different requirements of application traffic. The information of MAC layer and PHY layer are not both considered in routing decision. In this study, reliability estimator and latency estimator are introduced to provide cross-layer information required. Then a routing algorithm named LQ-routing is presented for multiconstrained QoS optimization to improve the performance of reliability and latency for critical Smart Grid traffic.

2.4. QoS Based Optimization of WSNs for Smart Grid

Various traffics of Smart Grid have different QoS requirements on communication latency and reliability. Ericsson classified the communication of Smart Grid into 3 types (real-time operational, administrative operational, and administrative communications) with different QoS requirements [42]. Detailed latency and reliability requirements of Smart Grid applications presented in Table 1 [43] are different with each other as well. Therefore, satisfying the QoS requirements is one of major technical challenges of WSNs in Smart Grid applications.

Table 1

Smart Grid applications and their QoS requirements.

Application	Delay (s)	Packet success rate (%)
Smart metering (AMI)	2	99-100
Price signaling	2	99
Automated feeder switching	1-2	99
Demand response	2	99-100
Emergency response	0.5	99-100
Residential energy management	1	90
Building automation	1-2	90
SCADA	2–5	99
Distributed generation	2	99
Distributed management	2	99
Overhead transmission link monitoring	1	90
Outage management	2	99

Many WSN-based Smart Grid studies like [6–8] are concerned with applications, but they neglect the fact that the IEEE802.15.4 standard of WSN is without QoS scheme.

In literature, some QoS optimization studies are carried out with the aim to provide QoS for Smart Grid in layered WSN protocols. For MAC layer, a MAC algorithm is presented in [44] to add the QoS to WSN protocols by providing differentiated service for Smart Grid traffic of different priorities. Two medium-access approaches, namely, DRX and FDRX data transmission, are introduced in [45] to deal with the delay and service requirements of Smart Grid. For NWK layer, the well-known on-demand, table-driven, and QoS-aware routing protocols are investigated in [46] in terms of packet delivery ratio, end-to-end delay, and energy consumption to show the advantages and disadvantages of each routing protocol type in different Smart Grid spectrum environments. Multipath and single-path QoS-aware routing algorithms are investigated in [47] to evaluate their service differentiation capabilities in reliability and timeliness domains under harsh Smart Grid environmental conditions.

The above layered QoS studies [44–47] try to improve and analyze the adaptability and performance of WSN in Smart Grid application. However, optimization in a layer does not represent the best solution for the entire WSN protocol stack. Consequently, cross-layer QoS optimization becomes increasingly important. A recent cross-layer study for Smart Grid [45] is mainly focused on spectrum access in Cognitive Radio Sensor Networks for Smart Grid. However, it does not explore the end-to-end performance of media access and multihop routing.

Therefore, in this paper, we aim to introduce a cross-layer optimization in PHY, MAC, and NWK layer protocols to provide a better end-to-end QoS performance for multihop WSN in Smart Grid.

3. Wireless Link Quality Evaluation in PHY Layer

In this section, we first analyze the affecting factors on Link Quality metric PRR. Then an LQE is presented to evaluate link quality between neighbor nodes.

3.1. Affecting Factors on Link Quality in Smart Grid

In IEEE802.15.4 based WSN, OQPSK modulation is employed by PHY layer. The bit error rate (BER) of OQPSK can be calculated as [48]

\begin{matrix} P_{e} = Q (\sqrt{2 \cdot γ_{(d)} \cdot \frac{B_{N}}{R}}), \end{matrix}

(1)

where $B_{N}$ denotes noise bandwidth of transceiver. R is the data rate of communication in kbps. $γ_{(d)}$ denotes SNR of radio communication as a function of distance d between transmitter and receiver. $γ_{(d)}$ is defined as

\begin{matrix} γ_{(d)} = 1 0^{(P_{r} (d) - P_{n}) / 10}, \end{matrix}

(2)

where $P_{r} (d)$ is the received signal strength as a function of distance d in dBm, while $P_{n}$ is the Background noise strength in dBm.

In order to verify integrity of packet, CRC (cyclic redundancy check) is adopted in IEEE802.15.4. An error may lead to the failure of packet reception. Therefore, the packet reception rate can be calculated as

\begin{matrix} P_{rr} = {(1 - P_{e})}^{8 l}, \end{matrix}

(3)

where l is the packet size in byte.

Consequently, theoretical Link Quality model can be expressed as (4), which is derived by inserting (1) and (2) into (3):

\begin{matrix} P_{rr} = {(1 - Q (\sqrt{2 \cdot 10^{(P_{r} (d) - P_{n}) / 10} \cdot \frac{B_{N}}{R}}))}^{8 l} . \end{matrix}

(4)

As a whole, the affecting factors on Link Quality are the background noise $P_{n}$ and the received signal strength $P_{r} (d)$ . In this study, $P_{n}$ and $P_{r} (d)$ in different Smart Grid environments are measured and analyzed as follows.

3.1.1. Background Noise in Smart Grid Environments

The noise interfering with WSN communication may be derived from background noise of environment and the interference of other WSN nodes. To measure background noise in Smart Grid environment, only one CC2530 based WSN node that is working in receive energy scan mode is used to avoid interference from other homogenous nodes. The measured channel is set as the 26th channel of IEEE802.15.4 to avoid interference from WiFi devices, since the 26th channel does not overlap with IEEE802.11 channel. The background sampling interval is 20 ms, and the background sampling result is the maximum received energy value in sampling duration of 0.2 ms. Figure 2 shows three kinds of Smart Grid environments, where background noise is measured by CC2530 based WSN node. Figures 3(a), 3(b), and 3(c) present Background Noise samples in the respective Smart Grid environments. The bar charts in Figures 3(d), 3(e), and 3(f) express corresponding relative frequency (RF) of the measured background noise energy in three Smart Grid environments. Background noise in Smart Grid environments fluctuates rapidly with kind of impulse features. It may be increased by 10–20 dBm with low relative frequency. Measured energy of background noise beside distribution lines is greater than that of other 2 environments about 2 dBm, which may result in the reducing of SNR by 1.5 times computed by (2).

Figure 2

Smart Grid environments of experiments. (a) Distribution line, (b) substation, (c) residence area.

Figure 3

Background noise samples and corresponding relative frequency in three Smart Grid environments.

3.1.2. Received Signal Strength in Smart Grid Environments

By log-distance path loss model, received signal strength can be expressed as

\begin{matrix} P_{r} (d) = P_{t} + L_{c} - PL (d_{0}) - 10 n \log_{10} (\frac{d}{d_{0}}) + X_{σ}, \end{matrix}

(5)

where $P_{r} (d)$ is wireless received signal strength in dBm with distance of d in kilometer. $P_{t}$ is transmission power of sender in dBm. $L_{c}$ is the actual power loss in antenna and circuit. $PL (d_{0})$ is free space path loss in dB at the reference distance $d_{0}$ in kilometer, which can be computed by $PL (d_{0}) = 32.44 + 10 n_{0} \log_{10} (d_{0} (km) \cdot f (MHz))$ (in this study $d_{0} = 1$ km). n is the path loss exponent, which is environment related. $X_{σ}$ indicating shadow fading caused by multipath interference is a normal random variable with zero mean and standard deviation σ in dB.

We use a pair of CC2530 based WSN nodes to estimate the actual power loss $L_{c}$ , path loss exponent n and standard deviation σ of shadow fading $X_{σ}$ . In order to improve accuracy, the distances d of two nodes in (5) are set as 2.5 m, 4.0 m, 6.3 m, 10.0 m, 15.8 m, 25.1 m, 39.8 m, and 63.1 m, respectively. ( $10 \cdot \log_{10} (d)$ is linearly increased from 4 to 18) and the height is set as 2 m to measure nine sets of received signal strength data in each Smart Grid environment. And then, those data tested in different distances are used for parameter estimation by ordinary least square (OLS). With 95% confident level, the estimated values are summarized in Table 2.

Table 2

Estimated parameters of signal strength in Smart Grid environments.

Smart Grid environments	$\hat{n}$	$\hat{σ}$	${\hat{L}}_{c}$
Distribution line	2.465	2.439	−15.092 dBm
Power station	2.253	2.263	−15.225 dBm
Resident area	2.357	2.504	−15.065 dBm

In Table 2, the estimated hardware-related actual power loss $L_{c}$ is relatively invariant, which is reasonable since the same experimental hardware is used in three Smart Grid environments. The environment-related parameters (path loss exponent n and standard deviation σ of shadow fading $X_{σ}$ ) are varied in different environments. It is important to note that a little change of exponent n will seriously influence link quality. For example, n increases by 0.1 will result in the decreases of SNR by 1.5 times.

It can be concluded by the experiment and analysis above that the Background Noise and received signal strength are mainly affected by different factors in Smart Grid environment, when the WSN node is homogeneous. Furthermore, Link Quality varies dramatically in accordance with changes in environments. This observation justifies the need of LQE for QoS Optimization in WSN based Smart Grid.

3.2. Link Quality Estimator

In our LQE, PRR is used as the metric of Link Quality. ${Prr}_{v_{i}, v_{j}}$ denotes the link quality estimation result computed by node $v_{i}$ between $v_{i}$ and its neighbor $v_{j}$ . And ${prr}_{v_{i}, v_{j}}$ represents the last computed PRR between $v_{i}$ and $v_{j}$ in the window $w_{p}$ . Supposing that $v_{i}$ receives r ACK message from $v_{j}$ after it transmits $w_{p}$ packets to $v_{j}$ , ${prr}_{v_{i}, v_{j}}$ is described as

\begin{matrix} {prr}_{v_{i}, v_{j}} = \frac{r}{w_{p}} . \end{matrix}

(6)

The selected value of $w_{p}$ will critically affect the computed result of ${prr}_{v_{i}, v_{j}}$ . For example, if $w_{p}$ is too great, the time spent on computing ${prr}_{v_{i}, v_{j}}$ will increase. Accordingly, the sensitivity of LQE drops. On the other hand, if $w_{p}$ is too small, the changes of r will make ${prr}_{v_{i}, v_{j}}$ fluctuate in a comparatively large range, which affects the stability of LQE result. By combining sensitivity, stability of LQE, and field test result in this paper, the value of $w_{p}$ is selected as 10.

After transmitting $w_{p}$ packets to $v_{j}$ , $v_{i}$ resets the count number for window $w_{p}$ and updates the Link Quality estimation result ${Prr}_{v_{i}, v_{j}}$ by

\begin{matrix} {Prr}_{v_{i}, v_{j}} = β_{p} \cdot {Prr}_{v_{i}, v_{j}} + (1 - β_{p}) \cdot {prr}_{v_{i}, v_{j}}, \end{matrix}

(7)

where $β_{p}$ representing the degree of weighting decrease is a constant smoothing factor between 0 and 1. A lower $β_{p}$ discounts older ${Prr}_{v_{i}, v_{j}}$ faster, but the estimation result accuracy may be reduced. A higher $β_{p}$ takes more history value of ${Prr}_{v_{i}, v_{j}}$ into account, but the estimation result cannot efficiently reflect the change of ${Prr}_{v_{i}, v_{j}}$ . To balance the performance of accuracy and efficiency of estimation results, $β_{p}$ is selected as 0.6 in this paper.

Different from PRR of LQE introduced in [15], the LQE is computed in sender side, which can be used to reduce the overhead of exchanging link quality information among nodes.

4. Quality of Service in MAC Layer

Smart Grid traffics require that WSN's MAC Protocol provides QoS with latency bounds for various traffics, which means that the node sending critical packet should preferentially access media channel. It is important to guarantee the reporting time within a given bound even in the worst case. This section presents a novel QoS-enhanced MAC for IEEE802.15.4. The QoS-MAC protocol provides differentiated services for data traffic with varied priorities.

4.1. CSMA/CA of IEEE802.15.4

IEEE802.15.4 MAC layer protocol supports two medium access modes: the slotted mode (beacon-enable mode) and unslotted mode (nonbeacon-enable mode) [49]. In the slotted mode, the MAC layer accesses the medium by a beacon-based superframe structure. The superframe is divided into active and inactive periods. During an inactive period, the node turns off the device radio and then sleeps in order to save energy. In the unslotted mode, arbitration of medium access is achieved by the CSMA/CA scheme [49], since nodes never sleep. The unslotted CSMA/CA flow chart is shown in Figure 4.

Figure 4

Unslotted CSMA/CA flow chart of IEEE802.15.4.

When there is a packet needing to be transmitted, node starts CSMA/CA by initializing number and period backoff. Then it delays a backoff period and performs CCA (clear channel assessment). If channel is idle, the node starts to transmit the packet, otherwise, it would adjust backoff period and restart it until it reaches the max number of backoffs. If max number of backoffs is reached and the channel is still busy, the packet will be dropped.

4.2. QoS-MAC Algorithm

In our QoS-MAC, the data packets will be buffered in the two queues based on their requirements for timely delivery: the critical traffic, considered as high priority (HP) packets (require immediate delivery with guaranteed delay and delivery rate), is pushed into HP queue; the regular traffic, considered as low priority (LP) packets (which typically have more tolerance in network delay and delivery rate), is pushed into LP queue.

Our QoS-MAC algorithm has three rules, one for competition within the node, and the other two for competition within a network among different nodes.

Rule 1.

For CSMA/CA mechanism in the node, any packets in the LP queue will not be serviced if the queue of the HP packets is not empty.

Rule 2.

When competing with each other for channel access in a network, the nodes with HP packets have fixed short backoff period, more backoffs, and more frequent CCA detection, while the nodes with LP packet use long random backoff period, less backoffs, and less frequent CCA detection.

Rule 3.

In CSMA/CA, CCA detection time of node sending HP packet is short, meanwhile CCA detection time of node sending LP packet is longer than the sum of backoff periods and CCA detection time of node sending HP packet.

These CSMA/CA rules are simple enough to be embedded into any network stack without additional expenses (i.e., overhead). Figure 5 shows the time frame of four nodes contending the channel access during the same period. Assuming that Node 1 and Node 2 have HP packets to be sent while Node 3 and Node 4 only have LP packets to be sent. At the beginning of the time fragment, Node 3 occupies the medium, later Node 1 and Node 2 receive their HP packets and start to service them. After they detect that the channel is busy, their CSMA/CA initiates the backoff period. At time 0, Node 3 completes its transmission and releases the channel for contention and automatically begins backoff for its next LP packet if there is any. During time slots 0 and 1, although Node 4 started its CCA detection first, Node 1 has shorter CCA detection interval because it has HP packets to be sent (Rule 2). Though Node 2 also has HP packets to be sent, its CCA finished before Node 3 releases the channel. Hence, at time 1, Node 1 starts its transmission of HP packets by interrupting the service of Node 4 (sending LP packets), which is forced by another backoff period after its CCA detection ends.

Figure 5

Nodes channel access competitive timing diagram.

At time 2 Node 1 finishes transmission and releases the channel before starting its short backoff time. In this case, Node 2 acquires the right to use wireless channel since Node 3 has not yet finished its CCA detection time (servicing LP data) even though its CCA starts earlier. Node 4 has not finished its backoff period either.

At time 4, Node 4 finishes its backoff period and starts CCA detection right after Node 2 finishes transmission and releases the channel. At this time, Node 1 still detects a busy channel because it performs CCA detection before Node 2 releases the channel.

Although Node 2 starts CCA later than Node 4 between times 4 and 5, it sends its HP packets again at time 5 since the CCA detection period and the backoff period of Nodes 3 and 4 (both servicing LP packets) have been increased (Rule 3) to ensure the channel will be open for nodes servicing HP data.

The QoS-MAC algorithm realizes unfair contention according to the priority of traffic among nodes by those 3 rules, which maximize the utilization of media channel and reduce the contention influence on HP traffic. More importantly, they are simple and easy to be transplanted in IEE802.15.4 with no additional overhead.

4.3. Collision Rate Estimation

In unslotted CSMA/CA of IEEE802.15.4 standard, attempting transmission packet is collided and then dropped when it reaches the maximum numbers that the CSMA/CA algorithm was required to backoff. The Collision Rate of HP packet is updated in a window of $w_{d}^{h}$ transmissions. $n_{d}^{h}$ denotes number of packets dropped in $w_{d}^{h}$ transmissions. The Collision Rate $c r_{v i}^{h}$ of HP packet of node $v_{i}$ in counting window $w_{d}^{h}$ is expressed as $c r_{v_{i}}^{h} = n_{d}^{h} / w_{d}^{h}$ . Hence, after $w_{d}^{h}$ transmissions complete, node $v_{i}$ updates its dropping rate by the WMEWMA method express as

\begin{matrix} c r_{v i}^{h} = β_{c} \cdot c r_{v i}^{h} + (1 - β_{c}) \cdot \frac{n_{d}^{h}}{w_{d}^{h}}, \end{matrix}

(8)

where $β_{c}$ represents the degree of weighting decrease and $0 < β_{c} < 1$ . Similar with $w_{p}$ and $β_{p}$ , the values of $w_{d}^{h}$ and $β_{c}$ affect the sensitivity and stability of $c r_{v i}^{h}$ as well.

4.4. MAC Layer Latency Estimation

The latency of packet in MAC layer includes (1) channel access delay: the time of the packet waiting for channel access contention in CSMA/CA, (2) transmission delay: the time of the transmitter from starting to send packet to receiving ACK. For node $v_{i}$ , $ca d_{v_{i}}^{l}$ , and $ca d_{v_{i}}^{h}$ , denote channel access delay of low priority packet and high priority packet, respectively. $ca d_{v_{i}}^{l}$ (or $ca d_{v_{i}}^{h}$ ) is updated when the last transmission is LP (or HP) packet, which is estimated by the EWMA method as

\begin{matrix} ca d_{v_{i}}^{l} = β_{d} \cdot ca d_{v_{i}}^{l} + (1 - β_{d}) \cdot (t_{backoff}^{l} + t_{CCA}^{l}), \end{matrix}

(9)

\begin{matrix} ca d_{v_{i}}^{h} = β_{d} \cdot ca d_{v_{i}}^{h} + (1 - β_{d}) \cdot (t_{backoff}^{h} + t_{CCA}^{h}), \end{matrix}

(10)

where $t_{backoff}^{l}$ ( $t_{backoff}^{h}$ ) is the total backoff delay of sending LP (HP) packet. $t_{CCA}^{l}$ ( $t_{CCA}^{h}$ ) is the totally clear channel assessment delay of LP (HP) packet when staying in queue. $β_{d}$ represents the degree of weighting decrease, $0 < β_{d} < 1$ .

Assuming that the size of LP and HP packets is equal. The transmission delay, ${T_{t}}_{x}$ , and LP and HP packets are also considered equal. Thus the total latency of node $v_{i}$ transferring an LP (HP) packet is denoted by ${Tr}_{v_{i}}^{l}$ ( ${Tr}_{v_{i}}^{h}$ ), which can be expressed as ${Tr}_{v_{i}}^{l} = ca d_{v_{i}}^{l} + T_{t x}$ ( ${Tr}_{v_{i}}^{h} = ca d_{v_{i}}^{h} + T_{t x}$ ).

4.5. QoS-MAC Simulation

To study the performance of our QoS-MAC, we conducted experiments using QualNet© network simulation software for two test scenarios designed for a Smart Grid WSN.

In the experiments, we considered a worst situation that many critical status data should be reported in short term when a failure occurred in Smart Grid. The sudden failure will result in bursty of critical traffic (High priority).

Two test scenarios are designed to study the network behavior of both types of traffic. Since the HP packet is bursty critical traffic and LP packet is continuous periodic regular traffic, we analyzed the network throughput, goodput (which is defined as the ratio of achieved throughput and the traffic arrival rate), delay, and collision rate when the HP traffic arriving rate ( $λ^{h}$ ) suddenly increases from (a) Constant Bit Rate of 8 kbps to (b) Constant Bit Rate of 16 kbps, while changing the LP traffic arriving rate ( $λ^{l}$ ) from 0.5 kpbs to 23 kbps on each node. As the result, the maximum traffic through the network for test scenario (a) ranges from 8 kps to 310 kbps, and 16 kbps to 390 kbps for test scenario (b). The PHY protocol of simulation is on the 26th channel of IEEE802.15.4 with the bandwidth of 250 kbps. Therefore, two test scenarios are designed to study the network behavior of both types of traffic for (a) when there is sufficient bandwidth and (b) when the bandwidth is more than saturated—demonstrating the performance for high priority data using our QoS-MAC protocol.

The network topology of our QoS-MAC simulation takes a star shape as shown in Figure 6. Since the process of medium access of node only interacts with its neighbor nodes, a network with 10 source nodes and a sink node is adopted, and all nodes are neighbor with each other. Each source node generates HP and LP packets and then sends them to sink node. Others parameters are set as in Table 3.

Table 3

Parameters setting in simulation of QoS-MAC.

	Queue size (bytes)	Packet size (bytes)	Backoff period (ms)	Number of backoffs	CCA duration (ms)
HP	500	50	0.64	15	0.016
LP	500	50	1.28–9.92	5	0.7

Figure 6

Network topology of QoS-MAC algorithm simulation.

It is worth noting that the default settings of IEEE802.15.4 standard are as follows: the backoff period is in the range 0 ms to 9.92 ms (which equals $2^{B E} - 1$ aUnitBackoffPeriod); the maximum number of backoff is 5; and CCA duration time is 0.128 ms. On the basis of the presentation of QoS-MAC algorithm above, the modification of parameter-setting guided by those three rules is to realize the following QoS functions: (1) reducing the backoff period for HP and increasing the minimum backoff time for LP, which is used to reduce HP packet's latency time and to prevent HP packet from being interrupted by LP packet, (2) increasing the number of backoff time of HP packet to provide more opportunities for accessing medium channel, which increases the reliability of HP packet, and (3) reducing CCA duration of node sending HP packet to enable it to interrupt other nodes sending LP packet.

In both scenarios, all the nodes try to send generated traffic simultaneously; thus the worst case scenario is simulated for the WSN. When the data traffic in the network exceeds the maximum bandwidth supported by IEEE802.15.4 (250 kpbs), the quality of service affects all network performance measures as shown in Figures 7 and 8.

Figure 7

Performance of QoS-MAC $λ^{h} = 8$ kbps (a) throughput (b) goodput (c) delay (d) collision rate.

Figure 8

Performance of QoS-MAC $λ^{h} = 16$ kbps (a) throughput (b) goodput (c) delay (d) collision rate.

From Figure 7, we observe that both the throughput and goodput of HP traffic are degraded a little even when the network bandwidth is saturated. Its throughput reaches to above 60 kbps (with total traffic generated as 80 kbps), while its goodput hovers above 80%. The LP traffic, on the other hand, shows quite a different trend. As $λ^{l}$ increases, its throughput rises until reaching platoon of 100 kbps, far less than the 310 kbps traffic arriving rate. As a result, its goodput drops to around 40%. The network delay of the HP traffic is consistently and significantly lower than that of the LP traffic, with 50% of improvement. The collision rate (as shown in Figure 7(d)) of the HP traffic is virtually 0, while the rate for LP traffic continues to increase by 40%. It is also observed that with more traffic in the network, the performance disparity between HP and LP data is increasing.

Figure 8 shows the experiment results for test scenario (b) when the traffic arriving rate of HP traffic $λ^{h}$ is 16 kbps while changing the traffic arriving rate of LP traffic $λ^{l}$ from 0.5 kbps to 23 kbps with 1 kpbs increment. These results show the same trend as those in the test scenario (a), in which the performance disparity between HP and LP data traffic is even bigger.

In summary, with the performance analysis of both test scenarios for a WSN, the QoS-MAC improves the quality of service of HP packets with boundaries for network delay, goodput, and collision rate.

5. Quality of Service in NWK Layer

The function of the QoS Optimization of NWK is shown in Figure 1, in which static routing table is created from known geographic information of WSN topology. Transmission reliability and latency are estimated by reliability estimator and latency estimator, respectively, based on the information from PHY layer and MAC layer. Finally, the LQ-routing algorithm chooses the node which satisfies the reliability and latency requirements as the next hop node.

5.1. Static Routing Table

In Smart Grid WSN, the locations of sensor nodes which are attached to electrical equipment are stationary. The geographic coordinate values of them are known or measurable. Assuming that all the coordinate data of nodes are recorded in network location table in every node. The network location table is built in advance and is updated only when the geographic locations of any nodes are changed by flooding-based broadcast.

In this study, Neighbor nodes Set is defined as a set of nodes located in the communication range of current node. As presented in Figure 9, ${v_{j}, v_{k}, v_{l}, v_{m}, v_{n}, v_{p}, v_{q}}$ are the neighbor nodes set of $v_{i}$ . Static routing table is defined as a set of alternative neighbor nodes geographically closer to the destination for transmission packet. As shown in Figure 9, when Node $v_{i}$ is sending a packet to node $v_{c}$ , $v_{i}$ 's Static routing table, $Sr t_{v_{i}, v_{c}}$ , is defined as a subset of neighbor nodes set that located in the $\pm θ$ ( $θ < π / 2$ ) angle sector area along with the line connecting $v_{i}$ and $v_{c}$ .

Figure 9

Schematic diagram of a WSN's neighbor nodes Set.

5.2. Reliability Estimator

The reliability estimator in NWK Layer computes the reliability of transmission between the current node and its neighbor nodes. Ideally, the reliability of a transmission depends on the reliability of wireless link as in [37], but in the MAC layer, the packet may be dropped in the process of CSMA/CA of IEEE802.15.4 standard. Therefore, in our reliability estimator, both the collision rate in MAC layer and Link Quality in PHY layer are collected to estimate the reliability.

The collision rate $c r_{v_{i}}$ introduced in Section 4 indicates the probability that the packet cannot be sent out successfully in MAC layer. The Link Quality metric $P_{rr} (v_{i}, v_{j})$ introduced in Section 3 denotes the packet reception ratio between node $v_{i}$ and its neighbor node $v_{j}$ . Therefore, a successful transmission means that the packet is not only sent out from $v_{i}$ but is also received by $v_{j}$ . Furthermore, the reliability estimator result between $v_{i}$ and $v_{j}$ for HP packets can be computed by

\begin{matrix} η_{v_{i}, v_{j}}^{h} = P_{rr} (v_{i}, v_{j}) \cdot c r_{v_{i}}^{h} . \end{matrix}

(11)

5.3. Latency Estimator

On the routing path from source to destination of multihop transmission, the total communication delay is the sum of latency of each hop. The latency estimator aims to estimate the latency between its neighbor nodes.

Assuming that $v_{j}$ exchanges MAC layer transfer delay ( $T r_{v_{j}}^{l}$ , $T r_{v_{j}}^{h}$ ) and the number of packets in queue ( $L_{v_{j}}^{l}$ and $L_{v_{j}}^{h}$ ) with its neighbor node $v_{i}$ by HELLO message as in [37], thus, for HP packet, the latency of choosing $v_{j}$ as the next hop node, denoted as $d_{v_{i}, v_{j}}^{h}$ , waits for all the packets currently in $v_{j}$ 's HP queue to be transferred. $d_{v_{i}, v_{j}}^{h}$ can be expressed as

\begin{matrix} d_{v_{i}, v_{j}}^{h} = {Tr}_{v_{j}}^{h} \cdot L_{v_{j}}^{h} . \end{matrix}

(12)

5.4. LQ Routing Algorithm

LQ routing algorithm is to select the best next hop node according to the reliability and latency requirements of packet, which is dependent on the QoS requirements of Smart Grid application. For example, according to Table 1, if a packet carries data of Smart metering, its reliability requirement will be 99%-100% and its latency requirements will be 2 seconds. It is important to note that those requirements are for end-to-end performance. Therefore, in multihop WSN, two additional QoS parameters latency and reliability requirements in each hop are defined for the LQ routing. Since latency is cumulative and reliability is multiplicative over all single hops on a routing path, one hop requirements are computed by

\begin{matrix} T_{hop} = \frac{T_{req}}{n_{maxhop}}, \\ η_{hop} = \sqrt[n_{maxhop}]{R_{req}}, \end{matrix}

(13)

where $T_{req}$ and $R_{req}$ represent the latency and reliability requirements of Smart Grid application. $T_{hop}$ and $η_{hop}$ are latency and reliability requirements of each hop. And $n_{maxhop}$ is the maximum hops number of location-based routing in a WSN.

From (13), it can be found out that $T_{hop}$ and $η_{hop}$ are related to Smart Grid application and network size. When network size is too big (where $n_{maxhop}$ value is also too big), both the requirements of end-to-end and one hop cannot be satisfied; therefore, the value of $n_{maxhop}$ in the topology design and deployment has to be limited. For example, WSN can be divided into many subnetworks and a sink node is deployed in each subnetwork to limit network size as presented in [10]. It is worth noting that the topology design is beyond the scope of this paper. For a deployed WSN and a certain Smart Grid application, $T_{hop}$ and $η_{hop}$ are considered as constants. Consequently, the LQ routing algorithm is aimed at satisfying $T_{hop}$ and $η_{hop}$ on each hop. It is presented as shown in Algorithm 1.

Algorithm 1: LQ-routing algorithm.

(1) $v_{i}$ is ready for sending a packet to destination $v_{c}$

(2) if the packet has been sent

(3) break

(4) if $v_{c}$ is neighbor node of $v_{i}$

(5) $v_{c}$ is selected as routing node

(6) else

(7) Initialization forwarding candidate set $N_{v_{i}}$ = ⌀

(8) if packet has latency requirement of each hop $T_{hop}$

(9) for $v_{j} \in {Srt}_{v_{i}, v_{c}}$

(10) if $d_{v_{i}, v_{j}}^{h} < T$

(11) $N_{v_{i}} = N_{v_{i}} \cup {v_{j}}$

(12) if $N_{v_{i}} = = ⌀$

(13) $N_{v_{i}} = {v_{j}}$ , ( $d_{v_{i}, v_{j}}^{h} = \min$ ( $d_{v_{i}, v_{k}}^{h}$ ), $v_{k} \in {Srt}_{v_{i}, v_{c}}$ )

(14) else

(15) $N_{v_{i}} = {Srt}_{v_{i}, v_{c}}$

(16) If packet has reliability requirement of each hop $η_{hop}$

(17) if $\max$ ( $η_{v_{i}, v_{k}}$ ) $\geq$ $η_{hop}$ , $v_{k} \in N_{v_{i}}$

(18) $v_{k}$ is selected as routing node

(19) else

(20) for (m = 1, $m \leq$ sizeof( $S_{v_{i}}$ ), m++)

(21) $1 - \overset{m}{\overset{︷}{(1 - η_{v_{i}, v_{j}}) (1 - η_{v_{i}, v_{k}}) \dots (1 - η_{v_{i}, v_{l}})}} \geq η_{hop}$

(22) m nodes $v_{j}$ , $v_{k}, \dots, v_{l}$ are selected as routing nodes

(23) else

(24) $v_{i}$ randomly selects a node from $N_{v_{i}}$ as routing node

(25) if the processing packet has reliability or lantency requirement

(26) Buffer the packet into HP queue

(27) else

(28) Buffer the packet into LP queue

When node $v_{i}$ is ready for transmitting a packet to sink node $v_{c}$ , firstly it checks whether this packet has been sent before. If it has been sent, this packet will be dropped to prevent routing congestion. Then $v_{i}$ checks whether node $v_{c}$ is $v_{i}$ 's neighbor node, if it is, $v_{c}$ is selected as next-hop routing node. If on the contrary, $v_{i}$ starts routing node selection within its Static routing table.

In the routing node selection process, firstly $v_{i}$ initializes forwarding candidate set $N_{v_{i}} = ⌀$ . If the packet has latency requirement $T_{hop}$ of each hop, all $v_{i}$ 's neighbor nodes satisfying latency requirement $T_{hop}$ are placed into set $N_{v_{i}}$ . Otherwise, the neighbor node of $v_{i}$ with least latency is placed into set $N_{v_{i}}$ . If the packet does not have latency requirement $T_{hop}$ , all nodes in Static routing table $Sr t_{v_{i}, v_{c}}$ are placed into set $N_{v_{i}}$ .

Secondly, if packet has reliability requirement $η_{hop}$ on each hop, node $v_{i}$ checks whether there is any neighbor node in set $N_{v_{i}}$ that can satisfy $η_{hop}$ . If so, this node is selected as next-hop routing node. Otherwise, m neighbor nodes, whose total multipath reliability satisfies $η_{hop}$ , are selected as next-hop routing nodes. In other words, when single neighbor node (single-path routing) cannot satisfy reliability of Smart Grid application, multineighbor nodes (multipath routing) are adopted to improve reliability performance of routing algorithm. If packet does not have reliability requirement $η_{hop}$ on each hop, $v_{i}$ randomly selects a node from set $N_{v_{i}}$ as next-hop routing node.

Finally, the packet is buffered in corresponding queue according to its priority.

5.5. LQ Routing Simulation

We conducted experiment to study the performance of LQ routing algorithm in comparison with other states of the art location-based routing protocols: GFW [50], SPEED [38], MMSPEED [39], and DARA [51], which are all implemented in QualNet© with the same simulation scenario setting. GFW makes greedy forwarding decisions using location information of nodes which is not a QoS algorithm. It is chosen as a benchmark representing basic location-based routing. SPEED and MMSPEED are well known location-based QoS routing. SPEED is designed to address latency requirement by maintaining a desired delivery speed, and MMSPEED is designed to improve both delay and reliability performances with a high probability. They are chosen as contrasts to provide service differentiation and probabilistic QoS guarantees in the latency and reliability domains. DARA is also a location-based QoS routing, which considers both reliability and delay, and defines two kinds of packets: critical (HP) and noncritical (LP) packets. We chose DARA as a contrast because its optimization objective is similar to LQ routing.

The topology of simulation is presented in Figure 10. A total of 49 sensor nodes are uniformly deployed in a 2 × 2 km² square area in QualNet© software. A sink (coordinator) node is located in the center. All the other nodes generate packets with the same rate and sent to sink node. Communication range of each node is about 500 m. The internal distance between nodes is 300 m. The QoS-MAC introduced in Section 4 is adopted in the LQ routing simulation, the PHY layer is 26th channel at 2.4 GHz defined by IEEE802.15.4 and the bandwidth is 250 kbps. The coefficient $β_{p}$ , $β_{c}$ and $β_{d}$ are set as 0.6, 0.7 and 0.7 respectively. Both the Windows size $w_{c}$ and $w_{d}^{h}$ of WMEWMA are set as 10 packets. θ of Static routing table definition is set as $6 0^{°}$ . The end-to-end latency and reliability requirements of HP packets are 500 ms and 0.95 respectively. LP packets have no requirement. The maximum hops number of location-based routing in WSN in Figure 10 is 5 hops. Therefore, for each HP packet, its $T_{hop}$ and $η_{hop}$ are set as 100 ms and 0.989 respectively. LP packet has no reliability and delay requirement. The traffic generation rate of each node is set as Constant Bit Rate of 1 kbps. The proportion of HP packet of the traffic of each node is increased from 10% to 100%. HELLO message is sent every 10 s between neighbor nodes.

Figure 10

Network topology of LQ routing algorithm simulation.

Figure 11(a) presents simulation results of average end-to-end PRR of both LP and HP packets (referred to as average PRR). From Figure 11(a), it could be found out that LQ, DARA and MMSPEED increase their average PRR as a function of HP proportion, while SPEED and GFR perform stably which are maintained at around 0.55 and 0.61 respectively. That indicates the former three algorithms support reliability QoS, while the latter two algorithms do not, which is in line with the previous expectations. Although average PRR of MMSPEED and DARA increases, LQ provides 25–30% better improvement of average PRR performance than MMSPEED and 10–20% more improvement than DARA. Therefore, LQ clearly performs best compared with other algorithms.

Figure 11

Simulation result of end-to-end PRR.

Figure 11(b) shows the simulation results of average end-to-end PRR of HP packets (referred to as HP PRR). In the figure we can observe that HP PRR of LQ is kept nearly at 1 when HP packets increase. On the other hand, HP PRR of MMSPEED dropped from 0.902 to 0.811, which means the reliability of QoS performance of MMSPEED is limited. HP PRR results of SPEED and GFR are not changed with HP packets increasing, because they do not support reliability QoS without differentiating HP and LP packets. The HP's reliability performance of LQ clearly outperforms other four algorithms on improving end-to-end reliability.

Figure 12(a) presents average end-to-end delay (referred to as average delay) of all packets. In the figure, the average delay of LQ routing algorithm is decreased from 320.2 ms to 204.4 ms when ratio of HP packets increases from 0.1 to 1, this proves our LQ routing algorithm provide latency QoS performance for HP packet. The average delay of GFW is almost not changed when ratio of HP packet increases because it not supports latency QoS. SPEED's and MMSPEED's average delay decrease a little with HP increases, which shows that SPEED and MMSPEED can provide QoS for communication delay. DARA provides less communication delay than GFW, SPEEED and MMSPEED. However, its average delay is about 30 ms more than LQ. In general, LQ spent least time for end-to-end communication.

Figure 12

Simulation result of end-to-end delay.

Figure 12(b) presents simulation results of HP packet's average end-to-end delay (referred to as HP delay). It is clearly that end-to-end delay for all algorithms is stable, which indicates the process of selecting routing path for HPs are not affected by the HP packets rise. As depicted in Figure 12(b), HP delay of LQ is between 175.4 ms to 207.3 ms, it performs 7–20% better than DARA (198.0 ms to 243.8 ms) and much better than other two QoS algorithms, SPEED (289.3 ms to 365.5 ms) and MMSPEED (263.9 ms to 309 ms).

In summary, with the compare analysis of GFW, SPEED, MMSPEED and DARA, our LQ-routing enhances either the end-to-end PRR or end-to-end delay for WSN.

6. Experiment Results and Discussions

In this section, the proposed Cross-layer QoS Optimization of WSN was implemented and tested in real Power Grid environment in Huanghshan city of China. It is compared with other two communication methods, GPRS and ZigBee, whose backgrounds are introduced as follows.

(1) GPRS. All distribution transformers in Huangshang city had been equipped with transformer supervisory terminal unit (TTU). All the TTU data are transmitted to DMS by GPRS (general packet radio service in wireless public cellphone network) model. Since GPRS is for public network, its communications are usually affected by other cellphone devices.

(2) ZigBee. A demonstration project “Wireless Sensor Networking Research and Practice for Power Industry in Small-and-Medium Sized Cities”, which has been designed and implemented in the Distribution Grid of Huangshan City, is aimed to gather operation data of distribution transformer and send into Distribution Manager System (DMS). In the project, ZigBee network was adopted to replace GPRS to construct a dedicated wireless network for TTU communication. The demonstration project includes a total of 64 ZigBee router/sensor nodes and 7 coordinator/Sink nodes. All the 71 nodes are divided into 7 subnetworks, and there is a Sink node in each subnetwork. However, ZigBee network also has disadvantages as it has no QoS mechanism.

A CC2530 based WSN testbed (node) is used for implementing our Cross-layer QoS Optimization algorithms, which aims to test the performance cross-layer algorithms proposed in this paper. Since Huangshan power company only allows us to implement the experiment in a subnetwork of the demonstration project for avoiding unpredictable effects on operation of the DMS, only a subnetwork with 14 nodes is used for experiment. The topology is shown in Figure 13.

Figure 13

Location information of experiment in Huangshan city; yellow nodes are sensors, red node is sink. The size of map is 410 m × 470 m.

In the experiment CC2530-based WSN testbed/node (shown in Figure 14) with power amplifier (CC2591) and high-gain antenna (10 dBi) is adopted to realize Cross-layer QoS protocols. The radio transmit power of the WSN node is 20 dBm, and the Line-of-Sight communication range is nearly 1.5 km. WSN testbed was operated in the ISM band at 2.4 GHz. A simplified WSN protocol based on an open source ZigBee stack, namely Freakz, was adopted in the testbed node. The protocols were operated on Contiki, a sensor and embedded operation system written in C programming language. The functions were modified, such as CCA detection, the CSMA/CA backoff mechanism, and routing algorithm of Freakz as presented in the cross-layer optimization algorithms.

Figure 14

WSN testbed adopted in experiment.

14 Nodes were installed on the distribution transformer as in Figure 15. The height of antennas is range between 2 m and 5 m according the height of electrical utilities. In the experiment, each node generated a HP packet and a LP packet per second. The packet size is 40 Byte. Since the end-to-end performances are unavailable without absolute synchronization network, we test the round trip delay and reliability performances. All the nodes transmit packet to sink node, and then sink node replies an ACK message to let each node knows the round trip delay and round trip reliability. All other parameters of our MAC and routing algorithms are the same with Table 2 and LQ-routing simulation setting in Section 5.

Figure 15

Node and antenna installation of different distribution transformers in the experiment.

In order to evaluate the improvement of our algorithms, the performances of Cross-layer QoS algorithms are compared with ZigBee protocol. Besides, since GPRS is widely used in TTU communication in china, our results are compared with it as well.

The original communication of TTU adopted GPRS and realized by SIM900 modules that can be controlled by AT cellular command interface. SIM900 connects with public cellphone network and internet after it is initialed. By the AT command “AT+CIPSTART”, the DMS server's IP address and port are assigned. Then, all the TTU data sent to SIM900 will be put into a TCP packet and be directly sent to DMS server. DMS server also can send inquiry and control commands in TCP packet to SIM 900.

The ZigBee network of demonstration project is composed by XBee Pro modules, in which, a standard IEEE802.15.4 MAC and PHY protocols and ZigBee specification were applied. All XBee Pro modules with the same pre-set PAN ID select an idle channel and then construct a network by self-organization. Therefore, we need to set PAN ID in advance. The transit power of XBee Pro is set as 17 dBm. Outdoor RF line-of-sight communication range of XBee Pro is about 1.6 km. The parameters of XBee Pro's MAC layer use default values of IEE802.15.4, for example, the backoff period is in the range 0 ms to 9.92 ms, the maximum number of backoff is 5, and CCA duration time is 0.128 ms. Routing protocol of XBee Pro is Ad hoc On-Demand Distance Vector (AODV) [52].

The experiment results in 24 hours of CC2530-based WSN testbed (referred to as QoS-MSN) are analyzed by comparing with the delay and reliability results of ZigBee and GPRS, which are extracted from DMS log files of those 14 nodes. The results are shown in Figure 16.

Figure 16

Experiment result of round trip delay and round trip reliability.

As in Figure 16, Round trip delays of HP packets of our QoS-WSN were around 0.5 s, which is less than either 1.1 s of ZigBee or 3.2 s of GPRS. But the Round trip delays of LP packet are around 1.4 s, which is more than ZigBee. Similar with round trip delay, the round trip reliability of HP packet is about 99%, about 1% higher than GPRS and 3% better than ZigBee. The round trip reliability of LP packet is almost the same with ZigBee. From the experiment, the following results can be obtained.

(1) Our QoS-WSN can obtain better performances than either ZigBee or GPRS.

(2) Compared with ZigBee, which adopts standard IEEE802.15.4 standard, Our QoS-WSN can effectively allocate limited wireless channel resources to HP traffic, that is the reason why the performance of HP packet is better but LP packet is worse than ZigBee.

(3) Compared with ZigBee, our QoS-WSN realized the end-to-end multiobject (reliability and/or latency) QoS optimization as expected.

7. Conclusion

In this paper, a comprehensive study on cross-layer QoS optimization of WSN for Smart Grid communication has been presented. On account of the diversity of application data in Smart Grid, the traffic has been classified into two categories according to their QoS requirements. All the algorithms are implemented in three layers of WSN protocol, which contains a LQE introduced to evaluate link quality under harsh environment in PHY layer. A QoS-MAC algorithm was presented to avoid disorderly contention among WSN nodes and to effectively allocate the limited wireless channel resources to two classes of packets, and an LQ routing algorithm provided an end-to-end multiobject (reliability and/or latency) QoS optimization route for the multihop transmission. Although all the three algorithms were implemented in different layers, they were operated in a cross-layer manner, which were different from traditional layered protocols. By comparing with the state-of-the-art protocols which provide end-to-end QoS for WSN, our algorithm obviously improved the performance of reliability and latency, since Link Quality estimation and wireless channel contention were taken into account.

Finally, the experiments were conducted in the demonstration project designed and implemented in the Distribution Grid of Huangshan City in China. With encouraging outcomes, the experiment showed that the proposed algorithms provided QoS with differentiated traffic based on their requirements. In addition, the proposed cross-layer optimization algorithms posted better critical traffic performances in comparison with those of the ZigBee protocols and the Distribution Grid traditional GPRS technologies.

Footnotes

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgment

This work was supported by National Natural Science Foundation of China under Grants 51307041 and 51177034.

References

NIST Framework and Roadmap for Smart Grid Interoperability Standards Release 2.0 2012 http://www.nist.gov/smartgrid/upload/NIST_Framework_Release_2-0_corr.pdf

IEEE Standards

IEEE Standard for Local and metropolitan area networks—part 15.4: Low-Rate Wireless Personal Area Networks (LR-WPANs) amendment 3: Physical Layer (PHY) specifications for low-data-rate, wireless, smart metering utility networks

IEEE Std 2012 802.15.4g-2012

Gungor

V. C.

Hancke

G. P.

Opportunities and challenges of wireless sensor networks in smart grid

IEEE Transactions on Industrial Electronics 2010 57 10 3557 3564

10.1109/TIE.2009.2039455

2-s2.0-77956607253

Ahmed

M. H. U.

Alam

M. G. R.

Kamal

Hong

C. S.

Lee

Smart grid cooperative communication with smart relay

Journal of Communications and Networks 2012 14 6 640 652

2-s2.0-84872849121

10.1109/JCN.2012.00030

Fateh

Govindarasu

Ajjarapu

Wireless network design for transmission line monitoring in smart grid

IEEE Transactions on Smart Grid 2013 4 2 1076 1086

10.1109/TSG.2013.2241796

2-s2.0-84878267420

Erol-Kantarci

Mouftah

H. T.

Wireless sensor networks for cost-efficient residential energy management in the smart grid

IEEE Transactions on Smart Grid 2011 2 2 314 325

2-s2.0-79957769025

10.1109/TSG.2011.2114678

Fernandes

R. F.

Fonseca

C. C.

Brandao

Ferrari

Flammini

Vezzoli

Flexible wireless sensor network for smart lighting applications

Proceedings of the IEEE International Instrumentation and Measurement Technology Conference (I2MTC ’14)

May 2014

Montevideo, Uruguay

434 439

10.1109/I2MTC.2014.6860782

Munhaw

Suryadevara

N. K.

Mukhopadhyay

S. C.

Gill

S. P. S.

WSN based utility System for effective monitoring and control of household power consumption

Proceedings of Instrumentation and Measurement Technology Conference (I2MTC '14)

2014

1382 1387

Guo

Delay minimization for reliable data collection on overhead transmission lines in smart grid

Proceedings of the IEEE Computing, Communications and IT Applications Conference (ComComAp '13)

April 2013

Hong Kong

IEEE

147 152

2-s2.0-84881228649

10.1109/ComComAp.2013.6533626

10.

Sun

Wang

Qian

Zhang

QoS modeling of wireless sensor networks for smart distribution power grid communication

Automation of Electric Power Systems 2011 35 20 59 64

2-s2.0-80555155682

11.

Baccour

Koubaa

Mottola

Zuniga

M. A.

Youssef

Boano

C. A.

Alves

Radio link quality estimation in wireless sensor networks: a survey

ACM Transactions on Sensor Networks 2012 8 4, article 34

10.1145/2240116.2240123

2-s2.0-84865016804

12.

Benkic

Malajner

Planinsic

Cucej

Ž.

Using RSSI value for distance estimation in Wireless sensor networks based on ZigBee

Proceedings of the 15th International Conference on Systems, Signals and Image Processing

June 2008

303 306

10.1109/IWSSIP.2008.4604427

2-s2.0-52949111029

13.

Srinivasan

Dutta

Tavakoli

Levis

An empirical study of low-power wireless

ACM Transactions on Sensor Networks 2010 6 2, article 16

2-s2.0-77749264928

10.1145/1689239.1689246

14.

Tang

Wang

K.-C.

Huang

Channel characterization and link quality assessment of IEEE 802.15.4-compliant radio for factory environments

IEEE Transactions on Industrial Informatics 2007 3 2 99 110

10.1109/TII.2007.898414

2-s2.0-34249862901

15.

Cerpa

Wong

J. L.

Potkonjak

Estrin

Temporal properties of low power wireless links: Modeling and implications on multi-hop routing

Proceedings of the 6th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC '05)

May 2005

414 425

2-s2.0-29844449933

10.1145/1062689.1062741

16.

de Couto

D. S. J.

Aguayo

Bicket

Morris

A high-throughput path metric for multi-hop wireless routing

Wireless Networks 2005 11 4 419 434

2-s2.0-23044444929

10.1007/s11276-005-1766-z

17.

Woo

Culler

D. E.

Evaluation of Efficient Link Reliability Estimators for Low-Power Wireless Networks 2003

Computer Science Division, University of California

18.

Senel

Chintalapudi

Lal

Keshavarzian

Coyle

E. J.

A Kalman Filter based link quality estimation scheme for wireless sensor networks

Proceedings of the 50th Annual IEEE Global Telecommunications Conference (GLOBECOM '07)

November 2007

Washington, DC, USA

875 880

10.1109/GLOCOM.2007.169

2-s2.0-39349117692

19.

Baccour

Koubâa

Youssef

F-LQE: a fuzzy link quality estimator for wireless sensor networks

Wireless Sensor Networks 2010 5970

Berlin, Germany

Springer

240 255 Lecture Notes in Computer Science

10.1007/978-3-642-11917-0_16

20.

Gronkvist

Traffic controlled spatial reuse TDMA in multi-hop radio networks

Proceedings of the 9th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications

September 1998

1203 1207

2-s2.0-0032301560

21.

Arumugam

Umamaheswaran

Infuse: a TDMA based reprogramming service for sensor networks

Proceedings of the Second International Conference on Embedded Networked Sensor Systems (SenSys '04)

November 2004

281 282

2-s2.0-27644578526

22.

Richardson

T. D.

Nicopoulos

Park

Narayanan

Xie

Das

Degalahal

A hybrid SoC interconnect with dynamic TDMA-based transaction-less buses and on-chip networks

Proceedings of the 5th International Conference on Embedded Systems and Design

2006

23.

Macedo

Grilo

Nunes

Distributed latency-energy minimization and interference avoidance in TDMA wireless sensor networks

Computer Networks 2009 53 5 569 582

10.1016/j.comnet.2008.10.015

2-s2.0-61349155478

24.

Saxena

Roy

Shin

Dynamic duty cycle and adaptive contention window based QoS-MAC protocol for wireless multimedia sensor networks

Computer Networks 2008 52 13 2532 2542

10.1016/j.comnet.2008.05.009

ZBL1181.68079

2-s2.0-49049113023

25.

Yang

Ulukus

Delay minimization with a general pentagon rate region

Proceedings of the IEEE International Symposium on Information Theory (ISIT '10)

June 2010

1808 1812

10.1109/ISIT.2010.5513302

2-s2.0-77955676936

26.

Yeh

Delay-optimal rate allocation in multiaccess communications: a cross-layer view

Proceedings of the IEEE Workshop on Multimedia Signal Processing

December 2002

404 407

27.

Yeh

Minimum delay multi-access communication for general packet length distributions

Proceedings of the Allerton Conference on Communication, Control, and Computing

2004

1536 1545

28.

Yang

Ulukus

Delay minimization in multiple access channels

Proceedings of the IEEE International Symposium on Information Theory (ISIT '09)

July 2009

2366 2370

10.1109/ISIT.2009.5205978

2-s2.0-70449511643

29.

van Dam

Langendoen

An adaptive energy-efficient MAC protocol for wireless sensor networks

Proceedings of the 1st International Conference on Embedded Networked Sensor Systems

November 2003

171 180

2-s2.0-18844402855

30.

Polastre

Hill

Culler

Versatile low power media access for wireless sensor networks

Proceedings of the 2nd International Conference on Embedded Networked Sensor Systems (SenSys '04)

November 2004

95 107

2-s2.0-26444451023

31.

Feng

Dou

Yuping

Analysis and compare of slotted and unslotted CSMA in IEEE 802.15.4

Proceedings of the 5th International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM '09)

September 2009

Beijing, China

1 5

2-s2.0-73149121921

10.1109/WICOM.2009.5303580

32.

Kulik

Heinzelman

Balakrishnan

Negotiation-based protocols for disseminating information in wireless sensor networks

Wireless Networks 2002 8 2-3 169 185

10.1023/A:1013715909417

ZBL1012.68975

2-s2.0-0036499227

33.

Ahvar

Serral-Gracià

Marín-Tordera

Masip-Bruin

Yannuzzi

EQR: a new energy-aware query-based routing protocol for wireless sensor networks

Wired/Wireless Internet Communication 2012

Berlin, Germany

Springer

102 113

10.1007/978-3-642-30630-3_9

34.

Lee

Yoo

Kim

Energy aware routing in location based ad-hoc networks

Proceedings of the 4th International Symposium on Communications, Control, and Signal Processing (ISCCSP '10)

March 2010

2-s2.0-77953854542

10.1109/ISCCSP.2010.5463499

35.

Dorigo

Gambardella

L. M.

Birattari

Martinoli

Poli

Stützle

An energy-efficient ant-based routing algorithm for wireless sensor networks

Ant Colony Optimization and Swarm Intelligence 2006 4150

Berlin Heidelberg

Springer

49 59 Lecture Notes in Computer Science

10.1007/11839088_5

36.

Sumathi

Srinivas

M. G.

A survey of QoS based routing protocols for wireless sensor networks

Journal of Information Processing Systems 2012 8 4 589 602

10.3745/JIPS.2012.8.4.589

2-s2.0-84874176241

37.

Djenouri

Balasingham

Traffic-differentiation-based modular QoS localized routing for wireless sensor networks

IEEE Transactions on Mobile Computing 2011 10 6 797 809

2-s2.0-79955480266

10.1109/TMC.2010.212

38.

Stankovic

J. A.

Abdelzaher

T. F.

A spatiotemporal communication protocol for wireless sensor networks

IEEE Transactions on Parallel and Distributed Systems 2005 16 10 995 1006

10.1109/TPDS.2005.116

2-s2.0-27444435578

39.

Felemban

Lee

C.-G.

Ekici

MMSPEED: multipath multi-SPEED protocol for QoS guarantee of reliability and timeliness in wireless sensor networks

IEEE Transactions on Mobile Computing 2006 5 6 738 753

2-s2.0-33646456047

10.1109/TMC.2006.79

40.

Huang

Fang

Multiconstrained QoS multipath routing in wireless sensor networks

Wireless Networks 2008 14 4 465 478

2-s2.0-46449105138

10.1007/s11276-006-0731-9

41.

Djenouri

Balasingham

LOCALMOR: localized multi-objective routing for wireless sensor network

Proceedings of the IEEE 20th Personal, Indoor and Mobile Radio Communications Symposium (PIMRC '09)

September 2009

IEEE

10.1109/PIMRC.2009.5449740

2-s2.0-77952839317

42.

Ericsson

G. N.

Cyber security and power system communicationessential parts of a smart grid infrastructure

IEEE Transactions on Power Delivery 2010 25 3 1501 1507

2-s2.0-77954005795

10.1109/TPWRD.2010.2046654

43.

Shah

G. A.

Gungor

V. C.

Akan

O. B.

A cross-layer QoS-aware communication framework in cognitive radio sensor networks for smart grid applications

IEEE Transactions on Industrial Informatics 2013 9 3 1477 1485

10.1109/TII.2013.2242083

2-s2.0-84882955779

44.

Sun

Yuan

Wang

Han

Zhang

Quality of service networking for smart grid distribution monitoring

Proceedings of the 1st IEEE International Conference on Smart Grid Communications (SmartGridComm '10)

2010

373 378

10.1109/SMARTGRID.2010.5622072

45.

Al-Anbagi

Erol-Kantarci

Mouftah

H. T.

Priority- and delay-aware medium access for wireless sensor networks in the smart grid

IEEE Systems Journal 2013 8 2 608 618

10.1109/JSYST.2013.2260939

2-s2.0-84885350103

46.

Temel

Ş.

Gungor

V. Ç.

Koçak

Routing protocol design guidelines for smart grid environments

Computer Networks 2014 60 160 170

10.1016/j.bjp.2013.11.009

2-s2.0-84894751910

47.

Sahin

Gungor

V. C.

Kocak

Tuna

Quality-of-service differentiation in single-path and multi-path routing for wireless sensor network-based smart grid applications

Ad Hoc Networks 2014 22 43 60

10.1016/j.adhoc.2014.05.005

2-s2.0-84901384008

48.

Zamalloa

M. Z.

Krishnamachari

An analysis of unreliability and asymmetry in low-power wireless links

ACM Transactions on Sensor Networks 2007 3 2

1240227

10.1145/1240226.1240227

2-s2.0-34250320573

49.

Wireless medium access control and physical layer specifications for low-rate wireless personal area networks

2006 IEEE Std. 802.15.4

50.

Karp

Kung

H. T.

GPSR: greedy perimeter stateless routing for wireless networks

Proceedings of the 6th Annual International Conference on Mobile Computing and Networking (MOBICOM '00)

August 2000

243 254

2-s2.0-0034547115

51.

Razzaque

M. A.

Alam

M. M.

Mamun-Or-Rashid

Hong

C. S.

Multi-constrained QoS geographic routing for heterogeneous traffic in sensor networks

IEICE Transactions on Communications 2010 91B 8 2589 2601

2-s2.0-67649892178

10.1093/ietcom/e91-b.8.2589

52.

Samir

R. D.

Belding-Royer

E. M.

Perkins

C. E.

Ad Hoc On-Demand Distance Vector (AODV) Routing 2003