Modeling and parameter analysis of IEEE 802.15.4-based networks and the metering application

Abstract

As the support of wireless sensor networks expands to various application scenarios, the communication environments and the performance requirements of different application scenarios vary a lot. To cope with different communication environments and performance requirements, both data transmission ability and medium access ability are equivalently important. In this article, a joint analytical model is proposed to fully and precisely estimate networks’ communication performance, energy efficiency, and scalability. In the proposed model, both the physical layer’s and medium access control layer’s key parameters are taken into consideration. By comparing with OPNET-based simulation model, the rationality of the proposed analytical model is first validated under a wide range of network scenarios. Then, a series of simulations under general network scenarios and metering network scenarios are conducted. With these simulations, the performance of adjusting both layers’ parameters in improving communication performance and energy efficiency was proved superior to single-layer’s parameter optimizations. Finally, by comparing the available range of different key parameters’ optimal value under different network scenarios, the maximum backoff numbers and the minimum backoff exponent are considered to be the most effective parameters for metering network optimization.

Keywords

IEEE 802.15.4 CSMA/CA transmission power control data packet size periodic traffic metering networks

Introduction

Wireless sensor networks (WSNs) have penetrated many kinds of applications, from habitat monitoring to industrial control, for its advantages of low deployment costs, easier to installation, maintenance and reconfiguration, as well as inherent intelligent-processing capability over traditional wired devices.¹ For example, as one of the emerging applications of WSNs, new smart meters have superseded the traditional counterpart–wired electricity, water, and gas meters.² While, as the support of WSNs expands to various applications, the network configurations and performance requirements diverse a lot. In general IEEE 802.15.4-based metering networks, the scale of network usually ranges from tens to thousands of meters.³ And the data acquisition interval ranges from milliseconds to days.¹ Salam and Khan⁴ list some critical issues that can degrade overall network efficiency for WSNs, which include energy efficiency, reliability, latency, climate, dynamic network, and security. Since sensor nodes are battery-powered while WSNs are generally designed to support applications in long-term deployments, the object of high energy efficiency usually contradicts with high reliable and real-time communication. These contradictions become more outstanding in metering networks. In metering network, maintaining high reliability and low latency communication is indispensable. However, the performance of communication in WSNs usually fluctuates over time and space. And the harsh environment that metering networks exist makes communication links even more unreliable. For example, the multipath effect, frequency selective fading, and data delay often occur in the underwater acoustic communication due to the low transmission rate (generally ≤100 kbps).⁵ Besides, the metering network may have to accommodate more and more devices, which demands the network with good scalability. To deal with these challenges, single-layer’s analysis is not sufficient anymore. Hence, this article takes the key parameters of the physical layer and medium access control (MAC) layer into consideration and constructs an overall analytical model to investigate the joint impact of these parameters on networks’ reliability, time efficiency, energy efficiency, and scalability.

The contributions of this article are as follows:

Without directly assuming the distribution of network traffic as previous works do, in this article, the network traffic on each channel is mathematically approximated as a Poisson’s process by characterizing the generation process of network traffic with a series of transformations among different stochastic processes.

An accurate and comprehensive analytical model for the analysis of IEEE 802.15.4 is proposed. The characteristic of sensor nodes is based on the TelosB mote Module.⁶ On the physical layer, we take both modulation way and handshake mechanisms into consideration. On the MAC layer, by leveraging the properties of Poisson’s process, a simple but accurate formula of calculating the probability of successful channel access is obtained.

The effect of adjusting different layer’s parameters on improving networks’ communication performance and energy efficiency is simulated and analyzed in both general networks and metering networks.

According to the optimal value of key parameters under different network scenarios, we find out that the maximum backoff numbers and the minimum backoff exponent are the most effective parameters for metering network optimization.

The remainder of the article is organized as follows: related work is given in section “Related works.” Section “System model” presents the WSN network traffic analysis, the analytical model, and critical issues. In section “Critical issues,” three critical issues concerning data transport performance and energy efficiency are proposed and analytically characterized. In section “Simulation and analysis,” rational analysis and performance evaluation of the proposed analytical model and critical issue optimizations are presented. Section “Conclusion” provides our concluding remarks and discussions of the results.

Related works

IEEE 802.15.4 has been thoroughly investigated and many models together with optimization approaches have been proposed in the literature.^7–27 Most of the proposed analytical models focus on either the physical layer or on the MAC layer. While in fact, the performance of communication is affected by both layers.

On the physical layer, in the quest for more accurate wireless channel characteristics, most analytical models adopt empirical data–based channel models.^8–19 While the assumption of transmission features and corresponding energy dissipation characteristics differs a lot. The models in the literature^8,9 assume the transmission power to be fixed. The models in the literature^10–12 assume the transmission power can be adjusted continuously within a predefined range. While in the literature,^13,14 they adopt discrete transmission power levels which are based on the characteristics of real sensor motes, such as Mica2 motes. Besides, within these studies, few of them consider the two-way handshaking mechanism completely. Although the model in the literature¹⁵ considers the energy dissipation of both data packet and ACK packet transmission, it ignores the failure probability of ACK transmission. Inspired by Kurt et al.,¹⁶ many analytical models of physical layer start to take both successful probability and energy dissipation of two-way handshaking mechanism into consideration.^17–19 Similarly, in this article, we model the physical layer according to the characteristics of the MICAz mote and take both modulation way and handshake mechanism into consideration.

Inspired by Bianchi’s²⁰ work, many models of the IEEE 802.15.4 MAC layer have been proposed in the literature.^7,21–27 The majority of them are the Markov chain based.^21–27 Park et al.²¹ propose a general Markov chain model for the analysis of the IEEE 802.15.4 MAC layer. They took the retransmission limit and handshaking mechanism into consideration. However, the model in Park et al.²¹ assumed a saturated traffic input which is not suitable for the majority appliances where the network does not always have packets to send. Jung et al.²² propose an analytical Markov chain model under unsaturated traffic conditions. In their proposed model, they considered superframe structure, retransmission limit, and handshaking operation, which is better reflected the characteristics of the IEEE 802.15.4 MAC layer. However, since the Markov chain model adopts states to characterize the status and behaviors of wireless sensor nodes, a large number of states will be defined for a precise model. In some applications like industrial metering, the network may envisage a relatively large node population. The adoption of Markov chain model would render the whole model more complicated. Besides, another detail of Jung et al.²² that needs to be reconsidered is the assumption of each node’s data frame arrival rate. For many metering networks, nodes usually generate data periodically. Hence, successive data packets from the same node are always strongly correlated with each other, which is inadequate to model as a Poisson’s process.

To avoid the high computational complexity that the Markov chain model brings about in dense WSNs, the model in Elshabrawy et al.⁷ assumes the traffic of a WSN with thousands of metering devices follows the Poisson’s distribution. And the arrival rate of clear channel assessment (CCA) attempts is analytically derived as a Poisson’s process. The report success probability is finally evaluated according to the aggregate CCA attempts rate. However, the relationship between aggregate CCA arrival rate and the report success probability in Elshabrawy et al.⁷ cannot be formulated, which will definitely take additional calculations in simulations. Inspired by Elshabrawy et al.⁷ and Kurt et al.,¹⁶ we propose a more comprehensive and scalable analytical model for the analysis of IEEE 802.15.4. Instead of directly assuming the distribution of network traffic as Elshabrawy et al.⁷ and Jung et al.²² do, we provide a series of rigorous deductions on it. The MAC layer model in this article furtherly uses the properties of Poisson’s process and obtains a more detailed analytical model but with lower computation complexity than the model in Elshabrawy et al.⁷ Besides, in addition to evaluating the throughput and energy consumption issue, this article explicitly analyzes and formulates three critical issues for WSNs, that is, reliability, time efficiency, and energy efficiency. Finally, a typical metering network is analyzed objectively and comprehensively under different network scales and traffic loads.

System model

Overview

We consider a WSN working in the beacon-enabled mode with star topology. One base station at the center and multiple sensor nodes are uniformly deployed around the base station within carrier sensing range. On the physical layer, we assume the transmission of a data packet is confirmed with handshaking mechanism (i.e. each successful transmission is replied with an ACK packet by the receiver). The characteristics of sensor nodes in this research take the TelosB mote Module as a reference, to simulate the actual situation as possible. On the MAC layer, we only consider the contention access period (CAP) in the active period. Beacon period and contention-free period (CFP) in the active period, as well as the inactive period, are beyond the consideration.

Network traffic analysis

Each sensor node is assumed to send a data packet to the base station in every T seconds, which consists of $n_{T}$ sequential time slots. For one sensor node, the probability that there exists a data packet needs to be reported is $1 / n_{T}$ . For one channel shared by $N_{C}$ devices, the probability that there exists a data packet needed to be transmitted from this channel at each single time slot is

p_{C} = 1 - {(1 - \frac{1}{n_{T}})}^{N_{C}}

(1)

For each sensor node, its successive data packets usually have a very strong correlation since each node usually reports to the base station periodically. However, for one channel shared by several sensor nodes, the correlation between its two successive data packets is much weaker. Since successive data packets that arriving at one channel always come from different sensor nodes, which may be completely irrelevant no matter in terms of external surroundings or application objects. Hence, it is more adequate to approximate the case that $k$ data packets are transmitted through one channel within $n_{T}$ time slots as a Binomial process. And the corresponding probability is

p_{C}^{k, n_{T}} = (\frac{k}{n_{T}}) {(p_{C})}^{k} {(1 - p_{C})}^{n_{T} - k}

(2)

According to equation (1), for a given report period T, as the duration of each time slot decreases, the value of $n_{T}$ increases while the value of $p_{C}$ decreases. Hence, we can get the limit of $p_{C}^{k, n_{T}}$

\begin{matrix} lim_{n_{T} \to \infty, p_{c} \to 0} p_{C}^{k, n_{T}} = lim_{n_{T} \to \infty, p_{c} \to 0} \\ \frac{{(n_{T} p_{C})}^{k}}{k!} {(1 - p_{C})}^{- \frac{(n_{T} p_{C})}{p_{C}}} \frac{1}{{(1 - p_{C})}^{k}} \end{matrix}

(3)

As $n_{T}$ increases and $p_{C}$ approximates to 0, $(1 - p_{C})^{- 1 / p_{C}}$ approximates to $e$ and $1 / (1 - p_{C})^{k}$ approximates to 0. The limit of $p_{C}^{k, n_{T}}$ can be furtherly expressed as

lim_{n_{T} \to \infty, p_{c} \to 0} p_{C}^{k, n_{T}} = lim_{n_{T} \to \infty, p_{c} \to 0} \frac{{(n_{T} p_{C})}^{k}}{k!} e^{- (n_{T} p_{C})}

(4)

Let the product of $n_{T}$ and $p_{C}$ as $λ$ , according to equation (4) we can get the limit of $p_{C}^{k, n_{T}}$ approximates to Poisson’s distribution

lim_{n_{T} \to \infty, p_{c} \to 0} p_{C}^{k, n_{T}} = lim_{n_{T} \to \infty, p_{c} \to 0} \frac{{(λ)}^{k}}{k!} e^{- (λ)}

(5)

In the slotted CSMA/CA algorithm, the duration of one unit backoff slot is 320 μs. While for a sensor node in the network, the time interval between its two successive data packets is always counted by seconds. The value of $n_{T}$ is at the level of hundreds to thousands accordingly. Hence, the data packet transmission on one channel is reasonable to be approximated as a Poisson’s process with the arrival rate $λ$ as $n_{T} p_{C}$ .

Device mode analysis

A wireless sensor node is always composed of several modules, including sensors, radio module, antenna, microcontroller unit (MCU), and so on. Radio module, antenna, and MCU are three main components in charge of data transmission. In the following analyses, we only consider the operation modes and energy consumption of the radio module and MCU, since these two modules are major modules in terms of both device function and energy consumption. TelosB mote takes the Chipcon CC2420²⁸ as its radio module and adopts the TI-MSP430 microprocessor²⁹ as its MCU module. The Chipcon CC2420 radio module has four operation modes: transmit, receive, idle, and sleep. Different operation modes correspond to different power consumptions. The TI-MSP430 microprocessor has six operation modes: one active mode and five low-power modes, that is, from LPM0 to LPM4.²⁹

According to the IEEE 802.15.4 standard, a node can be in one of the following four states: transmit, receive, idle, and sleep. A sensor node enters in transmit state as long as it is sending data packet. When in transmit state, the radio module stays in transmit mode and the MCU module stays in active mode for the procession of data packet. A node stays in the receive state as long as it is receiving a packet or performing CCA. And then, its radio module turns to receive mode and the MCU module stays in active mode to process received data or to identify whether the channel is clear or not. If there is no packet to be transmitted while the node is still in the active period or it stays at the backoff period, the node is in idle state. And then, the radio module is in receive mode and the MCU module is assumed to be in LPM3. Because in LPM3, the main clock source and the timer oscillator should be enabled in the MCU module to support for scheduled wake-ups. When a node is in the inactive period, it stays in sleep state with its radio module in sleep mode and its MCU module is assumed to be in LPM4, that is, the deepest sleep mode. Hence, the power consumption of the mentioned modes is shown in Table 1.

Table 1.

Power consumption of CC2420 and ATmega128L in different states.

Devices	Modes	Energy consumption (mW)
TI-MSP430	Active mode	1.5
	Idle mode	0.006
	Power down mode	0.0006
CC2420	Transmit mode (maximum)	52.2
	Receive mode	59.1
	Idle mode	0.06
	Sleep mode	0.003 mW

Analytical model

The occurrence of a successful packet reception can be attributed to a series of layers. The proposed model is targeted to theoretically analyze the impact of physical layer and MAC layer on successful communication. For the physical layer and the MAC layer, packets are successfully received due to two reasons: successful channel access and successful packet transmission. In this analytical model, the key parameters of both the physical layer and the MAC layer that concerning the channel access operations and packet transmission are taken into consideration, as is shown in Figure 1. It should be noticed that the proposed model does not take the correlation and interaction between two layers into consideration. Since the impact of each layer on communication is independent. With the successful handshaking probability and the successful CCA probability, network performance metrics including network reliability, time efficiency, and energy efficiency can be estimated easily.

Figure 1.

Frame diagram of proposed analytical model.

According to the datasheet of the Chipcon CC2420 radio module, the radio module provides eight transmission power levels, which is listed in Table 2 together with corresponding output power and power consumption.

Table 2.

Transmission power levels with output powers and power consumptions.²⁸

Power level	Output power (dBm)	Power consumption (mW)
31	0	52.2000
27	−1	49.5000
23	−3	45.6000
19	−5	41.7000
15	−7	37.5000
11	−10	33.6000
7	−15	29.7000
3	−25	25.5000

Considering a data packet is transmitted at power level $l$ , the received signal power with a distance of $d (m)$ away from the transmitter can be expressed as

P^{rx} (l) (dBm) = P^{tx} (l) (dBm) - PL (d) (dB)

(6)

where $PL (d) (dB)$ is the path loss. In this research, the lognormal path loss model is taken as the reference since it can capture the random variations of path loss more accurately than either the two-ray model or the log-distance model does.³⁰ The form of lognormal path loss model can be expressed as

P L_{ij} (d) (dB) = PL (d_{0}) (dB) + 10 η \log (\frac{d}{d_{0}}) + χ

(7)

where $d$ is the distance between transmitter and receiver, $d_{0}$ is the reference distance, $PL (d_{0}) (dB)$ is the path loss at the reference distance, $η$ is the path loss exponent and $χ$ is a zero-mean Gaussian random variable with the standard deviation $χ (dB)$ to model large-scale fading effects. It should be noticed that the lognormal path loss model can be applied to general monitoring scenarios, such as industrial process monitoring and advanced metering infrastructures in the smart grid. While for wireless body area networks, additional losses should be added.³¹ Therefore, the signal to noise ratio (SNR) that due to the transmission power at level $l$ and the propagation distance of $d (m)$ can be expressed as

ψ (l, d) (dB) = P^{tx} (l) (dBm) - LF (d) (dB) - P_{n} (dBm)

(8)

To cope with bit errors caused by noise, in digital communication, signals are always been modulated before transmission. To present the effect of modulation in improving the reliability of communication, we tried to transmit a signal in two different ways: one is to transmit it with offset-quadrature phase-shift keying (O-QPSK) modulation and the other way is to transmit it directly without any modulation. In Figure 2, it can be found that the transmitted signal with modulation can always obtain lower bit error rate (BER) than without modulation, especially when the SNR is high.

Figure 2.

BER performance comparison between O-QPSK modulation and without modulation.

The BER of binary phase-shift keying (BPSK) modulation can be simply expressed with Q-function as

ber (l, d) = Q (\sqrt{2 ψ (l, d)})

(9)

where the Q-function is defined as

Q (x) = \frac{1}{\sqrt{2 π}} \int_{x}^{\infty} e^{- \frac{x^{2}}{2}} dx

(10)

For the O-QPSK modulation scheme, it consists of two BPSK modulations on in-phase and quadrature components of the signal. Hence, its BER can also be expressed as

ber (l, d) = Q (\sqrt{2 ψ (l, d) \cdot G_{p}})

(11)

where $G_{p}$ is the process gain and equal to 8 for CC2420 radios.

Hence, the probability of a successful packet reception of a $b$ bits packet transmitted at power level $l$ over a distance of $d (m)$ is

p_{tx}^{S} (l, d, b) = {(1 - Q (\sqrt{2 \cdot 8 \cdot ψ (l, d)}))}^{b}

(12)

And the corresponding failure probability is

p_{tx}^{F} (l, d, b) = 1 - {(1 - Q (\sqrt{2 \cdot 8 \cdot ψ (l, d)}))}^{b}

(13)

Based on the above analysis, the packet reception probability relies on the transmission power level, the communication distance, and the length of packet. Since all communications are assumed of adopting handshaking mechanism, successful communication is the combination of successful data packet reception and successful ACK packet reception. Hence, the probability of a successful handshake occurs when a $b$ bits data packet is transmitted from one sensor node with power level $l$ and replied with a $b^{ack}$ bits ACK packet from another node with power level $k$ from $d (m)$ away. Hence, the probability of a successful handshake can be expressed as

p_{HS}^{S} (l, k, b, b^{ack}, d) = p_{tx}^{S} (l, b, d) \cdot p_{tx}^{S} (k, b^{ack}, d)

(14)

And the probability of a failed handshake should be

p_{HS}^{F} (l, k, b, b^{ack}, d) = 1 - p_{HS}^{S} (l, k, b, b^{ack}, d)

(15)

On the MAC layer, sensor nodes start to perform CCA attempt as a new data packet arrives. Let $p_{CCA}^{S}$ depicts the probability of successfully passing CCA. Every time sensor nodes fail to access to the channel, a new CCA process starts. Considering all chances until the maximum number of backoffs $N_{B}^{(max)}$ is reached, the aggregate arrival rate of CCA attempts for one channel $λ_{CCA}$ can be represented in terms of $λ$

λ_{CCA} = λ \cdot \sum_{k = 0}^{N_{B}^{(max)}} {(1 - p_{CCA}^{S})}^{k}

(16)

Since the packet arrival rate at one channel is assumed to follow a Poisson’s distribution, the CCA attempt rate of a channel $λ_{CCA}$ also follows a Poisson’s distribution with fixed $N_{B}^{(max)}$ and $p_{CCA}^{S}$ . According to the slotted CSMA/CA algorithm, sensor nodes should perform the CCA process to check whether a channel is idle or not before transmission. If the channel is busy, it will wait for a random period of time and check again. The backoff period is a random number of unit backoff slots from 0 to $2^{BE} - 1$ . With the assistance of backoff, nodes can avoid the failure of CCA to some extent. Larger $N_{B}^{(max)}$ offers more chances of performing CCA. Lager backoff exponent $BE$ offers nodes more backoff delay options, which will enlarge the disparity in backoff duration among nodes to decrease the probability of CCA failure. IEEE 802.15.4 slotted CSMA/CA algorithm bounds the value of backoff exponent between minimum backoff exponent value $B E_{min}$ to maximum backoff exponent value $B E_{max}$ . It is to be noted that the maximum value allowed for minimum BE by the IEEE 802.15.4 standard is equal to 8.³² Hence, the probability of successfully passing CCA $p_{CCA}^{S}$ is affected by $N_{B}^{(max)}$ and $B E_{min}$ .

As is shown in Figure 3, according to whether the time interval between two successive CCAs on one channel is larger than one complete handshaking duration or not, the probability of successful CCA can be classified into two cases. According to case 1, if the time interval between the current CCA attempt and previous CCA attempt on one channel is larger than one complete handshaking duration, the current CCA will succeed for the length of time interval is enough for handshaking. In this notation, the backoff period is included in the time interval for the first CCA of a new data packet. According to the theory of Poisson’s distribution,³³ the distribution of time interval between two consecutive CCA attempts follows the exponential distribution. Therefore, the probability of case 1 can be expressed as

p_{CCA}^{CI} = p [T_{n} \geq T_{HS}^{S}] = e^{- λ_{CCA} \cdot T_{HS}^{S}}

(17)

Figure 3.

Two kinds of time interval between successive CCA.

A complete operation of handshaking includes the time of transmitting data packet, the time of waiting for ACK packet, and the time of receiving ACK packet. Accordingly, the time of one complete handshaking operation $T_{HS}^{S}$ is

T_{HS}^{S} = T_{pkt} + T_{ack} + T_{wait}

(18)

where $T_{pkt}$ and $T_{ack}$ are the transmission time of data packet and ACK packet, respectively. Term $T_{wait}$ is the time interval between the end of data packet transmission and the beginning of ACK packet reception.

As is shown by case 2 in Figure 3, if the time interval between current CCA attempt and previous CCA attempt on this channel is no larger than one complete handshaking duration, the sensor node of performing current CCA attempt can take its previous backoff period as a reference to do the primary evaluation. Supposing that all nodes are staying at the same backoff stage $S$ , one node’s CCA attempt can be successful only if no CCA attempts succeed within the previous $T_{HS}^{S}$ , which can be expressed as

p_{CCA}^{BE} = 1 - [1 - {(1 - \frac{1}{2^{BE} - 1})}^{N_{C} - 1}] \cdot p_{CCA}^{BE} \cdot \frac{T_{HS}^{S}}{T_{bk, unit}}

(19)

where $T_{bk, unit}$ is the time duration of unit backoff slot. Hence, the probability of successfully passing CCA can be expressed as

p_{CCA}^{S} = 1 - Π_{i = 0}^{N_{B}^{(max)}} (1 - (p_{CCA}^{CI} + (1 - p_{CCA}^{CI}) \cdot p_{CCA}^{BE} (i)))

(20)

Critical issues

Reliability

Reliability is denoted as the probability of successful packet reception $p^{S}$ . According to previous analyses on the physical layer and the MAC layer, packets are unsuccessfully received due to two reasons: channel access failure and packet collision. Channel access failure happens when a packet fails to access the channel within $N_{B}^{(max)}$ times of CCAs. And a packet is discarded if it cannot be received within $n$ times of retry. Hence, the probability of successful packet reception can be expressed with the probability of successfully passing CCA $p_{CCA}^{S}$ and the probability of a successful handshake $p_{HS}^{S}$

p^{S} = \sum_{j = 1}^{n} p_{CCA}^{S} \cdot {(1 - p_{HS}^{S})}^{j - 1} \cdot p_{HS}^{S}

(21)

Time efficiency

Time efficiency is denoted as the excepted time interval for a successfully received packet from ready to be transmitted, until the transmission is successful and the ACK is received, that is, expected transmission latency $T_{DL}^{E}$ . The delay mainly comes from two operations: backoff operations on the MAC layer and handshaking operations on the physical layer. Based on the analytical model proposed earlier, the expected duration of the delay caused by backoff operation can be expressed as

\begin{matrix} T_{BF}^{E} = \\ \sum_{k = 0}^{N_{B}^{(max)}} \sum_{i = 0}^{k} (E (2^{min (BE + i, B E_{max})} - 1) T_{bkf} + 2 T_{CCA}) \cdot p_{CCA}^{S} {(p_{CCA}^{F})}^{k} \end{matrix}

(22)

The expected delay caused by successful handshaking is given in equation (18). The expected delay caused by failed handshaking includes two parts, that is, the transmission time of data packet $T_{pkt}$ and the maximum duration of waiting for acknowledgment $T_{wait}^{max}$

T_{HS}^{F} = T_{pkt} + T_{wait}^{max}

(23)

where the maximum duration of waiting for acknowledgment $T_{wait}^{max}$ is 864 µs.

Hence, the expected transmission latency with the retransmission times as $n$ can be expressed as

\begin{matrix} T_{DL}^{E} = \sum_{j = 0}^{n} (T_{BF}^{E} (j + 1) + T_{HS}^{F} j + T_{HS}^{S}) \cdot p_{HS}^{S} {(p_{HS}^{F})}^{j} \\ + (T_{BF}^{E} + T_{HS}^{F}) (j + 1) \cdot {(p_{HS}^{F})}^{j + 1} \end{matrix}

(24)

Energy efficiency

The operation of transmitting a packet is composed of performing CCA, backoff, data packet transmission, and ACK packet reception, which have been carried out earlier in the node’s behavior analysis. And the energy consumption of each operation can be estimated as the product of its duration and its power. With the expected latency of successful handshaking and failed handshaking, the expected energy consumption of handshaking operations can be derived similarly

E_{HS}^{S} = P^{tx} \cdot T_{tx}^{pkt} + P^{idle} \cdot T_{ack} + P^{rx} \cdot T_{rx}^{ack}

(25)

E_{HS}^{F} = P^{tx} \cdot T_{tx}^{pkt} + P^{rx} \cdot T_{wait}^{max}

(26)

The expected energy consumption caused by backoff operations can be expressed as

\begin{array}{l} E_{B F}^{E} = \sum_{k = 0}^{N_{B}^{(\max)}} \sum_{i = 0}^{k} (E (2^{\min (B E + i, B E_{\max})} - 1) T_{b k f} P^{i d l e} + 2 T_{C C A} P^{r x}) \cdot \\ p_{C C A}^{S} {(p_{C C A}^{F})}^{k} \end{array}

(27)

Hence, the expected energy consumption of successful transmitting a packet is

\begin{matrix} E^{E} = \sum_{j = 0}^{n} (E_{BF}^{E} (j + 1) + E_{HS}^{F} j + E_{HS}^{S}) \cdot p_{HS}^{S} {(p_{HS}^{F})}^{j} \\ + (E_{BF}^{E} + E_{HS}^{F}) (j + 1) \cdot {(p_{HS}^{F})}^{j + 1} \end{matrix}

(28)

In the following section, the energy efficiency is measured by energy per megabits (EPMs), the energy consumed by sending 1 MB of data with the unit as mJ

E_{MB}^{E} = E^{E} \cdot \frac{1 MB}{b \cdot p^{S}} .

(29)

Simulation and analysis

Simulation setup

In the following simulations, WSN devices are uniformly distributed within an area of 100 m × 100 m and operate in the 2.4-GHz industrial, scientific and medical (ISM) radio band with 16 workable channels. We borrow one of the path loss models from Kilic and Gungor³⁴ as $η = 1.64$ , $χ = 3.29 dB$ , $d_{0} = 1 m$ , and $PL (d_{0}) (dB) = 71.84 dB$ , which is derived from real smart grid environment measurements. The noise power $P_{n}$ at the receiver can be evaluated as 100 dBm according to Pöttner et al.³⁵ The power consumptions of a device in different modes are according to Table 1. Key parameters are assumed to follow the IEEE 802.15.4 standard³² and TelosB mote Module,⁶ and they are listed in Table 3 together with parameters of simulation scenarios.

Table 3.

Key parameters of simulation.

Parameter	Range	Default
$T_{bk, unit}$	–	320 µs
$T_{wait, max}$	–	864 µs
$T_{cca}$	–	128 µs
$N_{B}^{(max)}$	0–5	4
$B E_{min}$	3–8	5
$B E_{max}$	–	8
$P^{tx} (l)$	[0, −1, −3, −5, −7, −10, −15, −25] dBm	0 dBm
$P^{rx} (l)$		−100 dBm
$λ$	0.1–2 (packets per second)	0.5 (packets per second)
$N_{C}$	1–225	100
$b$ (with 8-byte header)	28–128 (bytes)	60 (bytes)
$b^{ack}$	–	12 (bytes)

The value of the time interval between the end of data packet transmission and the beginning of acknowledgment packet reception $T_{wait}$ is assumed as 360 µs, according to the measurement result of Pöttner et al.³⁵ In the following simulations, parameters are equal to the default value unless specified. And all the optimal values of key parameters are obtained through exhaust algorithms, which is reliable to search for the global optimum solution.

Rational validation

To validate whether the proposed analytical model is rational or not, a simulation model based on the OPNET simulator is taken as a reference.³⁶ The validation is carried out from two perspectives: different network scales and different traffic loads.

Figure 4(a)–(c) presents the result of validation from the traffic load perspective. From this perspective, 100 nodes are deployed within the square area with each node’s packet arrival rate $λ$ increases from 0.1 packets per second to two packets per second. Figure 4(d)–(f) presents the validation from different network scales, where from one node up to 225 nodes are deployed within the square area with a constant packet arrival rate of 0.5 packets per second.

Figure 4.

Performance comparisons between analytical models and simulated model under different traffic loads and different network scales.

As is shown in Figure 4(a) and (d), the reliability decreases slowly at first and then sharply as the number of devices increases and the packet arrival rate gets higher. Similarly, Figure 4(b) and (e) shows network with a larger scale and higher packet arrival rate will have a higher value of average delay. Figure 4(c) and (f) further compares the performance of two models in terms of power consumption. This could be explained in conjunction with the behavior of the physical layer and the MAC layer. A large number of nodes in dense networks will experience more serious fading on the link due to heavier interference and longer average transmission distance. A higher arrival rate of data packet brings a higher network load, which promotes more intensive CCA attempts and longer backoff duration, which then aggregate the competition of channel access and finally lead to substantial transmission failure.

As is shown in Figure 4, the proposed analytical model fits the simulation model well in terms of reliability, transmission latency, and EPM with small network scale and low packet arrival rate. Even though significant gaps appear as the number of nodes and the packet arrival rate increases, the analytical model always behaves as an upper bound of the simulation model in reliability and as a lower bound in transmission latency and EPM. From equation (1), we can find out large network scale and frequent data packet arrival bring each channel with higher value of $N_{C}$ and $n_{T}$ , which increases the value of $p_{C}$ . According to equation (3), smaller value of $p_{C}$ leads the value of $(1 - p_{C})^{- 1 / p_{C}}$ more close to $e$ , which makes the traffic on one channel approximates to a Poisson’s process.

Numerical analysis

The following simulation is performed under a typical scenario of 100 nodes deployed in the 100 m × 100 m area. To test the performance of both layer’s parameter optimization, default parameter setting according to IEEE 802.15.4, physical layer–focused parameter optimization and MAC layer–focused parameter optimization are also presented.

General network

In this part of analyses, the default parameter settings are taken as a baseline setting to compare with. And the comparison is elaborated from two perspectives, that is, communication performance and energy effectiveness. In communication performance comparison, we take the energy efficiency of default parameter settings as the upper bound $E_{MB, max}^{E}$ . Three optimization methods present their best communication performance with their EPM no more than $E_{MB, max}^{E}$ . Figure 5(a)–(c) presents the comparisons among different parameter settings given fixed 100 nodes but with $λ$ ranges from 0.1 to 2. As is shown in Figure 5(a), no significant gaps present among four sets of parameters in terms of successful packet reception probability when $λ$ is lower than 0.5. While gaps present as the arrival of data packet getting intense. So does transmission comparisons shown in Figure 5(b), the differences getting obvious as $λ$ is larger than 0.5. While the EPM curve of default parameters always behaves as an upper bound regardless of the variation of data packet arrival rate. Similarly, the performances of four different parameter settings are very close with the network scale of less than 30. While as the network scale increases, the differences getting more obvious in terms of both successful packet reception probability and expected transmission latency. However, the EPMs of three optimized parameter settings are always lower than the default setting. From Figure 5, we observe that optimization considering both layers outperforms the rest optimizations and default setting in terms of both network reliability and time efficiency under various kinds of network scales and traffic loads.

Figure 5.

Data transport performance comparisons among different optimization methods under different traffic loads and different network scales.

In energy effectiveness comparison, we present the best energy effectiveness of three optimization methods with their communication performance no worse than default parameter settings. As is shown in Figure 6, three optimized parameter sets are all managed to meet the data transport performance constraints defined by default setting with higher energy efficiency. It should be noticed that the advantage of optimization considering both layers is more obvious in networks with more than 100 nodes and packet arrival rate more than 1.8. Hence, the optimization considering both layers outperforms the rest optimizations in terms of both data transport performance and energy efficiency under different network scales and traffic loads, especially in large-scale networks with heavy traffic loads.

Figure 6.

Energy effectiveness comparisons among different optimization methods under different traffic loads and different network scales.

Metering network

Different from general networks, the metering network has stricter requirements on both network reliability and transmission latency. In this part of optimization, we set the constraint of successful packet reception probability $p_{min}^{S}$ as high as 90%. Yu et al.³⁷ give the average value of refresh interval of metering networks as 500 ms, which can be taken as the expected average transmission latency $T_{DL, max}^{E}$ . Hence, the object of this optimization is to find the minimum EPM while guaranteeing $p^{S}$ no less than 90% and $T_{DL}^{E}$ no more than 500 ms.

Figure 7 gives a comparison between different optimization methods. With the requirements of successful packet reception probability no less than 90% and transmission latency no more than 500 ms, adequate results are presented in Figure 7 with corresponding energy consumption. By comparing Figure 7(d) with Figure 7(a)–(c), we can find out that with the assistance of joint layer parameter optimization, sensor nodes can meet the requirements of data transport performance but with lower energy consumption than other optimizations. Besides, joint layer parameter optimization enables the best network scalability than other optimizations since it guarantees sensor nodes working in denser networks with heavier traffic loads while still meets the network performance constraints, which is impossible with the assistance of other optimizations.

Figure 7.

Comparison between different optimization methods targeting industrial metering networks: (a) default parameter settings, (b) physical layer parameter optimization, (c) MAC layer parameter optimization, and (d) joint layer parameter optimization.

The optimal parameter settings derived by joint layer parameter optimization is shown in Figure 8. Nodes in small-scale networks with low traffic load can send larger packets with lower transmission power, more backoff times, and small backoff exponent. While as the network scale and traffic load increase, nodes should send smaller packets with higher transmission power, fewer backoff times, and large backoff exponent. It should be noticed that although Figure 8(a) indicates the size of data packet can be adjusted widely, it will be affected by the requirement of metering applications, such as the precision of sampling. From Figure 8(c), we can find that the available range of transmission power’s optimal value is limited. However, Figure 8(b) and (d) indicates that both the maximum number of backoffs and minimum backoff exponent can be adjusted within a large range. Therefore, the maximum number of backoffs and the minimum backoff exponent should be adjusted prior to other parameters in metering network optimizations.

Figure 8.

Optimal parameter settings with different network scales and traffic loads targeting metering networks: (a) data packet size, (b) maximum number of backoffs, (c) transmission power, and (d) minimum backoff exponent.

Besides, in Figure 8(d), the optimal value of minimum backoff exponent changes with network scale and traffic load, which is different from the result shown in Elshabrawy et al.⁷ In networks with the traffic approximates to Poisson’s distribution, prolonging the duration of backoff has a limited contribution to successful packet reception probability. Since one of the basic properties of Poisson’s process is each event is stochastically independent of all the other events in the process.³³ On the contrary, larger value of backoff exponent brings higher average end-to-end delay, which will increase the probability of exceeding the transmission latency constraint $T_{DL, max}^{E}$ .

Conclusion

To cope with different application characteristics, both communication link states, which may subject to path loss and interference, as well as the competition for medium access, which becomes inevitable with a large number of devices, should be equally considered. In this article, a joint analytical model considering both the physical layer and the MAC layer was proposed at first. On the physical layer, the model referred to the actual working modes of the TelosB mote Module and formulated the relationship among transmission power level, transmission distance, data packet size, modulation way, and successful handshaking probability. On the MAC layer, based on the approximation of Poisson’s distributed traffic on each channel, a formula to calculate the probability of successful channel access was obtained. Simulation results suggested that the proposed analytical model can precisely estimate networks’ communication performance and energy efficiency of different network scales as well as traffic loads. Then, we compared the effect of adjusting different layer’s parameters on improving networks’ communication performance and energy efficiency. The results showed that adjusting both layers’ parameters can provide better reliability, time efficiency, and energy efficiency than only focusing on a single layer’s parameters. Finally, to fulfill the communication requirement of metering applications, key parameters’ optimal value under different network scales and traffic loads were presented. By comparing the available range of different key parameters’ optimal value, the maximum number of backoffs and the minimum backoff exponent are considered to be more suitable for the optimization of the metering network.

In the following research, several aspects of the proposed model will be improved. First and foremost, the proposed model should be compatible with more kinds of modulation schemes. Based on the improvement, the impact of different modulation schemes on improving the performance of communication should be tested. Second, the key parameters and key features of network layer should be considered. More than physical layer and MAC layer, the impact of routing on the network layer will also be analyzed. Finally, more WSN-based network applications should be analyzed. The analytical model can be modified according to the network scales, traffic patterns, and communication performance requirements of different application scenarios, so as to comprehensively and accurately estimate the communication performance, energy efficiency, and scalability of the network.

Footnotes

Handling Editor: Yanjiao Chen

Author contributions

Y.W. and W.Y. contributed to the conceptualization of this study. Y.W. and W.Y. contributed to the methodology of this study. Y.W. R.H., and T.W. contributed to the software. Y.W., T.W., and H.Z. contributed to the validation. Y.W. contributed to the writing—original draft preparation. R.H., W.Y., and H.Z. contributed to the writing—review and editing. W.Y. contributed to the funding acquisition.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Natural Science Foundation of China under grant (51874299), the National Science and Technology Pillar Program of China (2013BAK06B03), and the Fundamental Research Funds for the Central Universities (2018YJS004).

ORCID iD

Yipeng Wang

References

Güngör

VÇ

Hancke

. Industrial wireless sensor networks: applications, protocols, and standards. Boca Raton, FL: CRC Press, 2013.

Benzi

Anglani

Bassi

, et al. Electricity smart meters interfacing the households. IEEE Trans Ind Electron 2011; 58: 4487–4494.

Elshabrawy

Ashour

Robert

. Theoretical analysis of report success probability in IEEE 802.15.4-based smart utility networks. In: Proceedings of the 2016 IEEE 84th vehicular technology conference (VTC-Fall), Montreal, QC, 18-21 September 2016. New York: IEEE.

Salam

Khan

. IWSN-standards, challenges and future. IEEE Potent 2016; 35: 9–16.

Wang

Zhao

Cai

, et al. Bidirectional prediction-based underwater data collection protocol for end-edge-cloud orchestrated system. IEEE Trans Ind Inf 2019; 16: 4791–4799.

Data Sheet TelosB. Crossbow technologies, http://www2.ece.ohio-state.edu/;bibyk/ee582/telosMote.pdf (accessed 1 December 2020).

Elshabrawy

Shereen

Ashour

, et al. Report success probability/battery lifetime analysis of dense IEEE 802.15.4-based metering networks with hidden nodes. IEEE Sens J 2017; 17: 2259–2266.

Nandi

Kundu

. Optimal transmit power and packet size in wireless sensor networks in lognormal shadowed environment. Int J Sens Netw 2012; 11: 81.

Nandi

Kundu

. On energy level performance of adaptive power based WSN in shadowed channel. In: Proceedings of the international conference on devices and communications, Mesra, India, 24-25 February 2011, pp. 1-5. New York: IEEE.

10.

Xiang

Luo

Rosenberg

. Compressed data aggregation: energy efficient and high fidelity data collection. IEEE/ACM Trans Netw 2013; 21: 1722–1735.

11.

Kapnadak

Coyle

. Optimal non-uniform deployment of sensors for detection in single-hop wireless sensor networks. In: Proceedings of the 2011 8th annual IEEE Communications Society Conference on sensor mesh and Ad hoc communications and networks, Salt Lake City, UT, 27-30 June 2011, pp. 89-97. New York: IEEE.

12.

Leinonen

Codreanu

Juntti

. Distributed consensus based joint resource and routing optimization in wireless sensor networks. In: Proceedings of the 2012 conference record of the forty sixth Asilomar conference on signals, systems and computers (ASILOMAR), Pacific Grove, CA, 4-7 November 2012. New York: IEEE.

13.

Cotuk

Bicakci

Tavli

, et al. The impact of transmission power control strategies on lifetime of wireless sensor networks. IEEE Trans Comput 2013; 63: 2866–2879.

14.

Koulali

Kobbane

El Koutbi

, et al. Dynamic power control for energy harvesting wireless multimedia sensor networks. J Wireless Commun Netw 2012; 2012: 158-150.

15.

Datta

Kundu

. Packet size optimization for multi hop CDMA wireless sensor networks with nearest neighbors based routing. In: Proceedings of the 2012 third international conference on emerging applications of information technology (EAIT), Kolkata, India, 30 November-1 December 2012. New York: IEEE.

16.

Kurt

Yildiz

Yigit

, et al. Packet size optimization in wireless sensor networks for smart grid applications. IEEE Trans Ind Electron 2017; 64: 2392-2401.

17.

Yildiz

Kurt

Tavli

. Comparative analysis of transmission power level and packet size optimization strategies for WSNs. IEEE Syst J 2018; 13: 2264–2274.

18.

Elwekeil

Abdalzaher

Seddik

. Prolonging smart grid network lifetime through optimising number of sensor nodes and packet length. IET Commun 2019; 13: 2478–2484.

19.

Yildiz

Gungor

Tavli

. Packet size optimization for lifetime maximization in underwater acoustic sensor networks. IEEE Trans Ind Inf 2018; 15: 719–729.

20.

Bianchi

. Performance analysis of the IEEE 802.11 distributed coordination function. IEEE J Sel Areas Commun 2000; 18: 535–547.

21.

Park

Marco

Soldati

, et al. A generalized Markov chain model for effective analysis of slotted IEEE 802.15.4. In: Proceedings of the 2009 IEEE 6th international conference on mobile adhoc and sensor systems, Macau, China, 12-15 October 2009. New York: IEEE.

22.

Jung

Hwang

Sung

, et al. Enhanced Markov chain model and throughput analysis of the slotted CSMA/CA for IEEE 802.15.4 under unsaturated traffic conditions. IEEE Trans Veh Technol 2009; 58: 473–478.

23.

Al-Anbagi

Erol-Kantarci

Mouftah

. Time slot allocation in WSNs for differentiated smart grid traffic. In: Proceedings of the 2013 IEEE electrical power & energy conference, Halifax, NS, Canada, 21-23 August 2014. New York: IEEE.

24.

Yousefi

HHN

Kavian

Mahmoudi

. A Markov model for investigating the impact of IEEE 802. 15.4 MAC layer parameters and number of clusters on the performance of wireless sensor networks. Wirel Netw 2019; 25: 4415–4430.

25.

Wang

Yang

Han

, et al. An analytical framework for the IEEE 802.15. 4 MAC layer protocol under periodic traffic. Sensors 2020; 20: 3350.

26.

Alkama

Bouallouche-Medjkoune

Bachiri

. Modeling and performance evaluation of the IEEE 802.15. 4K CSMA/CA with priority channel access mechanism under fading channel. Wirel Pers Commun 2020; 115: 1-30.

27.

Gamal

Sadek

Rizk

, et al. Optimization and modeling of modified unslotted CSMA/CA for wireless sensor networks. Alex Eng J 2020; 59: 681–691.

28.

Texas Instruments. CC2420: 2.4 GHz IEEE 802.15. 4/Zigbee-ready RF Transceiver. 2006. https://www.ti.com/lit/ds/symlink/cc2420.pdf?ts=1606106746306&ref_url=https%253A%252F%252Fwww.google.com%252F

29.

Texas Instruments. MSP430 F149 datasheet, https://www.ti.com/product/MSP430F149 (accessed 3 June 2004).

30.

Kurt

Tavli

. Path-loss modeling for wireless sensor networks: a review of models and comparative evaluations. IEEE Antennas Propag Mag 2017; 59: 18–37.

31.

Abdel-Rahim

Habaebi

Chebil

, et al. An indoor path loss model for wireless sensor networks. Int J Ultra Wideband Commun Syst 2018; 3: 192–200.

32.

Association IS. IEEE Std 802.15. 4-2011, IEEE standard for local and metropolitan area networks—part 15.4: Low-rate wireless personal area networks (LR-WPANs). IEEE. Epub ahead of print 5 September 2011. DOI: 10.1109/IEEESTD.2011.6012487.

33.

Croarkin

Tobias

Filliben

, et al. NIST/SEMATECH e-handbook of statistical methods. NIST/SEMATECH, http://wwwitlnistgov/div898/handbook (accessed July 2006).

34.

Kilic

Gungor

. Analysis of low power wireless links in smart grid environments. Comput Netw 2013; 57: 1192–1203.

35.

Pöttner

Schildt

Meyer

, et al. Piggy-backing link quality measurements to IEEE 802.15.4 acknowledgements. In: Proceedings of the IEEE 8th international conference on mobile adhoc and sensor systems, MASS 2011, Valencia, Spain, 17-22 October 2011. New York: IEEE.

36.

Jurcík

Koubâa

. The IEEE 802.15. 4 OPNET simulation model: reference guide v2.0, IPP-HURRAY. Technical Report. HURRAY-TR-070509, May 2007.

37.

Gidlund

Åkerberg

, et al. Reliable RSS-based routing protocol for industrial wireless sensor networks. In: Proceedings of the IECON 2012-38th annual conference on IEEE Industrial Electronics Society, Montreal, QC, Canada, 25-28 October 2012, pp. 3231-3237. New York: IEEE.