Intelligent Time Synchronization Protocol for

Energy Efficient Sensor Systems

Jalil Boudjadar1 and Mads Mørk Beck2

1
Aarhus University, Denmark,
2
ReMoni ApS, Aarhus Denmark

Abstract. The ubiquity of sensors has offered great opportunities to

monitoring infrastructures to track the real-time state of observed en-
vironments. Monitoring systems rely on synchronous massive data to
calculate the environment state. However, collecting and communicat-
ing massive data usually leads to expensive energy consumption due to
high sampling frequency and communication cost for the sensors opera-
tion. Clamp-on wireless sensors are a flexible plug-and-play technology
easy to deploy where sensors operate on integrated batteries. This paper
introduces a highly accurate time synchronization protocol to coordi-
nate the operation of individual clamp-on sensors deployed as an energy-
constrained monitoring solution. While maintaining sensors highly syn-
chronized, the proposed protocol reduces drastically data communication
and energy consumption of the sensor network. To minimize the energy
consumption further, we optimize the protocol behavior using a genetic
algorithm. The protocol has been modeled in VDM for formal analysis,
implemented and compared to a state of the art protocol.

Keywords: Sensor networks, time synchronization, energy efficiency,

intelligent control

1 Introduction
The EU Commission has stated that all new buildings after 2021 should be
Nearly Zero Energy Buildings (NZEB) [1]. This will be achieved through inte-
grating local green energy resources in buildings, so that individual buildings
become almost self-satisfied in terms of energy; technically known by islanded
micro-grids [7]. This will require smart grid systems to guarantee profitable
deployment of such infrastructures. The reason behind this is that buildings
can generate enough green energy on their own to satisfy the private demand,
whereas the extra energy can be sold to the grid. Achieving a transparent prof-
itable business model for NZEB requires a monitoring of the building energy
generation and consumption with high accuracy and low latency, as false read-
ings could result in energy shortage or expensive bills.
Different monitoring solutions have been proposed to use sensors to mea-
sure energy and collect data, delivering thus a real time energy state for NZEBs
[11,12,13,14,8]. The conventional solutions often require many sensors that col-
lect large amount of data. This comes at a significant energy cost, both related
2 J Boudjadar et al.

to collecting, communicating and processing data. This is complaint with the

EU’s new green vision, and motivates the use of a fine grained monitoring model
where sensors are deployed at appliance level instead of just the in-going power
to the building. The rational behind this recommendation is to simply identify
energy waste within buildings and optimize the underlying energy management.
Achieving this find-grained monitoring requires a constellation of sensor net-
works, data communication and processing [15]. Data driven solutions rely on
collecting massive data to obtain accurate tracking of the energy dynamics over
time.
Clamp-on wireless sensors are an important low cost ingredient to build such
monitoring solutions given their flexibility and mobility [16]. A clamp-on sensor
enables to measure both voltage and current which are the main parameters in
calculating the energy consumption. Collecting both parameters from different
sensors must be synchronized so that energy consumption is calculated at the
timestamp at which the data is sampled. The main challenge is that such sensors
operate on batteries and have very limited computation resources to track the
time advancement, thus individual clocks can drift from the master node clock.
This requires the sensors to synchronize frequently with the master node in order
to calibrate their individual clocks. The time synchronization gets even worse in
case of a failure, e.g. loss of communication to the master node.
Different time synchronization protocols have been proposed for energy-
efficient operation of sensor networks [19,20,21,22,6,17]. The common denom-
inator of most of the time synchronization protocols is that the time accuracy is
obtained at the expense of energy consumption. In this paper, we propose an in-
telligent time synchronization protocol to coordinate the operation of individual
clamp-on sensors deployed as an energy-constrained monitoring solution. The
proposed protocol is able to deliver perfect time synchronization (0-delay) to
prevent the drifting of sensor clocks while increasing the energy efficiency of the
sensors. The protocol has been formally modeled in VDM and analyzed using
combinatorial testing against deployment constraints. Furthermore, a prototype
has been implemented and examined on an actual case study.
The rest of the paper is structured as follows: Section 2 introduces the overall
system architecture and modeling. Relevant related work is described in Section
3. Section 4 presents our energy-efficient time synchronization protocol. Analysis
of both time accuracy and energy efficiency of the proposed protocol is provided
in Section 5. Implementation is provided in Section 6. Finally, Section 7 concludes
the paper.

2 System Architecture and Modeling

This section describes the system architecture and elements, beside to what is
challenging in the time synchronization of clamp-on wireless sensor networks.
The architecture of the wireless sensor network (WSN) is depicted in Figure. 1.
It consists of a set of sensors sampling voltage and current data in real-time,
and a master node (sink) where all data is deposited. The master node commu-
 Time Synchronization for Intelligent Sensor Systems 3

nicates sensors data further to the cloud where control decisions are calculated,
i.e what energy resource to schedule?; when to sell energy to the grid?; etc. Sen-
sors operate on integrated batteries and communicate to the master node in a
wireless way to forward data and receive time synchronization messages given
that the sensors do not have physical clocks; reduced hardware to reduce energy
consumption of the sensors. The master node is able to track physical time given
that it has a clock. To prevent time drifting of sensors, the master node assists
by sending timestamp message (sync) frequently to the sensors to calibrate their
local time.
Fig. 1. Overall system architecture

panels

The energy consumption P at any time point t is calculated using the fol-
lowing equation:
P (t) = V (t).I(t). cos(φ)
where V (t) is the voltage at time t, I(t) is the current at time t and φ is the
angle between voltage and current. This requires that both voltage and current
must be taken at the same time point in order to obtain meaningful readings. If
the sensor readings are non-synchronous, the actual energy consumption state
will be inaccurate, which in turn leads to misleading control actuations.
4 J Boudjadar et al.

In Figure 2, one can Fig. 2. Data synchronization problem

see that although data
is sampled at the same Sensor1 Master node Sensor2
time point in sensor1
and sensor2, the arrival
of data from sensor1 to Current reading
Voltage reading
Master node is much
earlier than data from
sensor2. This in fact
due to the communi-
Current reading
cation environment and
Voltage reading
data size sent by dif-
ferent sensors. Thus the
master node cannot rely
on the arrival time to
merge data, rather on the time point at which data is sampled. The time instant
at which the data is sampled is integrated as a timestamp to data by the sen-
sor for each sample. This motivates the need for different sensors to have their
internal clocks synchronized.

Definition 1. A sensor Si is given by hTi , Ei , Di i where Ti denotes the actual

time of the sensor, Ei is the energy consumption of the sensor and Di is the
drifting of the sensor time compared to the time in the master node. A sensor
network is given accordingly by hM, (S1 , .., Sn )i where M = hC, Ei is the master
node with C a physical clock and E the energy consumption of the node, and
(S1 , .., Sn ) are sensors.

In fact, Ti is the number of clock ticks since the sensor is powered on. This
variable can be updated through a synchronization operation. The energy con-
sumption Ei is a critical attribute of the sensor as it determines the life time
and impacts how often the sensor can sample and send data to the master node.
As for behavior, the sensors experience four different operation phases: Standby,
Transmit, Receive and Measure. At Standby phase the sensor is inactive and con-
sume a very minimal energy consumption, whereas in the other three phases the
sensor is either sending, receiving or sampling data while consuming proportional
energy to the data throughput. The different phases can run with different time
intervals depending essentially on the sampling frequency and the time synchro-
nization protocol used to align sensors local time to the clock of the master node.
Further details of the sensors behavior and the underlying energy consumption
is provided in Section. 4.

3 Related Work
Time synchronization and energy efficiency for wireless sensor networks have
been studied thoroughly in the literature [19,20,21,22,5,18,23] for different ap-
plications. Beside to improving time synchronization accuracy and energy effi-
 Time Synchronization for Intelligent Sensor Systems 5

ciency, a challenging task is how to balance the trade-off between such conflicting
attributes [9,10].
The authors of [5] proposed a networked control to identify sensors synchro-
nization and estimate potential drifting errors. This might lead to much accurate
synchronization result but on the expense of energy given that a drifting of a
sensor would require a synchronization of the entire network. The larger the
network is, the more frequent synchronization packets are issued. Compared to
that, our protocol calculates the drifting error of sensors individually so that
only the sensors experiencing a drift will issue a sync request.
In [21], Ali et al introduced a time synchronization protocol based on averag-
ing consensus algorithm. The protocol relies on updating the sensors local time
following the expected communication time. This improves time synchronization
accuracy but with an increase in the needed number of communication instances
to estimate the actual communication time, by which the energy consumption
increases.
The authors of [20] proposed a time synchronization protocol for WSNs to
compensate the time discrepancy of sensors. It relies on a broadcast scheme and
timestamping mechanism to achieve low execution time and low network traffic
along with accurate synchronization. However, similarly to [21], performing a
broadcast communication every time a sensor drifts is expensive in terms of
energy and it is even not needed for sensors having clocks well synchronized.
The authors of [19] introduced a protocol, as an integration of two processes,
to reduce the energy consumption of sensors while maintaining high time accu-
racy. Such a protocol relies on the assumption that the master node knows the
complete topology of the network, which makes such a protocol not suitable for
ad-hoc networks due to the lack of flexibility.
The authors of [22] designed a high precision time synchronization based on
common performance clock source. Based on the mutual drift estimation, each
pair of activated sensors fully adjusts clock rate and offset to achieve network-
wide time synchronization without necessarily going through the gateway (mas-
ter node). The protocol considers stochastic communication to model random
packets loss. Given that our protocol does not consider packets loss, this will be
an attractive feature to add in future.

4 An Intelligent Time Synchronization Protocol

The challenge with wireless sensor networks have a single (master) node equipped
with a physical clock is that sensors may request time synchronization from
the master node for each data packet to send [6]. Although this guarantees
an accurate synchronization of sensors to the master node’s clock, it can end
up in draining the sensor batteries much faster. This situation gets even worse
when sensors operate with high sampling frequency. Such a naive synchronization
approach is depicted in Figure 3.
In order to improve the energy efficiency of WSNs while maintaining sen-
sors highly synchronized, we propose a new time synchronization protocol. The
6 J Boudjadar et al.

Fig. 3. Naive time synchronization

protocol dynamics as well as the underlying energy consumption are described

below.

4.1 Time Synchronization Dynamics

One way to reduce the sensors energy consumption is by reducing the size of
the data packets to send. The proposed protocol relies on the assumption that
each sensor is responsible for its own synchronization with the master node.
It first reduces the synchronization request packet by 4 bytes (out of 13 bytes
used in the Naive synchronization) thus enables the sensors to turn on their
transmitters/receivers for a shorter time to communicate with the master node.
Figure 4 depicts the packets structure of the proposed protocol. Given the high
number of data packets to communicate, the longer WSN runs the higher the
energy efficiency our protocol will achieve.
At a second stage, Fig. 4. Data packets for synchronization
the proposed proto-
col reduces the num-
ber of synchroniza-
tion requests/answers
to communicate be-
tween sensors and the
master node. This is
achieved by enabling
each sensor to use
the timestamps (syn-
chronization packets)
 Time Synchronization for Intelligent Sensor Systems 7

from the master node to calculate its own clock drifting. Based on its drifting,
the sensor decides to request new synchronization packets only if it has a consid-
erable drifting 3 . Otherwise, as long as the sensor internal time is synchronized
with the master’s clock no synchronization requests will be issued by the sen-
sor, thus no sync packet will be sent by master node to such a sensor. This will
reduce drastically the number of packets to communicate between sensors and
master node, which means longer Standby phase and higher energy savings for
the sensors.
Following such a be-
havior, the sensor can
Fig. 5. PI controller for time synchronization
be seen as a phase-
locked PI controller [4].
PI controller
Figure 5 depicts the
overall behavior of the Kp
sensor with respect to T_world Correction factor
Error
drifting estimation and
correction. The time in- Ki
terval to request a new
synchronization u(t) by
T_Estimate
the PI controller is cal-
culated as follows (dis-
cretized):
Xt
u(t) = Kp .Di (t) + Ki Di (t) − Di (t − 1)
0

where Kp and Ki are cumulative parameters to use for optimization. The drifting
(error) Di (t) of a sensor at time point t is calculated in turn as the difference
between the timestamp, sent by the master node, together with the the trip
time of the synchronization packet, and the sensor time (number of ticks) Ti (t)
at time point t.
In fact, the error calculation is performed as follows:
(Tworld − [Ti (t)]).F
Di (t) =
[Ti (t)] − [Ti (t − 1)]

where Tworld = C + R/2, C is the physical clock value provided by the master
node and R is the round trip time for a message communication between the
sensor and master node. F is a scaling factor defined as follows: F = C(t)−[Ti (t)].

4.2 Optimization
To improve the time accuracy and energy efficiency of the proposed proto-
col further, we have conducted an optimization process using a genetic algo-
rithm [2]. The optimization amounts at calibrating the PI controller parameters
3
By clock drifting we refer to the deviation of a sensor’s time from the physical clock
of the master node.
8 J Boudjadar et al.

(Kp , Ki , F ), so that optimal starting and result values are identified by which a
lower package count will be used for synchronization, i.e. minimize the amount
of sync packets needed before an actual drifting error occurs. Kp dictates how
much the controller should correct for the proportional error. Ki controls how
much of the previous error to correct for, and F is an integrator starting value
used to track the constant error.

Fig. 6. Sensors drifting without synchronization (ms)

400

200

0
Drifting time

-200

-400

Simulation time
0 100000 200000 300000 400000 500000

The genetic algorithm works by taking a random starting sets of values for
the control parameters and parsing them through a fitness function. In fact, the
fitness function weights the resulting errors for different parameter values. Such
a function is defined as follows:
Wi = |u(t)|.Count(t)
where Count(t) is the number of synchronization packets to send in order to
reach a satisfying synchronization between the sensor time and the master node
clock. The optimization results and sketch code are presented in Sections 5 and
6 respectively.

5 Analysis of the Time accuracy and Energy

This section analyzes the time accuracy and the improvement of the energy
efficiency of our time synchronization protocol.

5.1 Time Accuracy Analysis

Given that both the sensor network and our protocol are modeled using VDM,
we first formally analyze the time accuracy of our protocol against the base-
line delay requirement (500ms) imposed by our industrial partner. Furthermore,
 Time Synchronization for Intelligent Sensor Systems 9

Python simulations are used to analyze the time accuracy improvement through
optimization.
Without using any time synchronization mechanism, we can see that the
drifting error of a sensor can reach up to 400ms (Figure 6). Using combinatorial
testing of VDM [3], we succeeded to prove that the drifting obtained for our
protocol satisfies the baseline requirement. The drifting can either be positive or
negative.

Fig. 7. Sensor drifting using our protocol (ms)

10

-10
Time error

-20

-30

-40

-50

-60
Simulation time
0 100000 200000 300000 400000 500000

In fact, the drifting of our protocol is far below that requirement and does
not exceed 66ms (Figure 7).

Fig. 8. Sensor drifting after optimization

Normalized time error on a single sensor
1.0

0.8
Probability [%]

0.6

0.4

0.2

0.0
-1.00 -0.75 -0.50 -0.25 0.0 0.25 0.50 0.75 1.00
Clock error [time units]
10 J Boudjadar et al.

Using the optimization algorithm on the sensor behavior (modeled as PI con-

troller), the drifting time of the senors has been drastically reduced to converge
to zero with a maximum value of 1ms. Figure 8 depicts the probability distribu-
tion of the sensor drifting obtained through optimization. One can see that over
100 iterations the objective function of our optimization algorithm converges to
zero (Figure 9), which is the optimal value of the objective function.

Fig. 9. Optimization convergence

5.2 Energy Consumption Analysis

The energy consumption of the sensor amounts to the energy consumed through-
out the four behavior phases. The stay at each phase depends on the sampling
frequency and the output of the drifting calculation of the time synchronization
protocol. At Standby phase, the energy consumption EiS (t, t0 ) of a sensor i is in
fact the consumption rate Is accumulated along the stay duration.

EiS (t, t0 ) = Is .(t0 − t)

The next state is when the sensor is transmitting over the radio interface, this
is often a high energy operation. The energy consumption EiT during Transmit
phase is obtained as the energy consumed per packet IT , corresponding to the
energy consumption rate for transmission, times the number of packets obtained
as the time duration on the time to send one packet (Tm ).

t0 − t
EiT (t, t0 ) = It .
Tm
 Time Synchronization for Intelligent Sensor Systems 11

In similar way, the energy consumed during Receive phase EiR is calculated
as follows:
t0 − t
EiT (t, t0 ) = Ir .
Tr
where Ii is the energy consumption for receiving a single packet and Tr is the
time duration to receive the packet. We assume that during Transmit and Receive
phases, the sensor is active exactly for the time duration to send, respectively
receive, packets without waiting time between the different packets. As for Mea-
sure phase, the energy consumption is calculated as the consumption rate per
sample (Im ) times how many samples, obtained by the sampling duration on the
sampling frequency H.

t0 − t
EiM (t, t0 ) = Im .
1/H

Fig. 10. Energy consumption for a naive synchronization

1e7 Total power usage

3.0

2.5
Power [mA]

2.0

1.5

1.0

0.5

0.0 Simulation time

0 50000 100000 150000 200000 250000

Figure 10 depicts a simulation-based analysis of the energy consumption of a

sensor with a naive synchronization, i.e. a synchronization packet is requested for
each data sample. The energy consumption of our protocol, after optimization,
is depicted in Figure 11. One can see that our optimized protocol outperforms
quite well the naive synchronization approach.

6 Implementation and Experimental Results

The proposed time synchronization protocol has first been implemented in Python
for simulation and optimization purposes. Thereafter, in order to test the pro-
posed protocol in an actual setup and integrate it into the ReMoni cloud solu-
tion for remote monitoring 4 , it has been implemented in C language. To enable
comparison, the naive synchronization protocol has also been implemented in
Python.
4
https://www.remoni.com/solutions/
12 J Boudjadar et al.

Fig. 11. Energy consumption of our protocol after optimization

1e7 Total power usage

1.50

1.25

Power [mA]
1.0

0.75

0.50

0.25

0.0 Simulation time

0 100000 200000 300000 400000 500000

A sketch code for the main time synchronization function is depicted in

Listing 1.1. It is important to mention that, even for a large network, our op-
timization function converges to the optimal result after reasonable number of
iterations.

Listing 1.1. Python code for time synchronization

def SyncTime ( s e l f , Ctx , Crx , Trx , timeQa ) :
2 i f s e l f . t s . s t a t e == t i m e s y n c S t a t e s . FIRST BOOTED :
3 s e l f . t s . s t a t e = t i m e s y n c S t a t e s .CONVERIGN
4 s e l f . t s . T r e f = Trx + ( ( Crx − Ctx ) / 2 ) ∗ s e l f . t s .R
5 s e l f . t s . C r e f = Crx
6 else :
7 i f s e l f . t s . e r r == 0 :
8 s e l f . t s . b a d n e s s = abs ( s e l f . t s . e r r )
9 Tnow = s e l f . SyncTimeNow ( Crx )
10 C r t t = Crx − Ctx
11 Tworld = Trx + ( C r t t / 2 ) ∗ s e l f . t s .R
12 dC = ( Crx − s e l f . t s . C r e f )
13 s e l f . t s . e r r = ( Tworld − Tnow) ∗ dC/ (Tnow − s e l f . t s . T r e f )
14 s e l f . t s . i n t e g r a t o r = s e l f . t s . i n t e g r a t o r + (
15 s e l f . ts . err ∗ s e l f . ts . ki )
16 s e l f . t s .R = s e l f . t s . i n t e g r a t o r + s e l f . t s . kp ∗ s e l f . t s . e r r
17 s e l f . t s . T r e f = Tnow
18 s e l f . t s . C r e f = Crx
19 s e l f . t s . b a d n e s s = 0 . 9 ∗ s e l f . t s . b a d n e s s + (
20 0 . 1 ∗ abs ( s e l f . t s . e r r ) )
21 i f s e l f . t s . b a d n e s s < SYNC LEVEL :
22 s e l f . t s . s t a t e = t i m e s y n c S t a t e s .SYNCRONISED
23 e l s e :
24 s e l f . t s . s t a t e = t i m e s y n c S t a t e s .CONVERIGN
 Time Synchronization for Intelligent Sensor Systems 13

7 Conclusion

This paper proposed an intelligent energy-efficient time synchronization proto-

col for wireless (clamp-on) sensor networks. Sensors operate on battery and have
very limited hardware. The mission of the protocol is to maintain sensors time
synchronized with the clock of the master node so that the data sampling from
sensors is aligned. This results in obtaining accurate real-time state of the mon-
itored environment. The proposed protocol requests a synchronization packet
from the master node only if a sensor time is considerably drifting from the
master node clock.
The protocol has been formally specified in VDM to verify that the maximum
drifting does not violate the baseline requirement. Furthermore, an optimization
of the protocol has been conducted using a genetic algorithm to improve further
the time accuracy. The proposed protocol has been implemented and integrated
into the ReMoni cloud solution for remote monitoring. The analysis results of
the proposed protocol shows that it outperforms the state of the art protocols
with respect to time accuracy while it maintains a high energy efficiency.
A future work will be to study the trade-off between the time accuracy and
the energy efficiency and finding the optimal sampling frequency to reduce the
energy consumption much further.

References
1. European Commission.: NZEB Buildings. https://ec.europa.eu/energy/topics/
energy-efficiency/energy-efficient-buildings/nearly-zero-energy-buildings en
2. Jayachitra, A. and Vinodha, R.: Genetic algorithm based PID controller tuning
approach for continuous stirred tank reactor. Advances in Artificial Intelligence,
2014
3. T. Kulik, P. W. V. Tran-Jørgensen, J. Boudjadar and C. Schultz: ”A Framework
for Threat-Driven Cyber Security Verification of IoT Systems.” 2018 IEEE Inter-
national Conference on Software Testing, Verification and Validation Workshops
ICSTW, pp. 89-97.
4. Crowe, J. and Johnson, M.: Phase-Locked Loop Methods and PID Control. Springer
Publisher, pp. 259-296. 2005
5. Wang Ting, Cai Chun-yang, Guo Di, Tang Xiao-ming, Wang Heng.: Clock Synchro-
nization in Wireless Sensor Networks: A New Model and Analysis Approach Based
on Networked Control Perspective. Mathematical Problems in Engineering, vol 19.
2014
6. Ranganathan Prakash, Kendall Nygard.: Time Synchronization in Wireless Sensor
Networks: A Survey. International Journal of UbiComp 1(2), 2010
7. M. Banaei, M. Rafiei, J. Boudjadar and M. Khooban: ”A Comparative Analysis
of Optimal Operation Scenarios in Hybrid Emission-Free Ferry Ships.” In IEEE
Transactions on Transportation Electrification, vol. 6, no. 1, pp. 318-333, 2020.
8. Rastislav Ingeli, Miroslav Cekon.: Analysis of Energy Consumption in Building with
NZEB Concept. In Applied Mechanics and Materials 824:347-354, 2015
9. M. Rafiei, M. Khooban, M. A. Igder and J. Boudjadar: ”A Novel Approach to
Overcome the Limitations of Reliability Centered Maintenance Implementation on
14 J Boudjadar et al.

the Smart Grid Distance Protection System.” In IEEE Transactions on Circuits and
Systems II: Express Briefs, vol. 67, no. 2, pp. 320-324, 2020.
10. M. Gheisarnejad, J. Boudjadar and M. Khooban: ”A New Adaptive Type-II Fuzzy-
Based Deep Reinforcement Learning Control: Fuel Cell Air-Feed Sensors Control.”
In IEEE Sensors Journal, vol. 19, no. 20, pp. 9081-9089, 2019.
11. Anna Magrini, Giorgia Lentini, Sara Cuman, Alberto Bodrato, Ludovica Marenco.:
From nearly zero energy buildings (NZEB) to positive energy buildings (PEB): The
next challenge - The most recent European trends with some notes on the energy
analysis of a forerunner PEB example. In Developments in the Built Environment,
Volume 3, 2020
12. Anna Magrini, Giorgia Lentini.: NZEB Analyses by Means of Dynamic Simulation
and Experimental Monitoring in Mediterranean Climate. In Energies Journal, vol
13. 2020
13. Cao, X.; Dai, X.; Liu, J.: Building energy-consumption status worldwide and the
state-of-the-art technologies for zero-energy buildings during the past decade. En-
ergy Build. 2016, 128, 198213
14. Van de Putte, S.; Bracke,W.; Delghust, M.; Steeman, M.; Janssens, A. Compar-
ison of the actual and theoretical energy use in nZEB renovations of multi-family
buildings using in situ monitoring. E3S Web Conf. 2020
15. Anwar Ul Haq, Hans-Arno Jacobsen.: Prospects of Appliance-Level Load Monitor-
ing in Off-the-Shelf Energy Monitors: A Technical Review. In Energies Journal, vol
11, 2018
16. ReMoni.: Clamp-on IoT Sensors https://www.remoni.com/products2/
product-overview/
17. Elson, Jeremy Girod, Lewis Estrin, Deborah.: Fine-Grained Network Time Syn-
chronization using Reference Broadcasts. ACM SIGOPS Operating Systems Review.
Vol 36, 2002.
18. Ganeriwal, Saurabh Kumar, Ram Srivastava, Mani.: Timing sync Protocol for
Sensor Networks. SenSys03: Proceedings of the First International Conference on
Embedded Networked Sensor Systems.
19. Li, Fen He, Guanghui Wang, Xu.: An Improved Hybrid Time Synchronization
Approach in Wireless Sensor Networks for Smart Grid Application. In HPCC con-
ference 2015.
20. Ki-Hyeon Kim, Won-Kee Hong, Hiecheol Kim.: Low Cost Time Synchronization
Protocol for Wireless Sensor Network. IEICE Transactions on Communications 92-
B(4):1137-1143. 2009.
21. Al-Shaikhi, Ali. (2018). Accuracy-Enhanced Time Synchronization Method for
WSNs Using Average Consensus Control. Proceedings of International Multi-
Conference on Systems, Signals and Devices. 2018.
22. Nan Xiong, Minrui Fei, Taicheng Yang, Yu-Chu Tian.: Randomized and Efficient
Time Synchronization in Dynamic Wireless Sensor Networks: A Gossip-Consensus-
Based Approach. Complexity Journal, vol. 2018
23. Upadhyay, D., Dubey, A.K. Santhi Thilagam, P. Time synchronization problem of
wireless sensor network using maximum probability theory. Intl J Syst Assur Eng
Manag 9, 517524 (2018).