SICE Annual Conference 2007
Sept. 17-20, 2007, Kagawa University, Japan
Time synchronized wireless sensor network for vibration measurement
Yutaka Uchimura1 , Tadashi Nasu 2 and Motoichi Takahashi2
1
Department of Mechanical Engineering, Shibaura Institute of Technology, Tokyo Japan
(E-mail: uchimura@sic.shibaura-it.ac.jp)
2
Kajima Corporation, Tokyo, Japan
(E-mail: {takahamo, tadashi-nasu}@kajima.com)
Abstract: Network based wireless sensing has become an important area of research and various new applications for
remote sensing are expected to emerge. One of the promising applications is structural health monitoring of building
or civil engineering structure and it often requires vibration measurement. For the vibration measurement via wireless
network, time synchronization is indispensable. In this paper, we introduce a newly developed time synchronized wireless
sensor network system. The system employs IEEE 802.11 standard based TSF counter and sends the measured data with
the counter value. It enables consistency on common clock among different wireless nodes. We describe the accuracy
evaluation by simulation studies when the size of nodes increased. The hardware and software specifications of the
developed wireless sensing system are shown. The experiments were conducted in a three-story reinforced concrete
building and results showed the system performs more than sufficiently.
Keywords: Sensor network, Vibration measurement, Wireless network, Time synchronization
1. INTRODUCTION
With the rapid progress of wireless communication
and computer network technology, wireless network
based sensing has become an important area of research
and various new applications for remote sensing are
expected to emerge. Applications include monitoring
physical or environmental conditions such as temperature, sound, motion or pressure without laying hundreds
of meters of signal wires that are prone to breaking.
One particularly promising application is monitoring the
structural health of buildings and civil engineering structures [1][2]. Since measuring objects are usually huge
and wiring very long cables is required, wireless data
transmission is highly beneficial.
Structure health monitoring often requires measuring
vibration data such as acceleration and velocity. Measured data are analyzed to obtain the resonance frequency, damping ratio and spectrum response. measurement For wireless vibration measurements, time synchronization is important because the vibration data are simultaneously measured at multipoint sensor nodes and
are often transmitted via multi-hop relayed by wireless
devices. As a result, when the data arrive at a measuring station, they are contaminated with an unknown (random) time delay. Even if each sensor node sends data
exactly at the same moment, the arrival time of the data
does not match. Thus received data needs to be adjusted
so as to maintain time consistency on a common time
axis. Because when the data are used for modal analysis, a time difference may be misunderstood as a phase
shift. To maintain precise time consistency among wireless nodes, time synchronization is indispensable. In this
paper we describe a new time synchronized wireless sensor network system for vibration measurement.
Time synchronization of network has been a matter of
concern, therefore several technologies such as GPS, radio ranging have been used to provide global synchro-
nization in networks. GPS based synchronization offers
very precise synchronization, however it is not always
available indoor place. In wired network, the Network
Time Protocol (NTP) has been developed that has kept
the Internet’s clocks ticking in phase [3].
Time synchronization is also important to determine
radioactive period that directly affects battery life. There
have been many studies on time synchronization which
aimed for conserving battery energy of network nodes.
RT-Link and B-MAC [5] (Berkeley MAC) focused on the
life of batteries. RT-Link[4] uses MAC (media access
control) based on slotted ALOHA and it employs independent AM carrier-current radio device for indoor time
synchronization. B-MAC[5] (Berkeley MAC) supports
carrier sense multiple access (CSMA) with low power
listening (LPL) where each node periodically wakes up
after a sample interval and checks the channel for activity
for a short duration of 2.5ms. The main concern of these
methods is battery life.
Meanwhile, our focus in time synchronization is not
only the battery life but also maintaining time consistency among all sensor nodes. When a sensor network is
used for vibration measurement and modal analysis, all
the data measured at different wireless sensor nodes need
to be aligned with a common clock, which need not be
the real world clock. Because the difference of the clock
among nodes can be misunderstood as a phase shift of
vibration, which seriously affects the modal analysis of
building vibration. Therefore, the purpose of time synchronization in this paper is essentially different from the
studies that focus on battery life.
The developed system employs IEEE 802.11[6] standards based wireless devices. These devices are the
de facto standard of wireless networks and so are readily available on the market with industrial-level reliability. The modulation uses DS-SS (Direct Sequence Spectrum Spread) and OFDM (Orthogonal Frequency Divi-
- 2940 -
PR0001/07/0000-2940 ¥400 © 2007 SICE
sion Multiplexing) which offer robustness against phasing and noise. Those are advantages against RT-Link
and Berkeley Mote which use noise susceptible wireless
modulation. IEEE 802.11 standard devices have a timing
synchronization function (TSF) by default. We propose
to utilize the function for synchronization among nodes,
sampling data interval and time stamping of measured
data.
In the following section, we describe the adverse effects on vibration sensing caused by a time delay in wireless network and the reason why time synchronization is
needed. Section III describes simulation studies when the
size of wireless nodes becomes large. In Section IV, we
describe a newly developed wireless sensing system and
also describe its hardware and software components. In
Section V, experimental evaluations of time synchronization precision are presented.
2. TIME DELAY ON WIRELESS
NETWORK AND TIME
SYNCHRONIZATION
Data transmission via a wireless network requires certain time periods from the moment when the transmitter
sends the data to the moment when the data arrives at
the receiver. In vibration sensing applications, sensors
usually need to be placed at several points on different
floors in the building. The measurement station, which
is composed of a host computer system, is placed at a
certain location in the building. In such cases, especially
in a high-rise building, the wireless radio waves of each
sensor node do not directly reach the station. Thus, the
network path needs to be multi-hop and the packets are
transmitted via the relay of the wireless sensor nodes on
the path.
This multi-hop relay introduces a time delay, which
corresponds to the duration from the moment when a
node sends a packet to the moment when the station
(host PC) receives the packet. The CSMA/CA (Carrier
Sense Multiple Access/Collision Avoidance) media access method, which is used in IEEE 802.11 standards, is
another cause of time delay, because before the packet
sending, it always needs to wait for a random back-off
time and also needs to send an ACK packet then wait for
a CTS packet before sending the measured data packet.
Moreover, the packet does not always go through and
may be lost, in which case it needs to be re-transmitted,
resulting in a considerable time delay.
Therefore, even if all the nodes intend to send packets
at exactly the same time, the arrival time may be different.
This fact results serious problem when measured data are
used for modal analysis. Fig. 1 (a) shows the motion of
a bending beam with three measured points (marked as
circles). As time proceeds, the beam moves from P1 to
P2 and P3 . Suppose the transmission of data on the tip
is delayed, it virtually seems that the tip stays at P2 position even though other points moves forward (see Fig.
1 (b)). So, we somehow need to calculate back the measured time and maintain consistency of received data on
2
(a )
2
2
2
!
2
2
!
(b )
Fig. 1 Adverse effect on vibration measurement due to
transmission delay
the common time line.
To resolve this issue, we propose to send the vibration
data together with the time stamp of the moment when the
data is measured. After the data are received, they can be
rearranged along the desired time line based on the time
stamp. This procedure is valid provided that all clocks of
the nodes are matched. However, accuracy of the quartzcrystal oscillator based clock is affected by many factors
such as temperature and change of current or voltage [7].
Frequency stability of oscillators used for PCs is
mostly around 10−4 (100 ppm). Imprecision of 100 ppm
corresponds to a 1 millisecond (1 msec) error in 10 seconds, or an 18-degree phase shift for 100Hz vibration.
Even though the resonant frequency of a building is low
(typically less than 10 Hz), a 1 msec difference between
the fastest clock and the slowest clock is the maximum
tolerance, so we set the target performance of time synchronization to be less than 1 msec.
2.1 Timing synchronization function
As previously mentioned, we use IEEE 802.11 standards based wireless device and utilize the timing synchronization function (TSF). IEEE 802.11 supports a few
wireless configurations such as infrastructure mode or ad
hoc mode and each mode uses different time synchronization policy. In the infrastructure mode, the access
point periodically generates beacons which contain the
current clock value, which is then copied by the all nodes.
In the ad hoc case (multi-hop case), the stations share the
beacon generation responsibility among themselves in a
distributed fashion. The developed system uses the ad
hoc mode for network configuration. Each node in the
ad hoc mode maintains its own TSF timer, and each time
the beacon interval time (default 100msec) elapses, the
node will wait a random delay then send a beacon. A
synchronization beacon contains a message header and
timestamp which is a 64 bit field containing the TSF timer
and rolls over after it reaches 264 − 1.
To ensure that the timestamp value in the beacon accurately represents the internal timer, the timestamp field
- 2941 -
M o n ito r
th e v a lu e
T S F c o u n te r
S a m p lin g
T im e r
T h re a d
E x te n t o f R a d io W a v e
a q u ire o n e s h o t v a lu e
A /D C o n v e rte r
Fig. 2 Block diagram of A/D conversion maintaining
the constant sampling time
is not filled in until after the CSMA deferral. This is
very important because as previously mentioned, CSMA
is one of the main reasons of the unknown time delay.
This procedure is done within MAC layer and it is implemented at the hardware level.
If a beacon arrives before the random delay timer has
expired, the station cancels the pending beacon transmission and the remaining random delay. Upon receiving a
beacon, a station sets its TSF timer to the timestamp of
the beacon if the value of the timestamp is later than the
station ʟs TSF timer. (It is important to note that clocks
only move forward and never backward.)
2.2 Measurement interval (sampling time)
Once the vibration measurement starts, CPU of each
node commands the embedded A/D converter to acquire a
one-shot value from the vibration sensor. The data acquisition is executed at a certain interval. This constant interval corresponds to the sampling time and it ticks along the
TSF timer. As shown in Fig. 2, a software task (sampling
timer thread) implemented in each node keeps monitoring the value of TSF timer. When the TSF timer elapses
and comes up to the next sampling time, the A/D converter acquires a one-shot sensor value. As long as the
TSF timers of all wireless nodes are synchronized, the
A/D samplings of all sensor nodes are processed at precisely the same moment.
3. SCALABILITY ANALYSIS BY
SIMULATION
IEEE 802.11 standards based TSF partially depends
on a stochastic manner. Possibility of chance to send a
beacon is equal, thus it may happen that the node whose
clock is fastest cannot be given the chance to propagate
one’s clock. This could increasingly occur as the size
of IBSS becomes larger, where IBSS is the independent
basic service set (in other words, the group of ad hoc
nodes which have the same communication ID), and it
affects the precision of clock synchronization of multinode. Therefore, we examined the stochastic effect on
synchronization precision by the Monte Carlo simulation
method. We made simulations under the following conditions:
1. Each node has a TSF clock, whose frequency stability is modeled as a Gaussian distribution.
2. The standard deviation σ is 33.3ppm (parts per million), i.e. 3σ, which contains 99.73% of the clocks,
W ire ls s N o d e
Fig. 3 Configuration of wireless nodes whose radio wave
can reach to all nodes
W ire ls s N o d e
Fig. 4 Configuration of wireless nodes whose radio wave
can reach to only the adjacent nodes
corresponds to 100ppm.
3. The beacon interval is 100 msec (10 Hz).
The simulations are run for two cases of node arrangements.
• Case 1: All nodes are within the range that radio
waves can reach, i.e. each node can send a beacon
directly to any other node. See Fig. 3.
• Case 2: All nodes are arranged in daisy chain configuration so that radio waves can reach two adjacent
nodes only (right and left nodes). Fig. 4 illustrates
the allocation of nodes. End-to-end communication
is possible only through multi-hop relay.
In both cases, each simulation trials correspond to 90
minutes elapse of time and we made 100 times of simulation trials by randomly selecting clock speed of attending nodes. During the simulation, we recorded the difference between the fastest clock and the slowest clock
every 100msec. Notice that the slowest clock does not
indicate a specified clock. The clock, which shows the
most delayed time at a certain moment, is the slowest
clock. The slowest clock may receive a beacon and adjust
its TSF timer; as a result, other clock can be the slowest.
Fig. 5 and Fig. 6 show histogram of case 1 simulation. Fig. 5 is the case when the total number of nodes
is 20 and Fig. 6 is the case of 100 nodes. As shown
in the histograms, the precision of time synchronization
gets worse as number of nodes becomes large. In 20-node
case, 99.9% of data is within 80μsec. In 100-node case,
it is within 473μsec. The median value of 20-node case
is 12μsec and that of 100 case is 38μsec.
Fig 7, 8 and 9 show the histogram of case 2 for the
daisy chain configuration cases. Fig. 7 is the case of 5
nodes, Fig. 8 is that of 20 nodes and Fig. 9 shows 100-
- 2942 -
x 10
5
2.5
2
2
1.5
1.5
Frequency
Frequency
2.5
1
1
0.5
0.5
0
0
20
40
60
80
100
120
Time (micro second)
140
160
5
12
2.5
10
2
8
1.5
4
0.5
2
100
200
300
400
Time (micro second)
500
40
60
80
Time (micro second)
100
120
x 10
0
0
600
4
6
1
0
0
20
Fig. 7 Histogram of simulation results in case 2 of 5
nodes
Frequency
Frequency
x 10
5
0
0
180
Fig. 5 Histogram of simulation results in case 1 of 20
nodes
3
x 10
50
100
150
Time (micro second)
200
250
Fig. 6 Histogram of simulation results in case 1 of 100
nodes
Fig. 8 Histogram of simulation results in case 2 of 20
nodes
node case. The median value in 5-node case is 11μsec,
66μsec in 20-node case, and 385μsec in 100-node case.
In 5-node case, 99.9% of the data maintain less than
64μsec, 174μsec in 20-node, and 909μsec in 100-node
case.
In the actual measuring environment in a building, the
configuration of wireless device allocation is likely to be
a mixture of case 1 and 2 as illustrated in Fig. 10. Several
wireless nodes are located on a floor and a few of them
act as relaying packets to the nodes on a different floor.
The target synchronization accuracy is to maintain within
1msec. Form the simulation results, synchronization precision is maintained even when the total number of nodes
reaches up to 100 nodes.
Table 1 and their block diagram is shown in Fig. 12. The
hardware of the node is composed of CPU, DRAM and
PC104 extension bus. The A/D converter is connected
via the PC104 bus. The A/D converter has 16CH single end input, and 6CH of them are connected to BNC
jack. In this system, sensors are not embedded within it.
Because the vibration sensing uses various sensors such
as accelerometers or velocity sensors. It depends on the
structure of the buildings. Additionally, sensors are usually supplied as complete products not component parts,
thus separation of sensors and wireless nodes are reasonable and make it more versatile.
The operating system (OS) is RT-Linux, which is an
extension of Linux to a real-time operating system. Software including OS is implemented on compact flash disk
and the disk can also store measured data. As a wireless device, we used IEEE 802.11a/b/g standard wireless
device to which an external antenna can be connected.
The multi-hop routing is autonomously configured by
applying OLSRʢOptimized Link State Routingʣprotocol [8]. The protocol is pro-active, table driven and
utilizes a technique called multipoint relaying for message flooding. OLSR also implements a popular optional
link quality extension. Thanks to its autonomous routing, even if an operator carries a wireless node and walk
4. DEVELOPMENT OF WIRELESS
MEASUREMENT SYSTEM
We developed wireless embedded measurement system which can be applicable for vibration sensing. The
system is composed of host PC and several identical wireless nodes. Any PC which has an IEEE 802.11 standard
wireless device can be the host PC. Fig. 11 is a photo
of a wireless node. Size of the node is 120 mm width,
70 mm depth and 50 mm height. Main components and
their functions of the wireless node are briefly shown in
- 2943 -
14
x 10
4
12
Frequency
10
8
6
4
2
200
400
600
800
Time (micro second)
1000
1200
Fig. 11 Wireless node with external antenna
Fig. 9 Histogram of simulation results in case 1 of 100
nodes
C P U
G e o d e
G X 1
W ire le s s n o d e
S O -D IM M
2 5 6 M B
S D R A M
R e la y
H o st P C
N S
5 5 3 0 A
A /D
C o n v e rte r
B N C
C o n n e c to rs
R e la y
C o m p a c t
F la s h
ID E
P C 1 0 4
0
0
P C M C IA
W ire le s s L A N
C A R D
A R 5 2 1 2 C h ip
E x te rn a l A n te n n a
Fig. 12 Block diagram of hardware components
Fig. 10 Measuring configuration in a building as mixture
of case 1 and 2
around the building, the multi-hop relaying maintains the
end-to-end communication path.
5. EXPERIMENTAL EVALUATION OF
SYNCHRONIZATION PRECISION
For the purpose of evaluating TSF based synchronization, we made experiments in a three-story reinforced
concrete building. The allocation of wireless node is depict in Fig. 13. Fig. 14 shows the wireless node in the
middle place. We used three wireless nodes which were
arranged to make two-hop transmission, i.e. the end-to-
end communication packets are relayed via the middle
node.
Wired signal lines are connected to all nodes
for the reference purpose and rectangular wave voltage is
supplied to generate hardware interrupt. Every moment
when the voltage of square wave steps up, an interrupt occurs exactly at the same moment in three wireless nodes
and TSF timer count is recorded in the interrupt handler
routine.
By comparing recorded TSF timer of three nodes,
we can evaluate the synchronization precision. The frequency of square wave was set to 10Hz, i.e. the TSF
timer value is recorded every 100msec. A measurement
was continued for 15 minutes thus total number of TSF
record sets was 9,000. After the measurement, the difference between the fastest TSF count and the slowest TSF
Table 1 Hardware Specifications of wirelss node
CPU
Geode GX1 300MHz
Chipset
NS Geode 5530A chipset
RAM
144 Pin SODIMM 256MB
External Storage
Compact Flash
Extention Bus
PC104
WLAN Chip
Atheros AR5212
A/D Converter
16CH, 16bit Resolution
Human Interface
LCD and input switches
W ire le s s N o d e
(M id d le n o d e )
W ire le s s
N o d e
W ire le s s
N o d e
3 F
W ire d L in e
(u se d fo r re fe re n c e
p u rp o s e o n ly )
2 F
1 F
Fig. 13 The allocation of wireless node
- 2944 -
ware and software specifications of a developed wireless
sensing system and experimental evaluations, which were
conducted in a three-story reinforced concrete building
showed good performance.
ACKNOWLEDGMENT
This research had been conducted while the first author belonged to Kajima Corporation.
REFERENCES
Fig. 14 The wireless node placed in the middle
7
x 10
4
6
Frequency
5
4
3
2
1
0
0
Fig.
50
100
Time (micro second)
150
200
15 Histogram of experiment with three wireless
nodes
count at each recorded set was calculated. We made 15minute measurements totally 15 times, thus total number
of samples were 135,000.
Fig. 15 shows histograms of the difference of TSF
count, which corresponds to synchronization precision.
As shown the figure, the most frequent value is around
25 micro sec. It supposed to be smaller by the simulation analysis, however the simulation analysis does not
take account for any processing time for synchronization. Therefore, this must be more realistic result when
implementing on a real hardware. Nevertheless, the result shows sufficiently high precise synchronization and
it maintained much better performance than targeted 1
msec precision.
[1] M. Ruiz-Sandoval, B. F. Spencer Jr. and N. Kurata,
”Development of a high sensitivity accelerometer for
the Mica platform”, Proc of the 4th Int. Workshop on
Structural Health Monitoring, 2003, pp. 1027-1034.
[2] J. P. Lynch et.al., ʠ Post-seismic Damage Assessment of Steel Structures Instrumented with Selfinterrogating Wireless Sensors, ʡProc. of the 8th National Conference on Earthquake Engineering, 2006,
pp. 18 - 22.
[3] D.L. Mills,”Internet Time Synchronization: The Network Time Protocol”,, Global States and Time in
Distributed Systems, IEEE Computer Society Press,
1994.
[4] A. Rowe, R. Mangharam and R. Rajkumar, ”RTLink: A Time-Synchronized Link Protocol for Energy Constrained Multi-hop Wireless Networks ”,
Int. Conf. on Sensors, Mesh and Ad Hoc Communications and Networks Reston, 2006.
[5] J. Polastre, J. Hill, and D. Culler, ”Versatile Low
Power Media Access for Wireless Sensor Networks”,
Proc. of ACM SenSys, 2004, pp. 95 - 107.
[6] ANSI/IEEE Std 802.11, 1999 Edition,
”http://standards.ieee.org/getieee802/download/802.111999.pdf”
[7] M.D.Lemmon, J.Ganguly, L.Xia,”Model-based
clock synchronization in networks with drifting
clocks”, Proc. of 2000 Pacific Rim Int. Symposium
on Dependable Computing, 2000, pp. 177-184.
[8] Engelstad, P.E. Tonnesen, A. Hafslund, A. Egeland,
G. ,”Internet connectivity for multi-homed proactive
ad hoc networks”, Communications, 2004 IEEE International Conference on , 20-24 June 2004, pp.
4050-4056.
6. CONCLUSION
For the vibration measurement via wireless network,
time synchronization is indispensable. In this paper, we
proposed a new time synchronized wireless sensor network system. The system employed IEEE 802.11 standard based TSF counter and sent the measured data with
the counter value. It ensured consistency on the common clock among different wireless nodes. The accuracy when the size of nodes increased was evaluated by
simulation studies and the result validated that accuracy
maintained within 1 msec. We also described the hard-
- 2945 -