KonTest: A Wireless Sensor Network

Testbed at Vrije Universiteit Amsterdam

Konrad Iwanicki,

Albana Gaba,

and Maarten van Steen

Vrije Universiteit, Amsterdam, The Netherlands

Development Laboratories (DevLab), Eindhoven, The Netherlands

{iwanicki, agaba, steen}@few.vu.nl
Technical Report IR-CS-045
Vrije Universiteit Amsterdam, August 2008
ABSTRACT
We introduce our 60-node indoor wireless sensor network testbed, distributed among six ofce
rooms. We outline the hardware architecture of the testbed and discuss the basic properties of
the internode connectivity graph. In particular, we show that a node in the testbed has from 8
to 31 high-quality neighbors and that the network diameter is between 4 and 5 for the lowest
transmission power. The testbed is currently being used as the main evaluation platform for novel
wireless sensor network protocols and systems.
Faculty of Sciences
Vrije Universiteit Amsterdam
De Boelelaan 1081A
1081 HV Amsterdam
The Netherlands
2 K. Iwanicki et al.
Contents
1 Introduction 3
2 Hardware and Organization 3
3 Basic Network Properties 3
3.1 Experimental Setup . . . . . . . . . . . . . 4
3.2 Experimental Results . . . . . . . . . . . . 4
4 Conclusions and Future Work 6
References 7
A Node to Room Mapping 7
KonTest: A Wireless Sensor Network Testbed at Vrije Universiteit Amsterdam 3
1 Introduction
Composed of thousands of tiny low-power embedded de-
vices that collaboratively monitor the environment, guide
vehicles, and predict potential faults in buildings, bridges,
roads, and rails, wireless sensor networks (WSNs) are an
important tier in the IT ecosystem [1]. They have emerged
to realize the vision of an embedded Internet, in which
networks formed by wirelessly interconnected computing
devices provide detailed instrumentation over many points
in large physical spaces. Such instrumentation constitutes a
powerful tool that can potentially transform whole elds of
science, engineering, and manufacturing [1].
However, due to inherent limitations of wireless sensor
nodes and their interactions with the surrounding environ-
ment, sound evaluation of systems and protocols devised for
WSNs is challenging. Real-world deployments to date have
made it evident that in WSNs the divergence between prac-
tice and theory is even more pronounced than in other large-
scale distributed systems [2, 3, 4, 5, 6]. In particular, low-
power wireless communication is extremely unpredictable
and, despite signicant research efforts, no models that
accurately capture all its intricacies exist [7, 8, 9, 10, 11].
Likewise, it is extremely difcult to accurately model the
lifetime of a WSN-based system as the power consumption
of a node depends on a number of dynamically changing
factors, such as the number of cycles burned by the CPU,
the current power mode the CPU is in, the activity on the
radio channel, and the remaining battery capacity, to name
a few. Such unpredictability and complexity severely limit
the accuracy obtained by evaluating a WSN-based system
designs analytically or with simulations. Therefore, sound
evaluation that predicts the real-world system behavior rela-
tively well should be performed with a real implementation
running on real hardware.
In this technical report, we present a 60-node indoor
testbed we have deployed for conducting such evaluations.
Rather than comparing our testbed with the existing ones,
we outline the architecture and general properties of the
testbed. The objective of this report is thus describing our
main experimental platform for WSN-oriented research.
The rest of the paper is structured as follows. First, in
Section 2, we outline the hardware architecture and the
organization of our testbed. Then, in Section 3, we show
basic network properties, which we obtained through ex-
periments. Finally, in Section 4 we conclude and discuss
possible future work.
2 Hardware and Organization
The main hardware components constituting our testbed are
presented in Table 1. The testbed includes 60 TelosB-class
nodes [12] (see also Fig. 1 and Table 2). Ten of the nodes
have been purchased at Moteiv Corporation,
1
and each of
1
http://www.moteiv.com/
Component Quantity
TelosB (Moteiv) w/ sensors 10
TelosB (Crossbow) w/ sensors 20
TelosB (Crossbow) w/o sensors 30
USB hubs 16
PCs 6
Table 1. The main hardware components of our testbed.
Fig. 1. A TelosB node (from Moteiv) with the sensor suite.
them contains a full sensor suite. The remaining fty come
from Crossbow Technology Inc.:
2
twenty of them are with
and thirty are without the sensor suite.
The nodes are located on the fourth oor of the Faculty
of Sciences of Vrije Universiteit Amsterdam. They are dis-
persed among six rooms as depicted in Fig. 2. Such a distri-
bution impacts wireless internode connectivity, as discussed
in the next section, thereby providing a realistic network
topology. In addition, it allows for collecting environmental
data from different parts of the building: southern, eastern,
and western.
To avoid troublesome battery-based power supply and to
enable low-overhead node retasking and statistic reporting,
each of the nodes is connected to a PC using a USB
network. This network consists of a number of cables and
hubs connected to six Pentium III PCs, one PC per room.
The wired USB network is used only as a power supply
and a reliable transport backbone for protocol statistics and
control commands. The protocols evaluated on our testbed,
in contrast, use the standard wireless communication. In
this way the interference between statistic gathering and
protocol operation is minimized.
3 Basic Network Properties
To obtain basic connectivity properties of our network, we
have written and deployed a simple TinyOS 2.0 application.
2
http://www.xbow.com/
4 K. Iwanicki et al.
Fig. 2. The node placement in our testbed. The exact mapping of nodes to rooms is presented in Appendix A
3.1 Experimental Setup
The test application works in periods. In the default con-
guration, each period lasts 30 seconds and the clocks of
the nodes are not synchronized. In every period, each node
broadcasts a heartbeat message at a uniformly random time
moment within the period. A heartbeat message of a node
consists of the nodes identier (2 bytes), a sequence num-
ber (4 bytes), and a Hello world from Konrad Iwanicki!
string (34 bytes including the terminating zero).
Nodes that hear the message transmission record the fol-
lowing metadata: the local time, the senders identier and
sequence number, and the information about the received
signal for the message, as provided by the CC2420 radio
chip in the form of the received signal strength indicator
(RSSI) and the link quality indicator (LQI). The recorded
message metadata are placed in a metadata queue.
At the end of every period, all the records from a nodes
metadata queue are transmitted over the USB network to
the nodes corresponding PC, where they are logged by a
Java frontend of our application. Therefore, ultimately the
log of each node contains ne-grained data specifying all
the messages the node has received and the signal quality
when receiving these messages. By using such data, we are
able to accurately assess wireless internode connectivity.
We have conducted four deployments of the application,
each lasting at least 24 hours. In this way, we could measure
connectivity during busy daytime, when we expected wire-
less interference caused by student and employee laptops,
as well as during quiet nights, when the building was
deserted (surprisingly, there was virtually no difference).
Each deployment varied in the node transmission power,
which was set globally for all nodes. We used the fol-
lowing transmission power settings: 25 dBm, 15 dBm,
5 dBm, and 0 dBm. This allowed us to investigate the
impact of the transmission range on the network density and
diameter. Since during the experiments the 5 nodes in room
P4.46 were not deployed, we used only the remaining 55
nodes in other rooms.
3.2 Experimental Results
For each pair of nodes, in each direction, we have ana-
lyzed RSSI, LQI, and the packet loss rate (PLR), which
was obtained by examining the sequence numbers of the
received messages. In this way, considering for instance the
average RSSI, we obtained a connectivity matrix in which
a cell in row i and column j represents the average RSSI
value for messages received by node i from node j. Such
sample matrices for the transmission power of 25 dBm
are depicted in Fig. 3, Fig. 4, and Fig. 5.
These gures show inherent clustering between the
nodes. Nodes in the same room are likely to be connected
with a high-quality link (high RSSI and LQI values, and a
low PLR value). There are obviously some exceptions, such
as nodes 02 and 03 or nodes 48 and 45. In addition, some
nodes in every room can communicate with some nodes in
other rooms, which ultimately results in a single connected
network. Finally, all three link quality metrics are strongly
correlated, again with an exception of a few links, which
KonTest: A Wireless Sensor Network Testbed at Vrije Universiteit Amsterdam 5
Metric Value Remarks
MODULE
Processor Performance 16-bit RISC
Program Flash Memory 48K bytes
Measurement Serial Flash 1024K bytes
RAM 10K bytes
Conguration EEPROM 16K bytes
Serial Communications UART 0-3V transmission levels
Analog to Digital Converter 12 bit ADC 8 channels, 0-3V input
Digital to Analog Converter 12 bit DAC 2 ports
Other Interfaces Digital I/O,I2C,SPI
Current Draw 1.8 mA Active mode
5.1 A Sleep mode
RF TRANSCEIVER
Frequency band

2400-2483.5 MHz ISM band

Transmit (TX) data rate 250 kbps
RF power -25 dBm to 0 dBm
Receive Sensitivity -90/-94 dBm minimal/typical
Outdoor Range 75 m to 100 m Inverted-F antenna
Indoor Range 20 m to 30 m Inverted-F antenna
Current Draw 23 mA Receive/Listen
21 mA Transmit (at 0 dBm)
1 A Off
SENSORS (Optional)
Visible Light Sensor Range 320 nm to 730 nm Hamamatsu S1087
Visible to IR Sensor Range 320 nm to 1100nm Hamamatsu S1087-01
Humidity Sensor Range 0-100% RH Sensirion SHT11
Resolution 0.03% RH
Accuracy 3.5% RH Absolute RH
Temperature Sensor Range -40

C to 123.8

C Sensirion SHT11
Resolution 0.01

C
Accuracy 0.5

C at 25

C
MECHANICAL
Battery 2X AA batteries Attached pack
User Interface USB v1.1 or higher
Size (in) 2.55 x 1.24 x 0.24 w/o battery pack
(mm) 65 x 31 x 6 w/o battery pack
Weight (oz) 0.8 w/o batteries
(grams) 23 w/o batteries

Programmable in 1-MHZ steps, 5-MHz steps for compliance with IEEE

802.15.4/D18-2003.
Table 2. The specication of TelosB (source: Crossbow).
matches the results reported by other groups [13, 7, 9].
Based on the connectivity matrices, we also generated
connectivity graphs. Each vertex in the graph corresponds
to a node. There exists an edge between two vertices if
the communication links between the nodes associated with
these vertices meet certain requirements. For the RSSI
metric, for instance, the RSSI value of the links in both
directions must be at least 90 dBm. For the LQI metric, in
turn, the LQI value of the links in both directions must be at
least 95. Finally, for PLR metric, the packet loss rate of the
links must be below 15%. These thresholds represent high-
quality links and were obtained from the results reported by
other groups [13, 7, 9].
For different transmission power levels and link quality
metrics, we computed certain graph-theoretical properties
of the connectivity graphs. The properties include: the graph
diameter (the maximal shortest path between all pairs of
nodes), the network density (the degree of a node in the
connectivity graph), and the clustering coefcient (the ratio
of the actual number of links between a nodes neighbors
to all the possible links between the nodes neighbors). The
results for different metrics and power level are shown in
Table 3.
It can be observed that in all congurations the network
110
100
90
80
70
60
50
node id
n
o
d
e

i
d
Average RSSI (RF Power: 25 dBm)
1 6 10 20 25 40 45 60 65 70 80 84 87
1
6
10
20
25
40
45
60
65
70
80
84
87
(a) average
0
20
40
60
80
100
120
node id
n
o
d
e

i
d
Standard Deviation of RSSI (RF Power: 25 dBm)
1 6 10 20 25 40 45 60 65 70 80 84 87
1
6
10
20
25
40
45
60
65
70
80
84
87
(b) standard deviation
Fig. 3. RSSI matrices for the RF power of 25 dBm.
is multi-hop with the diameter of 3 for the highest trans-
mission power and the diameter of to 4 or 5, depending on
the link quality metric, for the lowest transmission power.
A node has on average more than 15 and less than 30
high-quality neighbors. Moreover, the connectivity graph is
highly clustered, as mentioned previously. High clustering
is an inherent feature of wireless networks as, due to the
limited radio range, two nodes that have a common neigh-
bor are very likely to be each others neighbors as well [14].
To complete the picture, in Fig. 6 and Fig. 7 we show
the distribution of node degree and path length for the PLR
metric and the lowest transmission power.
The degree distribution (see Fig. 6) demonstrates that,
more than 50% of the nodes have at most 17 good quality
neighbors. Yet, there exist some nodes that have more than
6 K. Iwanicki et al.
0
10
20
30
40
50
60
70
80
90
100
node id
n
o
d
e

i
d
Average LQI (RF Power: 25 dBm)
1 6 10 20 25 40 45 60 65 70 80 84 87
1
6
10
20
25
40
45
60
65
70
80
84
87
(a) average
0
2
4
6
8
10
12
node id
n
o
d
e

i
d
Standard Deviation of LQI (RF Power: 25 dBm)
1 6 10 20 25 40 45 60 65 70 80 84 87
1
6
10
20
25
40
45
60
65
70
80
84
87
(b) standard deviation
Fig. 4. LQI matrices for the RF power of 25 dBm.
25 neighbors. In general, the node degree distribution is
highly non-uniform and varies between 8 and 31 neighbors.
Such non-uniformity is very common in real-world WSN
deployments.
The distribution of the path length (see Fig. 7) is more
predictable. A bit more than 30% of all 2970 paths are one
hop. This is a direct consequence of the high clustering
coefcient. More than 60% of the paths are at most two
hops, and nearly 90% are at most three hops. However, a
relatively large fraction of paths has the same length as the
network diameter. This implies that, to reach each other,
many nodes must forward messages over the distance equal
to the network diameter. Such long paths facilitate testing
of various routing algorithms.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
node id
n
o
d
e

i
d
Packet Loss Rate (RF Power: 25 dBm)
1 6 10 20 25 40 45 60 65 70 80 84 87
1
6
10
20
25
40
45
60
65
70
80
84
87
Fig. 5. The PLR matrix for the RF power of 25 dBm.
0
5
10
15
20
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
f
r
a
c
t
i
o
n

o
f

n
o
d
e
s

[
%
]
degree [nodes]
Fig. 6. Node degree distribution (RF power of 25 dBm).
4 Conclusions and Future Work
We introduced our 60-node indoor WSN testbed. We out-
lined the hardware architecture of the testbed and presented
the basic properties of the internode connectivity graph.
The graph has highly non-uniform node density and many
paths reaching the network diameter of 4 to 5 hops. We
believe that these properties of our testbed will enable sound
evaluation of many of the WSN protocols and systems we
have devised or hope to devise in the future.
During our experiments, we also identied a need for au-
tomating testbed reprogramming and experiment schedul-
ing. To this end, we plan to augment our testbed with some
job scheduling software, similar to the software for cluster
computers. This would allow multiple users to concurrently
access the testbed, which is simply a necessity if the user
population grows beyond a few persons.
KonTest: A Wireless Sensor Network Testbed at Vrije Universiteit Amsterdam 7
RF Power Diameter
Density Clust. Coeff.
AVG STDEV AVG STDEV
25 dBm 4 19.65 5.96 0.82 0.13
15 dBm 3 27.62 5.58 0.84 0.12
5 dBm 3 29.22 6.07 0.83 0.12
0 dBm 3 29.87 6.85 0.80 0.11
(a) RSSI metric
RF Power Diameter
Density Clust. Coeff.
AVG STDEV AVG STDEV
25 dBm 5 17.04 5.83 0.83 0.13
15 dBm 3 25.96 4.29 0.86 0.12
5 dBm 3 27.51 4.69 0.86 0.12
0 dBm 3 28.31 5.72 0.86 0.12
(b) LQI metric
RF Power Diameter
Density Clust. Coeff.
AVG STDEV AVG STDEV
25 dBm 4 17.87 5.57 0.78 0.13
15 dBm 3 27.00 4.78 0.84 0.12
5 dBm 3 28.24 4.75 0.84 0.11
0 dBm 3 29.04 5.03 0.82 0.11
(c) PLR metric
Table 3. Connectivity graph properties depending on RF power.
0
10
20
30
40
50
0 1 2 3 4 5
f
r
a
c
t
i
o
n

o
f

p
a
t
h
s

[
%
]
path length [hops]
Fig. 7. Path length distribution (RF power of 25 dBm).
Acknowledgments
We would like to thank all people that allowed us to deploy
the sensor nodes in their ofces: A. Bakker, P. Costa,
J. Domaschka, D. Gavidia, R. Kemp, K. Mitrokotsa, J. Nap-
per, N. Paul, V. Rai, L. Ranaweera, M. Rieback, and
J. Sowi nska. Without their cooperation the deployment
could not have been completed.
References
[1] D. Culler and W. Hong, Eds., Communications of the ACM, June
2004, vol. 47, no. 6, ch. Wireless Sensor Networks, pp. 3057.
[2] R. Szewczyk, A. Mainwaring, J. Polastre, J. Anderson, and
D. Culler, An analysis of a large scale habitat monitoring appli-
cation, in Proceedings of the Second ACM Int. Conf. on Embedded
Networked Sensor Systems (SenSys), Baltimore, MD, USA, Novem-
ber 2004, pp. 214226.
[3] G. Tolle and D. Culler, Design of an application-cooperative man-
agement systemfor wireless sensor networks, in Proceedings of the
Second European Workshop on Wireless Sensor Networks (EWSN),
Istanbul, Turkey, January 2005.
[4] N. Ramanathan, E. Kohler, and D. Estrin, Towards a debugging
system for sensor networks, International Journal of Network
Management, vol. 15, no. 4, pp. 223234, July 2005.
[5] K. Langendoen, A. Baggio, and O. Visser, Murphy loves potatoes:
Experiences from a pilot sensor network deployment in precision
agriculture, in Proceedings of the Twentieth IEEE International
Parallel and Distributed Processing Symposium (IPDPS), Rhodes
Island, Greece, April 2006.
[6] G. Werner-Allen, K. Lorincz, J. Johnson, J. Lees, and M. Welsh,
Fidelity and yield in a volcano monitoring sensor network, in
Proceedings of the Seventh USENIX Symposium on Operating
Systems Design and Implementation (OSDI), Seattle, WA, USA,
November 2006, pp. 381396.
[7] K. Srinivasan, P. Dutta, A. Tavakoli, and P. Levis, Some implica-
tions of low-power wireless to IP routing, in Proceedings of the
Fifth ACM Workshop on Hot Topics in Networks (HotNets), Irvine,
CA, USA, November 2006, pp. 3136.
[8] M. M. Holland, R. G. Aures, and W. B. Heinzelman, Experimental
investigation of radio performance in wireless sensor networks,
in Proceedings of the Second IEEE Workshop on Wireless Mesh
Networks (WiMesh), Reston, VA, USA, September 2006, pp. 140
150.
[9] K. Srinivasan and P. Levis, Rssi is under appreciated, in Proceed-
ings of the Third ACM Workshop on Embedded Networked Sensors
(EmNets), Cambridge, MA, USA, May 2006.
[10] H. Lee, A. Cerpa, and P. Levis, Improving wireless simulation
through noise modeling, in Proceedings of the Sixth International
Conference on Information Processing in Sensor Networks (IPSN),
Cambridge, MA, USA, April 2007.
[11] R. Fonseca, O. Gnawali, K. Jamieson, and P. Levis, Four-bit wire-
less link estimation, in Proceedings of the Sixth ACM Workshop on
Hot Topics in Networks (HotNets), Atlanta, GA, USA, November
2007.
[12] J. Polastre, R. Szewczyk, and D. Culler, Telos: Enabling ultra-low
power wireless research, in Proceedings of the Fourth Interna-
tional Conference on Information Processing in Sensor Networks
(IPSN) (Poster Session), Los Angeles, CA, USA, April 2005, p. 48.
[13] A. Woo, T. Tong, and D. Culler, Taming the underlying challenges
of reliable multihop routing in sensor networks, in Proceedings of
the First ACM Int. Conf. on Embedded Networked Sensor Systems
(SenSys), Los Angeles, CA, USA, November 2003, pp. 1427.
[14] S.-Y. Ni, Y.-C. Tseng, Y.-S. Chen, and J.-P. Sheu, The broadcast
storm problem in a mobile ad hoc network, in Proceedings of the
Fifth ACM Annual International Conference on Mobile Computing
and Networking (MobiCom), Seattle, WA, USA, August 1999, pp.
151162.
A Node to Room Mapping
Figure 8-Fig. 12 present the exact mapping of node iden-
tiers to physical node locations. This data may be useful
when analyzing connectivity matrices.
8 K. Iwanicki et al.
Fig. 8. Room P4.30.
Fig. 9. Room P4.40.
Fig. 10. Room P4.56.
Fig. 11. Room R4.20.
Fig. 12. Room R4.23.