You are on page 1of 12

WIRELESS COMMUNICATIONS AND MOBILE COMPUTING

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492


Published online 29 December 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.1089

RESEARCH ARTICLE

Interference mitigation in WSN by means of


directional antennas and duty cycle control
Kamil Staniec* and Grzegorz Debita
Institute of Telecommunications, Teleinformatics and Acoustics, Wroclaw University of Technology, Wroclaw, Poland

ABSTRACT
Network spanning algorithms, such as ZigBee-native and Stojmenovič, constitute a crucial element in the wireless sensor
network design, by determining its potential for reliability and fault-tolerance. The interconnections between nodes have a
great impact on the radio interference level present in such a network and may create a serious electromagnetic compatibility
issue in some cases. It can be proved that the total interference incurred by a statistical node can be diminished in two ways:
either by using directional antennas or by setting an upper limit on the duty cycle of each network node. Copyright © 2010
John Wiley & Sons, Ltd.
KEYWORDS
sensor network; SNIR; interference; directional antennas
*Correspondence
Kamil Staniec, Institute of Telecommunications, Teleinformatics and Acoustics, Wroclaw University of Technology, Wroclaw, Poland.
E-mail: kamil.staniec@pwr.wroc.pl

1. INTRODUCTION combat interferences from adjacent cells (for further read-


ing refer to Refs. [8,9]). The authors have demonstrated that
Since wireless sensor networks (WSN) have now been at a slightly augmented production costs associated with
being widely used, a new sort of problems has arisen. Apart equipping existing sensor nodes with directional antennas
from issues regarding topology creation, efficient routing and implementing beam-tracking (or -- switching) func-
procedures, sensing elements, etc. a strong attention must tionality, one may remarkably improve the transmission
be paid to the increasing density of sensor nodes (see also performance of the sensor network.
Ref. [1]) in many networks, which may lead to excessive
interference issues. The problem concerns networks that
gather information from a specified area with random dis- 2. INTRODUCTION TO ZIGBEE
tribution of transmitting nodes operating at the same radio SYSTEM (IEEE 802.15.4)
frequency. Interference issues in WSN are not a novelty
[2--7], however, in this paper two other aspects have been Out of the multiplicity of candidate solutions for effective
addressed, which have not been investigated elsewhere. The creation of sensor networks is the ZigBee standard based on
first aspect is the duty cycle (DC) with which sensor nodes IEEE 802.15.4 specification [10]. It defines a low-cost, low-
operate in the network; its values spanning from nearly range, power-saving system allowing data rates -- depending
0% (for sporadic transmissions) to 100% (for continuous on the frequency band, of 20, 40, and up to 250 kbits/s
transmissions). The latter can be treated as a worst-case and almost inexhaustible number of network nodes (typ-
scenario and should occur (if at all) occasionally in ically 16 bits allocated for addressing with or 64 bits using
bursts. However, one may easily imagine situations where extended addresses). It has been defined in three separate
simultaneous transmissions from multiple (even all) nodes bands with 27 distinct channels defined therein, as given in
occur -- for instance, when alert information about some Tables I and II.
emergency circumstances on a vast area needs (such as Depending on the application requirements, an IEEE
fire) to be immediately transferred to the sink node for 802.15.4 low-rate wireless personal area network (LR-
countermeasure decisions. WPAN) may operate in either of two topologies [10]: the
The second aspect discussed here is the application of star topology (Figure 1) or the peer-to-peer topology (Fig-
directional antennas in the transmitting sensor modules; a ure 2). In the star topology communication is established
well-known solution in, e.g., cellular networks planning to between devices and a single central controller, called the

Copyright © 2010 John Wiley & Sons, Ltd. 1481


Interference mitigation in wireless sensor networks K. Staniec and G. Debita

Table I. Channels definitions in ZigBee.


A peer-to-peer network can be ad hoc, self-organizing,
Frequency k-th channel and self-healing. It may also allow multiple hops to route
band (MHz) Channels (k) definition (MHz) messages from any device to any other device on the net-
work. Such functions can be added at the higher layer, but
868.0--808.6 k=1 Fk = 863.3
are not part of the IEEE 802.15.4 standard. Two different
902--928 k = 1, 2, . . . , 10 Fk = 906 + 2(k−1)
device types can participate in an IEEE 802.15.4 network;
2400--2483.5 k = 11, 12, . . . , 26 Fk = 2405 + 5(k−11)
a full-function device (FFD) and a reduced-function device
(RFD). The FFD can operate in three modes serving as
a personal area network (PAN) coordinator, a coordina-
tor, or a device. An FFD can talk to RFDs or other FFDs,
while an RFD can talk only to an FFD. An RFD is intended
for applications that are extremely simple, such as a light
switch or a passive infrared sensor; they do not have the
need to send large amounts of data and may only asso-
ciate with a single FFD at a time. Consequently, the RFD
can be implemented using minimal resources and memory
capacity.
As will be treated in detail in later sections, the peer-to-
peer topology beside offering network scaling advantages
does not account for two problems: the network survivabil-
Fig. 1. Star topology example in ZigBee. ity/fault tolerance and mutual radio interference. As for the
first problem, ZigBee offers no inherent feature (see Section
PAN coordinator. The peer-to-peer topology also has a 3) guaranteeing each node to have at least two other nodes
PAN coordinator; however, it differs from the star topology as candidates for connections (compare, for instance, with
in that any device may communicate with any other device Figure 2 where failure of almost any node would cause the
as long as they are in range of one another. Peer-to-peer entire extended structure to break up into separate discon-
topology allows more complex network formations to be nected parts). As for the second problem, it will be shown in
implemented, such as mesh networking topology. 9 and 10 that this native neighbor-searching (i.e., network
Applications such as industrial control and monitoring, spanning) mechanism may lead to the generation of severe
WSN, asset, and inventory tracking, intelligent agriculture, radio interference, especially in high-density and high DC
and security would benefit from such a network topology. sensor networks.

Table II. Basic ZigBee (IEEE 802.15.4) parameters [10].

Frequency (MHz) BW (MHz) Data rate (kbits/s) Receiver sensitivity (dBm) SNIR (dB) @1%PER

868--868.8 0.6 20 −85 Not specified


905--928 2 40
2400--2483.5 2 250 5--6

Fig. 2. Peer-to-peer topology example in ZigBee.

1482 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm
K. Staniec and G. Debita Interference mitigation in wireless sensor networks

Fig. 4. A result of MST network formation with a basic listen


window.
Fig. 3. A result of LMST network formation with four subnets.

4. STOJMENOVIC MINIMUM
SPANNING TREE ALGORITHM
Out of multiple concepts of modeling and building the
3. IEEE 802.15.4 (ZIGBEE) LOCAL WSN topology (as in Refs. [11,12]) some of the most pop-
MINIMUM SPANNING TREE ular ones are those described by Stojmenovič [13], initially
ALGORITHM intended for military applications. These include: MST,
LMST, relative neighbor graph (RNG), Gabriel’s graph
From the two general topologies for ZigBee networks intro- (GG), Delaunay triangulation (DT) and partital Delaunay
duced in Section 2 two further topologies can be derived, triangulation (PDT).
namely the mesh and tree, which gives altogether four The MST algorithm was chosen (for comparative pur-
possibilities regarding the network structure. The topology poses with LMST described in Section 3) as the most
determines the spanning method to be chosen by a designer, relevant to WSN design due to its outstanding energetic effi-
which in turn impacts the system functionality. The most ciency (a strategic parameter in WSN) and fault-tolerance.
commonly implemented methods for ZigBee are the min- The MST algorithm starts with the PAN coordinator
imum spanning tree (MST) and local MST algorithms broadcasting a beacon message heralding a new network
(LMST), optimal due to a low-computational complexity formation, which is to announce that a new network is about
and energetic efficiency (for more details see). Both are to be created. Soon after this broadcast the PAN coordina-
considered to be heuristic method for a fast network form- tor enters the listening mode to hear incoming response
ing. Out of these two, the LMST is more frequently used calls from nodes within the radio audible range. The suc-
since it allows to span the network without prior knowledge cessfully received calls are being collected for the duration
of the whole network state (although, as will be shown in of the, so-called, listen window which -- depending on the
Sections 8 and 9, both algorithms perform very similarly as manufacturer -- lasts for a few up to several seconds (val-
regards intra-network interference issues). ues experimentally measured by the authors with the use
Both methods have one feature in common -- it is the pres- of the Daintree Sensor Network Analyzer Software 100A,
ence of the PAN coordinator(s) to initiate the formation of Prof. Edition -- a commercial software for ZigBee networks
the network. In LMST (typical for ZigBee applications), analysis). Now, after forming connections to the PAN coor-
however, the formation process starts in a few points of dinator each connected node performs the same searching
the unorganized network independently (nodes 8, 17--19 routine to find nodes of its own neighbor set (further away
in Figure 3) -- several FFD’s start a network spanning from the PAN coordinator) and which have not connected
routine to create connections in a star-topology around to the PAN coordinator upon the primary broadcast. Those
themselves (hence ‘local’ in the algorithm name). Once all nodes that respond within the listen window are recruited
the local stars have been created, connections are made as neighbors. The procedure then continues throughout all
between these local individual stars to span the whole the nodes in the entire area until eventually the minimum
network. (global) spanning tree is formed (Figure 4).

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd. 1483
DOI: 10.1002/wcm
Interference mitigation in wireless sensor networks K. Staniec and G. Debita

For example, the node 16 in Figure 5 may choose to trans-


mit via nodes 12, 13, or 18. The greatest transmit power
(mind the energy consumption) would, of course, be neces-
sary if the link 12 were selected and this may happen with
a 33% probability since all links in the neighbor set are
equally probable.
If the transmit power cannot be further increased (e.g.,
a maximum has been reached), one must account for the
received signal strength S at node no. 12 being much lower
than if the transmission were directed to nodes 13 or 18.
As will be presented in 6, the ratio between S and the sum
in all other nodes’ emissions (plus noise) cannot decrease
below a threshold value defined by a standard (5 dB here)
for successful transmission. Therefore, it is very desirable
to achieve as high level of S as possible since interference
will definitely not decrease with distance.

Fig. 5. A result of MST neighbor formation with a doubled listen


window. 5. PROPAGATION MODELING IN
WIRELESS SENSOR NETWORKS
It is assumed that the nodes which respond the fastest will As is stated in Ref. [10]: ‘a well-defined coverage area does
be those that lie the closest to the broadcasting node, thus not exist for wireless media because propagation character-
they may be considered as minimum-pathloss connections istics are dynamic and uncertain. Small changes in position
as compared to the calls that arrived after the listen window. or direction may result in drastic differences in the signal
Algorithms with a small (basic) window form a network strength or quality of the communication link. These effects
initial structure. It is obvious that if the listen window is occur whether a device is stationary or mobile, as mov-
prolonged by a multiple of the basic listen window duration, ing objects may impact station-to-station propagation.’ This
more nodes will join in, thus forming redundant connections statement holds true, especially if one remembers that there
(referred to as a neighbor set). Based on this assumption, exists no universal propagation model fitting all possible
the topology-creating algorithms can be divided into two environments.
groups: In the presented work it is assumed that nodes are dis-
tributed in an open-space area, attached to low-height masts.
(1) Those with a small listen window. It can be derived that for such a scenario the pathloss L
(2) Those with a large listen window (i.e., a multiple of exhibits a two-slope (a TS model) characteristics given
the small/basic listen window). by Equation (1). Figure 6 depicts the received power Pr
attenuated by L with the distance from the transmitter in
The multiple by which the basic listen window is to be the TS model (plotted for the antenna heights of h1 = 5 m
extended is heuristic since neither distances between net- and h2 = 10 m, for better visualization of signal fades at
work nodes are known, nor are their geographical locations. these heights than at those accepted in simulations, namely
In Ref. [13] it is shown that the listen window size may be h1 = h2 = 1.5 m; Figure 6).
extended up to three time the basic size in order to achieve This phenomenon can be explained by observing that
satisfactory level of redundancy (and hence the network if only two waves are considered, i.e., the direct and the
reliability and fault tolerance). It is shown in Figure 5 how ground-reflected one, at distances close to the transmitter
the network structure of connections changes by doubling their respective phase differences upon reception will cause
the basic window, as compared with the small-window case the received signal to be either strengthened (if they arrive
in Figure 4. in-phase) or faded (if in counter phase). After some point
It should, however, be also noticed that there is a par- (called a breakpoint dbrk ) and further, the respective phase
ticular feature that may be potentially threatening. Namely, difference between both waves never exceeds π/2 hence
the algorithm enables connections that intersect other con- both diminish monotonously. In the discussed case where
nections, e.g., pairs (1, 2) and (3, 14) in Figure 5. In radio h1 = h2 = 1.5 m and λ = 12.25 cm the breakpoint distance
communications avoidance of such a situation is a strongly equals 231 m.
recommended practice, especially if directional antennas

are used to reduce interference (see a notice on this issue 32.45 + 20 logfMHz + 20 logdkm d ≤ dbrk
L= (1)
in Section 6). Another issue worth mentioning is that since 20 log(h1 h2 /d 2 ) d > dbrk
with an extended listen window most nodes have redundant
connections, it is a purely random matter which of them In the ongoing investigations the 2.4 GHz industrial, sci-
will be chosen for transmission of a current packet. entific, medical (ISM) band has been chosen since 868 MHz

1484 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm
K. Staniec and G. Debita Interference mitigation in wireless sensor networks

a solution, however, would require some sort of adaptive


antenna system to be installed on board each transmitting
node. Now, every time a new packet were to be sent, a neigh-
bor node would be randomly selected out of the predefined
set and the main lobe of transmitting antenna would switch
its direction in order to concentrate its energy toward the
selected neighbor.
The goal of the research was to study the standard
network-spanning algorithms with respect to intra-network
interference induced by them. For this purpose a mathe-
matical model has been proposed and implemented as a
software for calculating mutual interference levels expe-
rienced by each node in the network. As for modeling
propagation losses a simple TS model has been proposed for
the radio signal attenuation between communicating nodes.
It is considered a better-suited solution for low-height sen-
sor antennas since the pathloss formula proposed in Ref.
[10] for LR-WPAN systems seems to be too static and opti-
mistic (as was demonstrated in Refs.[14,15]). In this model,
the propagation losses follow a 20 dB per decade fade before
Fig. 6. TS propagation characteristics. the breakpoint distance, dbrk , and 40 dB per decade after.
The radio technology behind these investigations is IEEE
802.15.4, commercially known as ZigBee.
Depending on the world region and required data rate,
ZigBee may operate in different modes. These are specified
in Table II (based on Ref. [10]) together with some other key
system parameters to be accounted for in signal-to-noise-
interference ratio (SNIR) calculations. The signal-to-noise
and interference for the k-th node is given by Equation (2),
where S is the desired signal power in (W), kB the Boltz-
man’s constant in (J/K), T the ambient temperature in (K),
BW the channel bandwidth in (Hz), NF the noise factor, M is
the total number of network nodes, whereas the last term in
the denominator denotes the sum of radiations from all other
nodes in the network (excluding, of course, the radiation
from the k-th node for which SNIR is being calculated).

 
S
SNIRk |[dB] = 10 log  
N+ Ik
Fig. 7. General interference scenario for ZigBee. k
 
S
offers limited performance capabilities (low-data rates) and = 10 log   (2)
only a single channel whereas 915 MHz band is unavailable kB · T · BW · NF + Im
M/(k)
for license-exempt applications in Europe.

According to Ref. [10], ZigBee transmitters are expected


6. INTERFERENCE CALCULATION to operate at powers between −3 and 10 dBm, with 0 dBm
IN WIRELESS SENSOR NETWORKS being typical for 2.4 GHz ISM band (refer to Table II). A
compliant device shall be capable of achieving a sensitivity
Interference in WSN (see Figure 7) occurs mainly because of −85 dBm or better. Although no specific requirements
even if direct line-of-sight directions between communicat- can be found in Ref. [10] on SNIR for sub-gigahertz ZigBee
ing nodes are not interfered, radio modules in the nodes can implementations, of major concern is the ISM-band version
still hear transmissions from other nodes due to (most usu- (most common).
ally) omni-directional antennas installed in them. The use For the purpose of comparison between the two network
of directional radiators could aid the situation by constrain- spanning algorithm a software simulator has been created
ing emissions/receptions in only specified directions. Such comprising of two components:

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd. 1485
DOI: 10.1002/wcm
Interference mitigation in wireless sensor networks K. Staniec and G. Debita

Fig. 8. An example of connections in a WSN (Stojmenovič Fig. 9. An example of connections in a WSN (ZigBee algorithm).
algorithm).

possible connections has been established, the actual con-


(1) Matlab scripts performing the network spanning nection is picked at random out of this set. The situation
algorithms on the random scenarios generated by the applies to both ZigBee native and Stojmenovič algorithms.
C++ application. Now, if a distant connection is chosen it may happen
(2) A C++ Builder application for (a) generating ran- that the transmitted signal will suffer considerable loss,
dom node distribution scenarios to be fed into the especially after the breakpoint, dbrk (refer to Section 5).
Matlab scripts and (b) calculating SNIR based on The weak desired signal combined with the aggregated
the output files generated by the Matlab scripts. interference from other nodes may, in turn, drastically
decrease SNIR and possibly disable transmission. Such a
In order to study the behavior of the ZigBee network, mul- black scenario is depicted in Figure 9 (for the same nodes
tiple scenarios have been generated with a variable number arrangement as in Figure 8) where connections have been
of nodes, each located at 1.5 m above the ground and trans- determined with ZigBee algorithm. Node 29 in this example
mitting with the power 0 dBm (at 0 dB of gain). Nodes are may connect to nodes 25, 28, and 30. Although the lat-
scattered randomly over a square area of 500 m × 500 m. ter case is extremely unfavorable due to the high expected
As can be directly seen from an example scenario presented pathloss, it will be selected as a valid connection with 30%
in Figure 8 the ZigBee native network spanning algorithm probability.
allows wireless links to intersect. In the example the link
between nodes 3 and 17 is intersected by the link between
nodes 25 and 7; the same can be observed for node pairs 7. A CONCEPT OF ANTENNA
(16, 17) and (21, 20). DIRECTIVITY IN INTERFERENCE
Obviously, the chances of intersecting links will grow MITIGATION
with the number of density of nodes deployed in the area.
This is not of a great importance with omni-directional One of the two aspects to be analyzed in the presented work
antennas, but -- as will be demonstrated in Ref. [8] -- will is the idea of using directional antennas to combat inter-
have a significant influence on the intra-network interfer- ference introduced by other nodes. As is shown in Figure
ence if directional antennas are used. It should also be 10 interference is expected to be partially omitted if one
noticed that each node may (not necessarily will) have more assumes the use of directive antennas in the radio modules
than one possible neighbors to connect to after the network attached to sensors. In the example a desired transmission
spanning routine, which is supposed to assure redundancy session takes place between the sensor pair (1, 2) whereas
in connections (refer to Sections 3 and 4 for details). three other transmission sessions have also been established
An issue not to be underestimated is the fact that during in the vicinity: (3, 4), (5, 6), and (8, 7).
the alternative links selection these are not selected on a If directional antennas are used, only (3, 4) session will
minimum-pathloss criterion and the assumption that nodes interfere with the desired session. Otherwise, for omni-
first to respond are the closest to the initiator is not true directional reception -- all three foreign communications
in all circumstances. For instance, some closer nodes may would inject interference into node no. 2 thereby diminish-
simply be involved in response transmissions to other PAN ing SNIR observed by that node, as follows directly from
coordinators at the moment and will therefore not be able Equation (2). Obviously, in order to take most advantage
to respond at the moment. On the other hand, once a set of of this narrowed beamwidth, the beam switching must also

1486 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm
K. Staniec and G. Debita Interference mitigation in wireless sensor networks

boresight region, diminished by 25 dB with respect to the


intended transmission direction contained within θ 3dB .
The advantages of directionality have also been studied
in aspects addressing routing and medium access control
strategies for WSN, as in Refs. [16--20]. This paper, in turn,
is focused on the impact of directional antennas usage on
the physical layer of the WSN operation.

8. ANALYSIS OF RESULTS --
ANTENNA BEAMWIDTH INFLUENCE
ON INTERFERENCE
Fig. 10. An example of interference mitigation with the use of
directive antennas in radio modules.
With the assumptions in previous sections in mind, multiple
simulations have been carried out to determine the influence
of the antenna beamwidth on the level of experienced inter-
ference. The number of nodes was varied from 10 to 90 and
for each case number, 10 random scenarios were generated
with Monte--Carlo distributed nodes. For every individual
node SNIR was found in each of the 10 scenarios and aver-
aged. The final value of SNIR as seen in Figures 13--15 was
calculated as the average across all the nodes in all scenarios
for a given number of M nodes. Such an averaging proce-
dure was intended to provide statistical robustness to results,
Fig. 11. The concept of antenna directivity switching: (a) packet otherwise -- a per-instance analysis might exhibit flickering
#1 transmission; (b) packet #2 transmission. variations stemming from random choices of neighbors in
Stojmenovič and ZigBee native algorithms (as described in
be involved since a given node picks one of the candidate Sections 3 and 4). A typical value of minimum SNIRmin
nodes (once for each packet transmission) randomly out of defined for ZigBee (according to Table II) was indicated as
its neighbor set (see Figure 11). Obviously, the beamwidth a horizontal dashed line.
control comes at some additional costs of complicating the It can be clearly seen that the SNIR changes non-linearly
transmission/reception module. However, as will be shown with a strong dependence of SNIR with the beamwidth. In
in Sections 8 and 9, the expected improvement in SNIR the 0◦ beamwidth case SNIR is only limited by the ther-
justifies both effort and cost. mal noise and some rudimentary interference from other
The final remark regarding the implementation of the nodes (due to 25 dB signal attenuation in their directions)
directional antennas concerns the antenna radiation pattern and the pathloss incurred by the desired signal. As seen in
(ARP) shape. Since its exact form usually differs by manu- Figures 13--15, regardless of the algorithm, SNIR crosses
facturer, a generic simplified approach has been proposed to its lower bound (5 dB) when the antenna beamwidth, under-
model ARP provided the 3 dB (boresight) angle is known, stood as the θ 3dB angle discussed in Section 7, reaches ca.
specified as θ 3dB in Figure 12. 180◦ (which means that antennas radiate over one half the
The θ 3dB angle is defined as the angle spread on which horizontal plane and attenuate signals in the other half,
edges the transmitted power with a given ARP reduces by according to Figure 12) for 10 nodes. Statistically, this case
3 dB. For the sake of simplicity a straight-line sectorization halves the total number of nodes M in the visible range
has been proposed (gray area in Figure 12b) with a vesti- of each node. Quite surprisingly, however, SNIR responds
gial reception of the back- and sidelobes lying outside the most rapidly when the number of nodes is increased from

Fig. 12. A model of ARP: (a) exact; (b) simplified.

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd. 1487
DOI: 10.1002/wcm
Interference mitigation in wireless sensor networks K. Staniec and G. Debita

Fig. 16. Minimum beamwidth dependence on the number of


nodes in the network.

Fig. 13. Averaged SNIR versus antenna beamwidth for different


nodes densities - Stojmenovič algorithm (a small window).

Fig. 17. Averaged number of interferences versus antenna


beamwidth for different node densities.

Fig. 14. Averaged SNIR versus antenna beamwidth for different In any case, however, Stojmenovič algorithm in the small
nodes densities - Stojmenovič algorithm (a large window). window mode behaves similarly to ZigBee native algorithm
in that they both perform slightly poorer than the large-
window Stojmenovič algorithm (i.e., a narrower beamwidth
is necessary to achieve the SNIR of 5 dB as compared to the
large window algorithm).
Figure 17, in turn, shows how the number of interfer-
ences changes with the antenna beamwidth. Obviously, for
omni-directional reception (360◦ case), the total number of
interferences equals M-2 since a given node cannot interfere
with itself or with the node it currently transmits to. It should
also be noticed that ZigBee and Stojmenovič algorithm do
not exhibit significant differences in the interference perfor-
mance despite different principles in the network structure
formation (refer to Sections 3 and 4).

Fig. 15. Averaged SNIR versus antenna boresight for different


nodes densities - ZigBee native algorithm. 9. ANALYSIS OF RESULTS -- DUTY
CYCLE INFLUENCE ON
INTERFERENCE
10 to 20 (refer to Figure 16): the required beamwidth to sus-
tain the minimum SNIR has to be reduced by 45, 60, and In this section attention is paid to another factor which
45◦ for the ZigBee native and Stojmenovič algorithm (small affects the interference performance of a mesh WSN -- the
and large window), respectively. For all further increases in DC of the transmitting modules. As was mentioned in pre-
M, i.e., in the range 20--90 (see Figure 16), the necessary vious sections, the results presented here concerned the
reductions in the required beamwidth to sustain the min- blackest of the possible scenarios with an important con-
imum SNIR, are far less pronounced than for the 10--20 clusion that even under such unfavorable conditions the
nodes transitions, and equal 25, 35, and 20◦ . minimum SNIR can be met provided an adequate constraint

1488 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm
K. Staniec and G. Debita Interference mitigation in wireless sensor networks

Fig. 18. Averaged experienced number of interferences versus


DC of transmitting modules at 360◦ beamwidth.

Fig. 19. Averaged number of interferences versus antenna bore-


sight for different values of the DC.
is imposed on the antennas beamwidth. In most common
applications and in majority of cases, however, there is no
need for all the nodes to radiate simultaneously. An excep-
tion being some emergency situations when a disastrous
phenomenon is sensed by multiple nodes over a vast area
(e.g., a flood or fire). If so happens the DC, is considered
to be near 100%. Under normal operation, however, DC
is usually far less than this maximum [21] -- imposes an
upper bound of 1% on DC for devices operating at 868 MHz,
although no such regulation exists for the band under con-
sideration in this paper, i.e., 2.4 GHz ISM.
For the analyses, the most interference-prone scenario
with 90 nodes has been chosen. As was given in Figures
13--15, in order to maintain undisturbed communication the
antenna beamwidth had to be reduced down to 90◦ . It will
now be verified how this result is influenced by the DC. Fig. 20. Averaged SNIR versus DC of transmitting modules with
In the presented calculations a given node was assumed to Stojmenovič algorithm (a small window).
experience interference from another node if both follow-
ing conditions were satisfied: firstly -- the other node’s and
the tested node’s radiation patterns are within the visibility
range of each other and secondly -- the potentially interfer-
ing node is active (which comes at a probability equal to
DC). Otherwise, if any of the two conditions is not fulfilled,
the potentially interfering node is deemed either invisible
or inactive. Furthermore, as for the Stojmenovič algorithm,
only the small-window results were presented since almost
no difference with the large window was observed. It is best
demonstrated in Figure 18 which summarizes the number of
interference experienced by an average network node in the
worst-case omni-directional reception. A detailed view of
the number of interference as a function of the beamwidth is
provided in Figure 19. One can observe a linear dependence Fig. 21. Averaged SNIR versus DC of transmitting modules with
of this number with the DC (DC = 100% case the results ZigBee algorithm.
converge to those presented in Section 8).
From direct observation of Figures 20 and 21 it is also evi- into exponential relation between the minimum beamwidth
dent that the DC may dramatically change the interference (required to meet the minimum SNIR) and a DC.
situation from almost interference-free case even for omni- This can be seen in Figure 22 -- the strongest impact of
directional reception (DC = 1--5%) to heavy-interference the DC on the achieved results for DC change from 1 to
scenarios requiring serious narrowing of the beamwidth to 30%. In this range θ min undergoes a reduction from 360◦
sustain a proper radio transmission. down to ca. 150◦ (yielding 210◦ of total change).
Lastly, it should be noticed that despite the linear depen- For the remaining range of DC (i.e., 30--100%) the
dence of the experienced number of interferences with the changes in the minimum beamwidth scaled from ca. 150◦
DC of the network devices, this linearity translates itself down to 100◦ (yielding only 50◦ of total change). A best-fit

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd. 1489
DOI: 10.1002/wcm
Interference mitigation in wireless sensor networks K. Staniec and G. Debita

modules, is a considerable complication of the transceiver


electronics on the sensor boards which are by definition
meant to be rather small and inexpensive devices. Since the
authors’ intention was not to provide any recipe or solu-
tions in this respect, the results presented in the paper may
therefore be treated as encouragement to the research and
development in this area by demonstrating the expected ben-
efits in the physical layer obtained from the implementation
of antenna directivity.
The authors are in progress with devising a network
spanning algorithm -- alternative to the two above men-
Fig. 22. Minimum beamwidth versus DC of transmitting tioned (i.e., either Stojmenovič or ZigBee native), which
modules. possesses an inherent radio interference-mitigation feature.
This is to be achieved by avoiding establishment of cer-
exponential relation has also been found, given by Equation tain interference-generating connections at an early stage of
(3), to analytically express the minimum beamwidth θ min as the network design. Initial results prove that such an algo-
a function of the DC. However, it should be noticed that the rithm is viable. Moreover, some measurements have been
improvement in the network performance is discussed here performed with adequately reprogrammed ZigBee devices,
in the statistical sense -- as an average of all transmission indicating that the novel algorithm also (considerably) out-
sessions. Therefore, half of the total number of packets sent performs the popular methods described in Sections 3 and
in the network may expectedly be lost due to insufficient 4 with respect to the energetic efficiency (on the order of
SNIR. ca. 10 times).
The average mismatch between the three curves drawn
in Figure 22 and the best-fit curve equals 6.35◦ .
ACKNOWLEDGEMENTS
θmin = 446.7 DC−0.32 (3)
This paper has been written as a result of realization of the
project entitled: ‘Detectors and sensors for measuring fac-
10. CONCLUSIONS AND FURHER tors hazardous to environment -- modeling and monitoring
RESEARCH of threats.’ The project funded by the European Union via
the European Regional Development Fund and the Polish
The paper describes in detail two popular algorithms for state budget, within the framework of the Operational Pro-
spanning wireless sensor ad-hoc networks, namely ZigBee- gramme Innovative Economy 2007--2013. The contract for
native and Stojmenovič. It is stressed that in many instances refinancing no. POIG.01.03.01-02-002/08-00.
the mesh topologies created by them are not fault-tolerant
(i.e., in the case of a node failure no alternative paths are
present to bypass this node). Moreover, since links between
nodes in such a network are possible to intersect one another, REFERENCES
an electromagnetic incompatibility threat may arise, espe-
1. Mordachev V, Loyka S. On node density -- outage prob-
cially in the case of simultaneous transmissions from many
ability tradeoff in wireless networks. IEEE Journal on
nodes lying in the proximity of one another.
In order to study the intra-network interference effec- Selected Areas in Communications 2009; 27(7): 1120--
tively, a simple two-ray propagation model has been 1131.
proposed and demonstrated to be adequate for low-height 2. Zhou G, He T, Stankovic JA, Abdelzaher T. RID: radio
antennas. A simplified model for the antenna radiation interference detection in wireless sensor network. Pro-
characteristic has also been introduced. Based on these ceedings of IEEE INFOCOM 2005, vol. 2, 13--17 March,
models and the network-spanning algorithms, a software Miami, USA, 2005, pp. 891--901.
simulator has been proposed for calculating the SNIR in 3. Gewali L, Mohamad K, Tun M. Interference Aware
IEEE 802.15.4 (or ZigBee) networks. It has been shown Dominating Set for Sensor Network. Proceedings of
that even in scenarios with simultaneous emissions from Third International Conference on Information Technol-
all network nodes one may find an optimal antenna 3 dB
ogy: New Generations (ITNG’06), 0-7695-2497 -4/06,
beamwidth sufficient to maintain the minimum required
April 2006, Las Vegas, Nevada, USA, pp. 268--273.
value of SNIR. Another recommended means of decreas-
ing the intra-network interference is by controlling the DC 4. Xu W, Trappe W, Zhang Y. Defending wireless sen-
of the network devices, i.e., allowing access to the channel sor networks from radio interference through channel
with some probability, referred to in the paper as a DC. adaptation. Proceedings of ACM Transactions on Sen-
An obvious disadvantage of implementing directional sor Network, vol. 4, no. 4, article 18, Aug. 2008, pp.
antennas, at least for the state of the art on sensor radio 18.1--18.34.

1490 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm
K. Staniec and G. Debita Interference mitigation in wireless sensor networks

5. Hekmat R, Van Mieghem P. Interference power statis- and Wireless Communications (PWC) 2003, Sept. 23--
tics in ad hoc and sensor networks. Wireless Networks, 25, Venice (Italy), 2003, pp.590--600.
Springer: Springer Netherlands, vol. 14, no. 5, Oct. 2008, 18. Ramanathan R, Redi J, Santivanez C, Wiggins D, Polit
591--599. S. Ad hoc networking with directional antennas: a com-
6. Vakil S, Liang B. Balancing cooperation and interfer- plete system solution. IEEE Journal on Selected Areas
ence in wireless sensor networks. Proceedings of 3rd in Communications 2005; 23(3): 496--506.
IEEE Annual IEEE Communications Society Confer- 19. ElBatt T, Anderson T, Ryu B. Performance evalua-
ence on Sensor, Mesh and Ad Hoc Communications tion of multiple access protocols for ad hoc networks
and Networks (SECON), vol. 1, Sept. 2006, pp.198-- using directional antennas. Proceedings of IEEE Wire-
206. less Communications and Networking Conference
7. Mordachev V, Loyka S. Statistical properties of (WCNC), March 16--20, 2003, New Orleans (Louisiana),
electromagnetic environment in wireless networks, USA.
intra-network electromagnetic compatibility and safety. 20. Munari A, Rossetto F, Zorzi M. Cooperative cross layer
Proceedings of 20th International Symposium on EMC, MAC protocols for directional antenna ad hoc networks.
Sept. 13--17, Wrocław, 2010, pp.619--624. ACM Mobile Computing and Communications Review.
8. Tran-Xuan C, Koo I. An RSS-based localization scheme ACM: New York, NY, USA, 2008; 12 (2): 12--30.
using direction calibration and reliability factor informa- 21. CEPT, ERC Recommendation 70-03 Relating to the Use
tion for wireless sensor networks. KSII Transactions on of Short Range Devices (SRD), Annex 1: Non-specific
Internet and Information Systems 2010; 4(1): 45--61. Short Range Devices. Version of 16 October 2009.
9. Staniec K, Debita G. Antenna beamwidth control
for improving signal-to-noise ratio in wireless sensor
networks. Proceedings of 2010 14th International Con-
AUTHORS’ BIOGRAPHIES
ference on Intelligent Engineering Systems (INES), May
2010, pp. 103--107. Kamil Staniec, received his MSc
10. IEEE, IEEE Std 802.15.4TM -2006, Part 15.4: Wire- degree from Lodz Technical University
less Medium Access Control (MAC) and Physical layer (with honors) in 2001 and a PhD degree
(PHY) Specification for Low-Rate Wireless Personal from Wroclaw University of Technol-
Area Networks (WPANs). ogy (with honors) in 2006. Since 2006
11. Shooman ML. Reliability of Computer Systems and Net- he has been a scientist with Wroclaw
works -- Fault Tolerance, Analysis and Design, John University of Technology where he
is involved in research on the propa-
Wiley & Sons: 2002.
gation modeling and measurement in
12. Pióro M, Medhi D. Routing, Flow, and Capacity Design
indoor and mine environments. Apart from these issues,
in Communication and Computer Networks, Morgan he also coordinates the tele-information section in a multi-
Kaufmann Publishers: San Francisco, 2004. disciplinary project on sensors and detectors for monitoring
13. Stojmenovič I. Handbook of Sensor Networks -- Algo- factors hazardous to the environment (end of project in
rithms and Architectures, John Wiley & Sons: Ottawa, Dec. 2012), commissioned by the Polish Ministry of Sci-
2004. ence. His investigations in the project concern methods
14. Debita G, Staniec K. Reliable mesh network planning for reducing intra-network radio interference in wireless
with minimization of intra-system interference. Pro- sensor networks and improving their energetic efficiency.
ceedings of Broadband Communication, Information Since 2007 he has also been involved as a chief scientist
Technology & Biomedical Applications BROADBAND- with a project on the next-generation services and networks
where he has developed a deterministic model based on
COM’09, July 15--18, 2009, Wrocław, Poland, pp.
the ray launching for modeling the radiowaves propaga-
243--244.
tion in reverberation chambers, in both amplitude and time
15. Debita G, Staniec K. Algorithms for reliable networks domains.
deployment in mesh topology with flow control. Inter-
national Journal of Electronics and Telecommunications
2010; 56(1): 73--80. Grzegorz Debita, a graduate (MSc)
16. Jasani H, Yen K. Performance improvement using from Wroclaw University of Technol-
directional antennas in ad hoc networks. International ogy, Institute of Telecommunications,
Teleinformatics and Acoustics. He
Journal of Computer Science and Network Security
is currently employed at the Chair
2006; 6(6): 180--188.
of Radiocommunications and Tele-
17. Choudhury RR, Vaidya NH. Impact of directional anten- informatics where he is pursuing a
nas on ad hoc routing. Proceedings of IFIP Personal PhD dissertation (presently awaiting

Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd. 1491
DOI: 10.1002/wcm
Interference mitigation in wireless sensor networks K. Staniec and G. Debita

defense). His domain of research, in general, pertains reliability. Other topics of expertise include optimization
to the tele-information science systems design while his methods in radio systems for intra-network electromag-
main interest if focused on the application of planar netic compatibility and optimization of communication
graphs in wireless mesh networks with regard to their protocols.

1492 Wirel. Commun. Mob. Comput. 2012; 12:1481–1492 © 2010 John Wiley & Sons, Ltd.
DOI: 10.1002/wcm

You might also like