7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
1/8
Mitigating the Effect of Interference in WirelessSensor Networks
Nadeem Ahmed, Salil S. Kanhere and Sanjay JhaSchool of Computer Science and Engineering
University of New South Wales
Sydney, Australia
Email: {nahmed, salilk, sanjay}@cse.unsw.edu.au
AbstractPerformance of a deployed Wireless Sensor Network(WSN) is greatly influenced by the interference it is subjectto during operation. Degradation happens due to interferenceresulting in packet drops, retransmissions, link instability andinconsistent protocol behavior. We have conducted experimentsthat highlight the fact that interference caused by WiFi andco-channel contention significantly degrades the network perfor-mance of protocols. These potential sources of interference musttherefore be accounted for during the design stage of a WSNin order to achieve acceptable network performance. Based onthese observations, we have proposed a multi-hop multi-channeltopology control protocol RMMTC for WSN that takes intoaccount interference caused by WiFi networks in operation inthe vicinity and uses multiple channels at different frequencies tomitigate the effect of co-channel interference. This paper detailsthe design and performance evaluation of our proposed RMMTCprotocol using both simulations and empirical experiments. Inaddition, we have formulated the multiple channel assignmentproblem as an Integer linear program (ILP) and compared theperformance of our distributed protocol with the centralized ILPsolution. The simulation results show that RMMTC performsclose to the optimal centralized ILP and achieves a nine-foldreduction in the percentage of dropped packets when a densenetwork is subjected to interference from WiFi and co-channelcontention.
I. INTRODUCTION
Wireless Sensor Networks (WSN) have been actively re-
searched over the past decade and many protocols have been
proposed covering the MAC, routing and transport layers.
Recently, the focus of research in WSN is shifting from perfor-
mance evaluations using simulations studies to experimental
evaluations of proposed protocols in real-world scenarios. It
has been observed that most of the proposed protocols do not
perform as per design when subjected to real radio environ-
ments. One of the major causes of under-performance is the
interference issues in WSN. Degradation due to interference
results in packet drops, retransmissions, link instability and
inconsistent protocol behavior [1] [2] [3] [4] etc.
Interference for a ZigBee based WSN1 can be broadly
classified as internal, i.e., originating from within the network
e.g., multiple links within the same network that can hear
each other and external, e.g., WiFi interference. Effect of
internal interference is mitigated by careful topology selection,
choice of appropriate MAC layer or use of multi-channels.
Interference due to WiFi sources, on the other hand, is difficult
1In this paper, we use ZigBee based WSN and WiFi interference asillustrative examples. In general, the ideas discussed can be applied to anymulti-channel WSN operating in presence of external interference.
to anticipate due to several reasons. Firstly, WiFi operate
on the same frequency band (2.4 GHz) as ZigBee based
WSN and hence the WiFi transmissions appear as noise to
ZigBee. Secondly, the WiFi transmissions are likely to be
from far more powerful sources as compared with ZigBee
based sensor nodes thus considerably lowering the Signal to
Noise Ratio (SNR) for the ZigBee transmissions. Considering
that WiFi deployments are widespread and most realistic
scenarios where WSN would be deployed would very likelybe co-located with APs e.g., in urban buildings or factory
environment etc., these potential sources of interference must
be accounted for during the design stage of a WSN in order
to achieve acceptable network performance.
One potential solution to mitigate the effect of internal inter-
ference is to form a network topology based on multiple non-
interfering cliques operating on different channels [5]. Typical
WSN devices are capable of channel switching capabilities
that provide support for the use of multiple channels operating
at different frequencies e.g., CC2420 radios used for MicaZ
motes can use 16 different channels in the 2.4 GHz band
[6]. Multiple transmissions can take place on these orthogonal
channels to increase the spectral efficiency. Use of multiplechannels in WSN has been explored in previous research
efforts and multi-channel MAC protocols has been proposed to
improve the network throughput [7] [8] [9] [10]. Most of these
protocols assume the presence of multiple orthogonal channels
for parallel communications. One problem with use of multi-
channel in WSN is that channels are not truly orthogonal
i.e., multiple simultaneous transmissions on adjacent channels
do cause interference due to unwanted power received from
transmitters on adjacent channels [2], [5] etc.
In [11], we presented a study on the interference issues
for multi-channel WSN (partly reproduced in Section III). We
studied the effect of interference on off-the-shelf MicaZ motes
operating on different ZigBee channels. The study showed that
ZigBee transmissions are affected by both external (WiFi) and
internal (adjacent channel) interferences, depending on the net-
work topology and the presence of WiFi sources in the vicinity.
The study also highlighted that effect of adjacent channel
interference is more pronounced when ZigBee transmitters are
using variable transmission powers on adjacent channels.
Based on the above mentioned study, in order to improve the
network performance in presence of both internal and external
interference, we have proposed a distributed RSSI based Multi-
hop Multi-channel Topology Control protocol (RMMTC). This
35th Annual IEEE Conference on Local Computer Networks LCN 2010, Denver, Colorado
978-1-4244-8389-1/10/$26.00 2010 IEEE 160
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
2/8
protocol selects short, multi-hop communication neighbors
with adequate RSSI values to mitigate the effect of external
interference, and also reduces co-channel contention and ad-
jacent channel interference by careful channel assignment to
form multiple collision domains.
In this paper, we formulated the problem of RSSI based
multi-channel topology control as an Integer Linear Program
(ILP) that serves as a benchmark to compare our proposed
heuristic to the optimal solution. We perform a detailedand systematic simulation and empirical study to compare
the network performance of our the RMMTC protocol with
other multi-hop protocols such as simple hop count based
[12] and MintRoute [13]. Simulation results show that for
dense topologies, RMMTC protocol considerably improves the
network performance by achieving a nine-fold reduction in
percentage of drop packets due to interference (from about
45% to less than 5%).
This paper is organized as follows; Section II introduces
related work in the area of interference issues and multi-
channel communications in WSN. Section III describes our
preliminary experiments that establish the characteristics of
WiFi and co-channel interference in ZigBee based WSN.Network model for the topology control problem and the
protocol design for the RMMTC protocol is presented in
Section IV. Section V details the simulation and experimental
study to evaluate the performance of our proposed RMMTC
protocol. Finally conclusion is presented in Section VI.
I I . RELATED WOR K
There is a large body of research exploring the use of multi-
channels in WSN with the majority focussing on proposing the
use of multi-channel MACs to coordinate simultaneous use of
multi-channels [7] [8] etc. Wu et.al. in [5] proposed assigning
different channels to subtrees to avoid channel switching
and use of time-synchronized multi-channel MAC protocols.They proposed sampling the environment to discover ZigBee
channels for use that are least effected by WiFi and then
create node-disjoint multi-path trees for reducing the internal
interference. Our work is different in several ways. Firstly,
instead of finding ZigBee channels that are least effected by
WiFi, we assume that WiFi interference effects almost all of
the ZigBee channels. This assumption is more realistic as WiFi
sources operating on different channels can be turned on/off
effecting different ZigBee channels at different time instances.
Secondly, following the observation in [14] that higher RSSI
values for ZigBee based WSN motes have a strong correlation
with good packet reception rates, we use RSSI as the threshold
for selecting links in the topology that can withstand the WiFi
interference.
A white paper from CrossBow (manufacturer of MicaZ
motes) [6] reports the adverse effect of WiFi on ZigBee
transmissions. However, the reported values are much lower
than that observed by our link characterization experiments
(see Section III). Authors in [3] also confirmed through
experiments the adverse effect of WiFi interference on ZigBee
based networks and proposed to switch ZigBee channels once
interference is detected on the current ZigBee channel using
threshold based interference estimator. The base station in their
approach probes the deployed network for RSSI sampling. The
selected nodes along the path to the data source reports data
back to the base station that then finds the channel with least
interference. The base station then directs the node lying on
the path to the source to switch the channel. This approach
incurs high overhead for a many-to-one data transfer where all
the nodes are sending sensed data back to the bases station,
which is typical of most WSN. Hauer et.al. [15] recently
presented results from an empirical study for the effect of WiFiinterference for body area networks. They took multi-channel
measurements and proposed noise floor as the estimator for
link quality degradation. In [16], the authors proposed channel
surfing schemes by adapting channel allocations to avoid
interference. The proposal is reactive in nature as all nodes
initially work on the same channel and switching (of a group
or the whole network) only takes place when interference is
encountered. In contrast, our work pre-empts the influence of
interference by allocating multiple communications channels
to different subtrees rooted at the base station.
III. LIN K CHARACTERIZATION EXPERIMENTS
We have conducted empirical experiments to characterizethe link behavior when subjected to WiFi and adjacent channel
interference. The initial results appeared in a poster paper [11]
and some of the results are re-produced here to put this work
in proper context.
Fig. 1. RF Spectrum Utilization Showing WiFi (high peaks) and 802.15.4
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 2640
50
60
70
80
90
100
ZigBee Channel No
AveragePacketReceptionRate
0 dBm
25 dBm
Fig. 2. Effect of WiFi Interference
We first study the effect of WiFi interference on the
transmission from WSN nodes operating on different ZigBee
channels. We conducted our experimental study using the
MicaZ platform that provides 16 ZigBee channels operating
between 2.405 to 2.480 GHz frequency range (see Figure 1).
WiFi, one the other hand, has 11 channels which operate in
the same frequency band. As each channel in WiFi occupies
161
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
3/8
a wider RF spectrum than ZigBee (22 MHz as compared
with 3 MHz), a single WiFi channel can simultaneously
cause interference on four adjacent ZigBee channels e.g., WiFi
channel 1 occupies the same RF spectrum as channels 11-14
for ZigBee.
The experiments were performed indoors in an office build-
ing. A pair of transmitter and receiver nodes were placed 1.5m
apart on a table raised about 0.5m above the floor level. The
transmitter sends a total of 5000 packets at the rate of 20packets/second with each packet containing a unique sequence
number. The receiver logs the sequence number and RSSI of
each packet that it receives.
For interference purposes, we used FTP clients operating
at WiFi channels 1 and 11 with -55 dBm and -62 dBm
average received signal power (measured with a spectrum
analyzer, co-located with the receiver). We used two power
levels (maximum 0 dBm and minimum -25 dBm) for the Zig-
Bee transmitter and repeated the experiment for 16 different
ZigBee channels. The average RSSI recorded at the receiver
was -62 dBm and -88 dBm for the maximum and minimum
transmission level, respectively. We note that techniques for
creating reproducible interference proposed in [17] can alsobe utilized for these experiments.
Figure 2 shows that the transmitter is able to maintain a
successful packet reception rate (PRR) above 90%, when using
any of the ZigBee channels, at maximum transmission power.
For the case when the node is transmitting at lowest power
level, effect of WiFi interference is more pronounced with
average PRR values falling to about 44% (for channel 14).
11 12 13 14 15 16 17 1850
60
70
80
90
100
ZigBee Channel
AveragePacketReceptionRate
0 dBm
25 dBm
Mixed Mode(0 and 25 dBm)
Fig. 3. WiFi and Multi-Channel Interference (Channel 11-18)
We next conducted experiments with eight pairs of nodes,
each pair tuned to a separate frequency. Transmitters are
placed in one line at a distance of 1.5m from their respec-
tive receivers. Simultaneous transmissions occur at multiple
frequencies (channel 11-18 and channel 19-26 in separate
experiments). We conducted three different set of experiments.
In the first set, all nodes transmit at 0 dBm power. In the
second set, the node transmission power was reduced to -25
dBm. The last set, referred to as the mixed mode experiment,
nodes using odd number channels transmit at full power while
nodes operating at even number channels transmit at lowest
power. The WiFi channels 1 and 11 were in operation, both
with average received signal power of about -68 dBm.
Figures 3 and 4 show that nodes are able to maintain an
average PRR close to 100% when transmitting at 0 dBm (with
RSSI of about -60 dBm). When the transmission power is
19 20 21 22 23 24 25 2650
60
70
80
90
100
ZigBee Channel
AveragePacketReceptionRate
0 dBm
25 dBm
Mixed Mode( 0 and 25 dBm)
Fig. 4. WiFi and Multi-Channel Interference (Channel 19-26)
reduced to -25 dBm (with RSSI of about -88 dBm), PRR is
dropped to about 90% for channels that are under the influence
of WiFi (channels 11-14 and 22-23 are affected). For the mixed
mode experiment, nodes transmitting at 0 dBm are able to
maintain the PRR close to 100%. For nodes transmitting at
lowest power (even number channels) that are also under the
influence of WiFi interference, the PRR is dropped to 70% -
75% (channels 14 and 22).
In conclusion, our experimental study reveals the following
characteristics of multi-channel interference:
1) ZigBee transmissions are affected by both WiFi and
adjacent channel interference, depending on the received
signal power levels.
2) ZigBee sources are able to maintain good PRR (> 85%)when the received power from transmitter signal and
received power from interference signal differ by less
than 25 dB, for WiFi or adjacent channel transmissions.
3) Mixed mode experiments revealed that variable trans-
mission power WSN designed for reducing interference
are infact more prone to interference when a high power
neighbor is operating on an adjacent channel in addition
to interference from WiFi.
These characteristics serves as guidelines for designing anefficient multi-channel protocol for WSN e.g., ZigBee channel
selection should depend on the topology and whether there are
WiFi sources in operation in the environment. Effect of WiFi
interference can be mitigated by selecting short, multi-hop
communication as compared to long distance transmissions.
Similarly, adjacent channels can be utilized for a multi-channel
WSN by careful channel assignment. We follow these guide-
lines in the design of a distributed topology control protocol
discussed in the next section.
IV. NETWORK MODEL AND PROTOCOL DESIGN
Based on the recommendations from the link character-
ization experiments discussed in the previous section, wedetail here the design of a topology control protocol that can
alleviate the effect of external and internal interferences. We
first describe a graph theoretic network model that is used
to formulate the multiple channel topology control optimiza-
tion problem and then detail the design of a heuristic and
distributed topology control protocol.
A. Network Model
We represent the network as an undirected graph G(V, E)where V is the set of vertices (each v V corresponds
162
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
4/8
to a node in network). E is the set of edges representingthe possible wireless links between sensor nodes. Let Rtand Ri denote the fixed transmission range and interferencerange where Ri > Rt. Let d(u, v) represent the Euclideandistance between two nodes u, v V and K represent thenumber of available channels. For two nodes u, v V , directcommunication is only possible if the d(u, v) Rt and thetwo nodes operates on a common channel k K.
Let Rthresh represent the minimum threshold RSSI valuerequired to overcome the effect of interference. We first prune
the graph G to remove wireless links with RSSI < Rthresh.Let GR(V, E) represent the graph induced by RSSI basedlinks selection where GR G. Given the network graphGR(V, E), we calculate the interference value for a node asfollows:
Assume node vi transmits to node vj (vi and vj Vand both operate on same channel) using transmission power
Pi. The signal from vi will reach the receiver vj with signalstrength Sij
Skij = Pi/d(i, j) (1)
where is the pathloss constant. Let denote the back-ground noise value including the interference from WiFi
transmissions. The SNR at the receiver node vj is given by
SN R = Skij/ > Rthresh (2)
The reception of signal from node vi to vj is also influencedby co-channel interference. Since interference is an issue at
the receiver, let Ukj represent the set of nodes located withinRi distance from the receiver vj and operating on the samechannel k as vj . Co-channel interference Ikj at a receiver vjis thus given by
Ikj =uUk
j
Skuj (3)
The signal to interference and noise ratio (SINR) for the
receiver node vj combines the background noise interference and co-channel interference Ikj and is represented by
SINR = Skij/(Ikj + ) > Rthresh (4)
Note that background interference including the interfer-ence from WiFi is assumed constant for this work for all the
nodes in the topology. A node i causes interference at thereceiver node j if it operates on the same channel k and iswithin Ri of j.
Ykij =
1 if i Ukj0 otherwise
(5)
Let C be the cost function that associates a non-negativecost to each vertex in the network. As we want to minimize
the interference in our case, the cost is assumed to be the
number of nodes operating on the same channel and causing
interference at the receivers along the path from a node to the
sink. The total cost can be defined as the sum of the costs of
vertices.
Total number of nodes effecting a receiver j given by
Ckj =iUk
j
Ykij (6)
During the course of the topology selection, each vertex
vi V chooses a vertex vj V, i = j as its parent creatinglink Lij operating on a channel k for routing to the sink.
Lkij =
1 if i, j are assigned channel k0 otherwise
(7)
The output of the topology control protocol is a spanning
subgraph GT(V, E) where GT G.
The objective function for multiple channel assignment can
be defined as:
Find a network channel assignment graph GT(V, E) G, such that the interference on vertices V is minimized.
The problem can be formulated as an Integer Linear Pro-
gram (ILP) as follows;
Let dest represent the sink, and vkj represent a node j V
operating at channel k.
MinimizekK
vV\{dest}
Ckj vkj
subject to the constraints given by Equations 5, 7 and
vkj =
1 if j is assigned channel k0 otherwise
(8)
Centralized optimization can be performed on a given net-
work Graph G to select GR and subsequentally GT to arriveat an idealized network topology with channel assignment.
B. Protocol Design
The solution of the ILP formulated problem is not feasible
due to its complexity and centralized nature. Hence, we
introduce RMMTC, a distributed and heuristic multi-channel
protocol, that can be used to find an approximate solution
in polynomial time. We have designed the RMMTC protocol
based on the recommendations from our link characterization
experiments discussed in Section III. There are two important
design considerations. First is the choice of the threshold
used for categorizing the external interference (section III
characteristic No 1 & 2). As RSSI at the receiver governs
the interference, we chose RSSI threshold values for selecting
links that are resilient to the effect of external interference.
Second consideration is how to avoid the internal interference
(co-channel as well as adjacent channel, section III character-
istic No 3). We do careful channel assignment for utilizing
multiple channels for this purpose.
There are two popular choices for the design of a multi-
channel protocol. One can assign different channels to different
links. In this case, each node can use multiple channels. The
nodes in the local neighborhood coordinate and decide on the
channel allocations using a multi-channel MAC. This approach
introduces the channel switching and channel coordination
163
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
5/8
overheads. The other choice is to form different routing
subtrees rooted at the base station and assign a unique channel
for transmission to each subtree. All the nodes belonging to
a subtree work on the specified channel. Hence any single
channel MAC can be used without the need for expensive
channel coordination. Similar to the work presented in [5], we
used the later approach for the design of RMMTC protocol.
RMMTC protocol thus establishes links in the tree topology
with two objectives. First, it selects links in the topology thathave RSSI values above a certain RSSI threshold. As the
most popular data transfer pattern in WSN is the many-to-
one transfer where all the nodes send data packets to the base
station, RMMTC ensures that links in both directions have
RSSI values above a certain threshold value.
Secondly, RMMTC forms routing subtrees rooted at the
base station where each subtree can be assigned a non-
interfering communication channel. During the course of the
subtree formation, new nodes are added to a tree based on the
interference levels observed at their parent nodes. This is done
to reduce the internal interference.
As described in Section III, ZigBee based WSN has 16 com-
munication channels available for use out of which channel 15,20, 25 and 26 are least effected by external interference. We
have selected channel 26 as the control channel (required for
the neighborhood discovery and channel assignment process)
for the RMMTC protocol while channel 15, 20, and 25 are the
first channels to be assigned for use by the base station. Once
these channels are assigned, the base station uses channels 11,
23, 17, and 13 in this specific order. Remaining channels are
only assigned if required. This assignment order makes sure
that channels least effected by external interference are utilized
first, followed by alternate channels that would not cause
interference. Adjacent channels, prone to internal interference,
are only added in the topology if sufficient subtrees are formed
that require further unique channel assignment (section IIIcharacteristic No 3). We note here that proactive techniques to
detect channels that do not experience external interference can
also be employed and a list of good quality channels provided
to the base station for use [5]. We have not used this technique
for two reasons. Firstly, these network measurements are
expensive as these need to be taken at several locations in
the topology. Secondly, WiFi sources can be introduced or
switched off at any time in the environment and therefore
these network measurements need to be taken periodically and
topology reconfigured accordingly. We have thus designed the
RMMTC protocol without any presumed knowledge of the
interference environment. Any such knowledge can be utilized
for further enhancing the protocol performance.
We now describe the details of the protocol. Once nodes
are randomly deployed in the target area, the base station
announces its position in the topology by broadcasting HELLO
messages. Nodes that receive this HELLO message extract the
RSSI values and mark base station as 1 Hop neighbor in the
neighbor table. If the RSSI value is more than the Rthresh,the node reply back with an ACK packet. The base station
checks the RSSI values of the received ACK messages and in
case RSSI is more than the Rthresh, replies each such neighborwith a SELECT message containing a unique channel number.
Fig. 5. RMMTC Protocol Message Exchanges
Node on reception of the SELECT message notes the channel
number and marks the upstream link as valid. These nodes
now send a CONFIRM message to the base station.
For the rest of the discussion, we refer to the nodes already
assigned a channel No as level N nodes and the downstream
one hop child nodes as level N+1 nodes. For interference
calculations, nodes need to know the location of other nodes
operating on the same channel (part of the same subtree).
Once a node become a level N node, it broadcasts a REQ
message requesting neighbor information from nodes on thesame subtree. Nodes that are part of the same subtree, reply
back with information about their one-hop neighbors on the
same channel. The level N nodes now discover set Ukj (referSection IV-A) by forming a disc of radius Ri centered at itself.The existing interference Ie is thus equal to the cardinality ofset Ukj .
Once the Ie is calculated, Level N nodes announce theirposition and channel number using the HELLO messages
to the downstream nodes. Neighboring nodes may receive
multiple HELLO messages that pass the RSSI test. These
receiver nodes maintain a list of their eligible parents (level N
nodes) and send an ACK message to each one of them. Level
N nodes do not immediately reply to each of the receivedACK messages. Rather it maintain a list of all ACK messages
that pass the RSSI threshold. Let Nel represent the numberof potential level N+1 nodes. After timeout, level N nodes
sends SELECT messages to all the eligible level N+1 nodes.
SELECT messages contain the channel number, Ie and Nelvalues. Nodes receiving the SELECT messages compares the
Ie values, selecting the level N node with the lowest interfer-ence value. For equal Ie values, level N node with lowest Nelis selected. Ie values control the internal interference whileNel does the load balancing among branches of a subtree. Thisload balancing mechanism ensures that the tree grows in depth
(the chosen channel number continues to be selected down
the topology). The level N+1 node now sends a CONFIRM
message to their selected parent. Level N nodes now exactly
knows how many children have joined it. Figure 5 shows
these message exchanges where Node D sends a CONFIRM
message to Node B as the received Ie value for Node B isless than that at Node A (2 vs 5).
If no CONFIRM message is received by a level N node
despite sending a SELECT message to its children, a new
SELECT message with a force flag is sent again to one of
the children informing that parent node would become a leaf
node if it does not join that subtree. Child node receiving
164
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
6/8
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
Topology Formed by HopCount Assignment
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
2728
29
30
31
3233
34
35
36
37
38
39
40
41
42 43
44
45
46
47
48
49
51
5253
54
55
5657
58
5960
61
62
63
64
6566
67
68
69
70
71
72
7374
75
76
77
78
79
80
81
82
83
8485
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105106
107
108
109 110
111
112
113
114
115 116
117
118
119
120
(a)
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
Topology Formed by RMMTC Assignment
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
2728
29
30
31
3233
34
35
36
37
38
39
40
41
42 43
44
45
46
47
48
49
51
5253
54
55
5657
58
5960
61
62
63
64
6566
67
68
69
70
71
72
7374
75
76
77
78
79
80
81
82
83
8485
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105106
107
108
109 110
111
112
113
114
115 116
117
118
119
120
(b)
0 10 20 30 40 50 60 70 80 90 1000
10
20
30
40
50
60
70
80
90
100
Topology Formed by Centralized ILP Assignment
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
2728
29
30
31
3233
34
35
36
37
38
39
40
41
42 43
44
45
46
47
48
49
51
5253
54
55
5657
58
5960
61
62
63
64
6566
67
68
69
70
71
72
7374
75
76
77
78
79
80
81
82
83
8485
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105106
107
108
109 110
111
112
113
114
115116
117
118
119
120
(c)
Fig. 6. Topology Formation for HopCount, RMMTC and Centralized ILP
this message, sends a LEAVE message to its earlier selected
parent. The selected parent replies with SELECT message with
information of number of children attached to it. If the parent
has more than one child nodes, level N+1 nodes joins the
parent who has sent the SELECT message with the force flag
by sending CONFIRM message. If no reply is received for
the SELECT message with force flag, node can send the same
message to other level N+1 nodes from whom ACK were
received.This process is now repeated for each new level of nodes
till no ACK messages are received by advertising the HELLO
messages. These nodes become the leaf of the formed subtrees.
Nodes that do not receive any SELECT message in response
to their ACK messages, eventually joins the neighbor with
the highest RSSI value and become part of a subtree. If no
links are detected with RSSI above the threshold value, the
protocol has a fall back mechanism to select links with lower
RSSI values.
V. PERFORMANCE EVALUATION
In this section, we evaluate the network performance of
RMMTC by simulations and by experiments on a real testbed.
A. Simulation study
We implemented our proposed RMMTC protocol in NS2
discrete event simulator to study the performance of our
scheme in large networks to investigate its scalability. We
compared its network performance with other related topology
control protocols. The metric used for comparison is the
percentage of successful packet reception (PRR) at the base
station for all the nodes in the topology. Nodes are deployed
randomly in a 100m x 100m target area. Nodes after forming a
connected topology report their sensed data back to a centrally
located base station (at position 50,50) after every 3 seconds.
Note that as the default NS2 implementation does not support
multi-channels, we modified the MAC layer to discard packets
received from the physical layer on all channels except the one
in use at that node.
We assume that external interference from WiFi is present
on all the three commonly used WiFi channels (1, 6 and
11). We also assume that all the nodes in the topology
are influenced by same amount of interference from WiFi
transmissions. Thus the value of Rthresh is assumed as -75dBm for all the nodes. ZigBee transmissions with RSSI values
lower than -75 dBm experience packet drops probability @
10% for each 2 dBm decrease in RSSI values below -75 dBm
with maximum packet drop probability of 50% for all RSSI
values lower than -85 dBm. This packet drop policy has been
implemented at the modified MAC layer in NS2.We compared the performance of RMMTC with central-
ized ILP assignment and two other distributed schemes. For
centralized ILP topology assignment, we used MATLAB to
discover links with lowest interference values (Ie) to form
multi-channel routing subtrees. The topology is then providedto NS2 with parent information as static routes to evaluate its
network performance. The first distributed scheme included in
our simulation study is the hop count based topology formation
(HopCount) based on distance vector routing [12]. In this
protocol, topology is setup and routes are established based on
the received hop count values in the route broadcast messages.
The same topology is used for two simulation scenarios -
No WiFi and then subjected to WiFi interference (HopCount
with WiFi). The second distributed scheme included in our
comparison is the RSSI based topology formation which, is
part of the RMMTC protocol but only uses a single channel
(channel 13). Links in RSSI based scheme are only selected
if the RSSI values are above the Rthresh value. Note that thisRSSI based assignment takes care of external interference fromWiFi but does not take any measures to combat co-channel
interference and channel contentions. Each set of experiment
was repeated five times for different number of nodes.
80 100 120 14040
50
60
70
80
90
100
Number of Nodes
%agePacketReceptionRate
HopCount no WiFi
HopCount with WiFi
RSSI with WiFi
RMMTC with WiFi
Centralized ILP with WiFi
Fig. 7. Network Performance of RMMTC Protocol
Topology formed by HopCount, RMMTC and centralized
ILP solution are shown in Figure 6. Simulation results in
Figure 7 show that for HopCount PRR falls substantially
in the presence of WiFi interference. For 80 nodes, PRR
is reduced to about 69% from an average of about 98%
when no WiFi is present. Similarly, for 140 nodes topologies,
PRR reduces to 55% from about 94%. RSSI based topology
165
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
7/8
improves the packet reception rate by selecting links with
better RSSI values. However, PRR is still lower than the
HopCount - No WiFi scenario. The PRR is further improved
by RMMTC by forming different routing trees based at the
base station operating on non-interfering channels with RSSI
based links selection. For 140 nodes topology, the RMMTC
protocol reduces the percentage of dropped packets from 45%
to less than 5%, a nine-fold reduction.
Comparing the performance of RMMTC with centralizedILP assignment, RMMTC performs close to the ILP, especially
for dense topologies. In dense topologies, it is easier to find
better RSSI based links and to assign different channels to
routing trees due to presence of more neighbors.
20 40 60 80 100 120 1400
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF (No of Lost Packets)
HopCount No WiFi
HopCount With WiFi
RSSI based With WiFi
RMMTC With WiFiCentralized ILP With WiFi
Fig. 8. Empirical CDF for 120 Nodes Topology
Figure 8 displays the empirical CDF for number of lost
packets in a 120 node topology for the schemes under compar-
ison. Result shows that for HopCount - with WiFi almost 70%
of the nodes lost upto 32% of their packets (46 packets out of
140). RSSI-with WiFi improves the distribution (70% of the
nodes now lost less than 10% of their packets). RMMTC has
network performance closest to the centralized ILP assignment
as shown by the CDF. For RMMTC, 98% of the nodes lost
less than 9% of their packets.
B. Experimental Evaluation
Fig. 9. 12 Nodes Test Bed Topology
We conducted experiments to verify the operation and per-
formance of the RMMTC protocol in a real world setting. We
used standard off-the-shelf Crossbow MicaZ sensor nodes for
the experiments. These experiments were performed indoors
in an environment where multiple WiFi Access Points (APs)
were active on the commonly used WiFi channels (channels 1,
6 and 11) We also created WiFi interference through a laptop
running a FTP client on WiFi channel 1. There was another
WiFi AP active on channel 1. Channel 6 and 11, on the other
HopCount MintRoute RSSI RMMTC
10
20
30
40
50
60
70
80
90
100
%agePacketReceptionRate
Experimental Study: Comparison of Multi Hop Protocols
Min
Max
Fig. 10. Comparison of Multi-Hop Protocols
hand, had 1 and 3 active APs respectively at the time the
experiments were conducted.
We set up a 12 node indoor topology in a 2m wide L
shaped corridor where nodes are placed on inverted foam
glasses raised about 10 cm from the ground (See Figure 9).
The node transmission power level was set at -15 dBm to
ensure that multi-hop communication is required between the
nodes. Nodes send a packet every 0.5 sec to the base station.
These packets are carried over multiple hops to the base station
where the received packet is logged with the sender ID andits corresponding sequence number.
For comparison purposes, we implemented three of the
protocols, used in the simulation study, in TinyOS (namely;
HopCount, RSSI and RMMTC). In addition, we also included
the widely used MintRoute multi-hop routing protocol [13]
(implementation available in TinyOS 1.x distribution). In
this protocol, routes are setup and maintained depending on
the link quality estimations by snooping the received signal
strength and packet losses etc.
We compared the individual as well as cumulative Packet
Reception Rate (PRR) achieved for all the nodes using differ-
ent multi-hop topology control schemes. The default channel
used for all protocols is ZigBee channel 13 (chosen to beunder the WiFi influence of FTP client operating on Channel
1) except for RMMTC for which the default control channel
is 26 and channel 13 and 21 are used for data transfer phase.
Default MAC for MicaZ was used for all the experiments.
Rthresh was set at -82 dBm.Figure 10 shows the maximum and minimum PRR for
the compared schemes. The overall reception rate for the
HopCount protocol fluctuates between a maximum of about
68% to minimum value of about 47%. This can be attributed
to the fact that no consideration has been given to the potential
interferences during the topology construction. MintRoute
protocol and RSSI based protocol both show improvement in
the overall packet reception rate with values ranging between
79% and 66% for MintRoute and 84% and 62% for RSSI
respectively. This performance improvement is due to the
fact that both these protocols consider the link performance
parameters while selecting the link for topology construc-
tion/maintenance. This shows that even simple consideration
of RSSI values during topology construction can considerably
improve the network performance. In addition to reducing the
WiFi interference by selecting higher RSSI values, RMMTC
creates multiple collision domains (2 in this case) to con-
siderably reduce channel contention and improve the average
166
7/27/2019 Mitigating the Effect of Interference in Wireless Sensor Networks
8/8
1 2 3 4 5 6 7 8 9 10 11 12
10
20
30
40
50
60
70
80
90
100
Nodes
%agePacketReceptionRate
HopCount
(a)
1 2 3 4 5 6 7 8 9 10 11 12 13
10
20
30
40
50
60
70
80
90
100
Nodes
%agePacketReceptionRate
MintRoute
(b)
1 2 3 4 5 6 7 8 9 10 11 12 13
10
20
30
40
50
60
70
80
90
100
Nodes
%agePacketReceptionRate
RSSI
(c)
Fig. 11. Successful Packet Reception Rate for HopCount, MintRoute and RSSI based Protocol
2 4 5 7 9 11 12
10
20
30
40
50
60
70
80
90
100
Nodes
%agePacketReceptionRate
RMMTC (Channel 13)
1 3 6 8 10
10
20
30
40
50
60
70
80
90
100
Nodes
%agePacketReceptionRate
RMMTC (Channel 21)
Fig. 12. Packet Reception Rate for RMMTC
packet delivery rate to above 85%.
Figures 11 and 12 show the individual average PRR for all
the nodes in the topology. As expected, nodes located near
the base station shows better PRR than those further away.
Note that routing subtree working on channel 13 shows more
packet losses as compared with that on channel 21, due to the
WiFi interference from the FTP file transfer. This experimental
study has highlighted that the effect of WiFi interference can
be mitigated by selecting short, multi-hop links with adequate
RSSI values as compared to long distance transmissions.
Similarly, reducing the interference by incorporating multiplecollision domains based on multiple channels considerably
improves the performance.
VI . CONCLUSION
This work shows that the use of multiple channels and links
with adequate RSSI values can mitigate the effect of both ex-
ternal and internal interferences. We formulated an ILP model
for describing the multiple channel assignment to overcome
the effects of interferences. We proposed a fully distributed
multi-channel topology control protocol that is shown to work
close to the centralized optimal ILP solution. The performance
evaluation is done by carrying out discrete event simulations
and verified by deployment in a real test bed. The results showthat our proposed protocol can successfully achieve acceptable
network performance, more for dense networks, when exposed
to interferences.
REFERENCES
[1] S.Y. Shin, H. S., Park, S., Choi, and W. H., Kwon, Packet Error RateAnalysis of IEEE 802.15.4 under IEEE 802.11b Interference, 2005:Wired/Wireless Internet Communications (WWIC05).
[2] O.D. Incel, S. Dulman, P. Jansen, and S. Mullender, Multi-ChannelInterference Measurements for Wireless Sensor Networks, 2006:LCN,pages 694-701.
[3] R. Musaoiu-E. and A. Terzis, Minimizing the Effect of WiFi Interferencein 802.15.4 Wireless Sensor Networks, 2008: Int. J. Sensor Networks,Vol. 3, No. 1, pages 4354.
[4] K. Srinivasan, P. Dutta, A. Tavakoli, and P. Levis, An Empirical Studyof Low-power Wireless, 2010: ACM TOSN, pages 1-49.
[5] Y. Wu, J. A. Stankovic, T. He, J. Lu and S. Lin, Realistic and EfficientMulti-Channel Communications in Wireless Sensor Networks, 2008:Infocom, pages 1193-1201.
[6] Crossbow Technol ogy Inc., Avoi ding RF interferencebetween WiFi and Zigbee, 2004: Available at: http ://www.xbow.com/products/Pr oduct pdf files/W ireless pdf/ZigB eeandWiF iInterference.pdf.
[7] G. Zhou, C. Huang, T. Yan, T. He and J. A. Stankovic, MMSN: Multi
Frequency Media Access Control for Wireless Sensor Networks, 2006:IEEE INFOCOM.
[8] K. R. Chowdhury, N. Nandiraju, D. Cavalcanti, and D. P. Agrawal,CMAC: A Multi-channel Energy Efficient MAC for Wireless SensorNetworks,2006: IEEE WCNC.
[9] M.D. Jovanovic and G.L. Djordjevic, TFMAC: Multi-channel MACProtcocol for Wireless Sensor Network, 2007:TELSIKS07.
[10] Y. Kim, H. Shin and H. Cha. Y-MAC: An Energy-efficient Multi-channel MAC Protocol for Dense Wireless Sensor Networks. 2008:IPSN08.
[11] N. Ahmed, S. Kanhere and S. Jha, Multi-Channel Interference inWireless Sensor Networks, 2009: IPSN, pages 367-368.
[12] C. E. Perkins and P. Bhagwat, Highly Dynamic Destination SequencedDistance Vector Routing (DSDV) for Mobile Computers, 1994: ACMComputer Communication Review, pages 234-244.
[13] A. Woo, T. Tang and D. Culler, Taming the Underlying Challenges ofReliable Multi-hop Routing in Sensor Networks, 2003: SenSys, pages14-27.
[14] K. Srinivasan, and P. Levis, RSSI is under appreciated, 2006: Em-Nets06.
[15] J. Hauer, V. Handziski and A. Wolisz, Experimental Study of the Impactof WLAN Interference on IEEE 802.15.4 Body Area Networks, 2009:EWSN, pages 17-32.
[16] W. Xu, W. Trappe and Y. Zhang, Channel Surfing: Defending WirelessSensor Networks from Interference, 2007:ACM IPSN07, pages 499-508.
[17] C. A. Boano, Z. He, Y. Li, T. Voigt, M. Zuniga and A. Willing,Controllable Radio Interference for Experimental and Testing Purposesin Wireless Sensor Networks, 2009: SenseApp, pages 865-872.
167
Top Related