Mitigating the Effect of Interference in Wireless Sensor Networks

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


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


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


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


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


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


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


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


7/8

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


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


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


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


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


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


RMMTC (Channel 13)

1 3 6 8 10

10

20

30

40

50

60

70

80

90

100

Nodes


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

Mitigating the Effect of Interference in Wireless Sensor Networks

Documents

Transcript of Mitigating the Effect of Interference in Wireless Sensor Networks