Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

www.ietdl.

org
Published in IET Wireless Sensor Systems
Received on 21st April 2011
Revised on 4th September 2011
doi: 10.1049/iet-wss.2011.0069

ISSN 2043-6386

Co-operative routing for wireless sensor networks


using network coding
R.R. Rout1 S.K. Ghosh2 S. Chakrabarti1
1
G.S. Sanyal School of Telecommunications, Indian Institute of Technology, Kharagpur, West Bengal, India
2
School of Information Technology, Indian Institute of Technology, Kharagpur, West Bengal, India
E-mail: rashmirsays@gmail.com; skg@iitkgp.ac.in; saswat@ece.iitkgp.ernet.in

Abstract: Omni-directional two-way traffic flow using flooding leads to broadcast storm problem in a wireless sensor network
(WSN). The probabilistic routing protocols help in reducing the broadcast storm but lack reliability. Network coding is a new
paradigm that allows an intermediate node to encode incoming packets which improves the bandwidth efficiency and
reliability of the WSN. A network coding-based probabilistic routing (NCPR) scheme has been proposed, which is energy-
efficient, reliable and alleviates the broadcast storm problem in a clustered WSN. The focus of this work is to design the
network coding-based procedures for probabilistic routing protocols. In NCPR, a sensor node initialises a transmission
process in a WSN cluster by transmitting a packet from its sensed queue. Each neighbour node encodes its received packet
with its sensed packet using XOR network coding and transmits the coded packet with certain probability. Three network
coding-based procedures have been proposed for encoding and decoding of packets in intra-cluster and inter-cluster
communications by distributing roles among the sensor nodes. It has been shown that the NCPR scheme provides better
energy efficiency and reliability compared to the probabilistic routing and pure flooding schemes.

1 Introduction data packets to the next node. In comparison to the


traditional store and forward mechanism, the network
In recent years, the usage of wireless sensor network (WSN) coding technique improves the bandwidth efficiency and
is becoming popular as such networks can be deployed thus decreases the energy consumption [2, 5, 7].
for distributed monitoring, like environmental monitoring, In a wireless network, the wireless link status changes
seismic monitoring, healthcare assistance, military dynamically due to interference. Therefore the network is
applications [1] and so on. Traditional communication prone to packet loss. To mitigate the effect of such
paradigm in WSNs are many-to-one [2], one-to-many and problems, Al-Kofahi and Kamal [2] have proposed a
many-to-many. The many-to-one traffic flows from sensor scheme on survivability against link failure using network
nodes to sink node and the one-to-many traffic flows from coding. The reliability gain of network coding against link
sink node to sensor nodes. In many-to-many paradigm, two- failure and path failure has also been studied by Aly and
way traffic flows between sensor nodes. Usually, there is a Kamal [8] and Bahramgiri and Lahouti [9], respectively.
considerable amount of omni-directional redundant traffic Moreover, network coding has emerged as an efficient
flow in many-to-many paradigm. Even though the flooding technique [10 – 14] to tackle the broadcasting capabilities in
of data packets simplifies the working of WSN, it leads to a network. In a WSN, the packet loss by the intermediate
broadcast storm problem for many-to-many communication. sensor nodes may lead to a situation where the sink node
Probabilistic routing schemes [3, 4] are used to restrict the may not receive sufficient packets to decode. Thus the sink
flow of redundant messages with small routing overhead in node may fail to take appropriate global decisions. For a
WSN. However, these probabilistic schemes lack reliability cluster-based WSN, to take local decisions distributively,
due to the absence of request to send – clear to send (RTS – the intermediate sensor nodes should decode the network-
CTS) and acknowledgment mechanisms. Therefore alternate coded packets by hearing its neighbours from the same
suitable energy-efficient mechanisms are required to protect cluster or from the neighbour clusters.
the network from link failures (due to collision, contention This paper proposes a new routing scheme for clustered
etc.). WSN, incorporating an efficient form of network coding. In
The network coding technique is the basis for many the proposed scheme the intermediate sensor nodes as
bandwidth and energy-efficient transmission schemes in well as the sink node can encode and decode the packets.
wireless networks [5, 6]. With network coding, the This scenario is applicable for mission critical applications,
intermediate nodes of a network can appropriately encode such as battle field, where reliability is a major issue. For
the incoming data packets before transmitting the coded such many-to-many traffic flows, the clusters can share

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 75


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012
www.ietdl.org
information among themselves and take decisions locally Each node which wants to transmit the network coded data
without waiting for the sink node each time. The proposed computes a similar linear combination as given in (1). The
scheme helps in protecting the WSN against link failures. encoded packets are transmitted through the network with
Nguyen et al. [15] defined the network coding gain for the n coefficients. The encoding vector is used in the
single-hop communication. This definition of network receiver side to decode the data. In this paper, the XOR
coding gain, however, needs modification for multi-hop network coding is used. It is a special case of linear
communication in WSN. The scheme improves the network coding [16]. The encoded packets that are
network coding gain which is expressed in terms of transmitted in the network are elements in GF(2) ¼ {0, 1}.
transmission coding gain and reception coding gain. The In this case, bitwise XOR in GF(2) is used as an operation.
transmission coding gain is defined as the ratio of the
native data packets transmitted with network coding to the 1.1.2 Decoding process: A receiver node has to solve a
number of data packets transmitted without network coding set of linear equations to retrieve the original packets. The
using the same number of transmissions in the WSN. encoding vector q is received by the recipients with the
Furthermore, the reception coding gain is defined as the encoded data. Let a node has received a set (q 1, Y 1), . . . , (q m,
ratio of the native data packets received with network coding Y m). The notations Y j and q j denote the information symbol
to the number of data packets received without network and the coding vector for the jth received packet, respectively.
coding using the same number of receptions in the network. The following form of linear system (2) with m equations and
The major contributions of this present work can be n unknowns need to be solved for decoding operation.
summarised as follows:

n
† Three network-coding-based encoding and decoding Yj = qji Gi , j = 1, . . . , m (2)
procedures are designed for a clustered WSN by distributing i=1
functionalities among the sensor nodes. The encoding and
decoding procedures assist the sensor nodes to take decisions At least n linearly independent packets must be received by the
locally without waiting for global decisions from the sink recipients to decode the original native packets. Here, the only
node. unknown is Gi which contains the original packets transmitted
† Development of a network-coding-based probabilistic in the network. The n number of original packets can be
routing scheme by incorporating the customised network decoded by solving the linear system in (2) after getting n
coding procedures in probabilistic routing. The proposed linearly independent packets.
scheme exhibits better performance in terms of network
coding gain, energy efficiency and reliability. 1.1.3 Example of XOR network coding: An example of
† A detailed theoretical analysis and simulation have been XOR network coding is shown in Fig. 1. Three nodes such
carried out for the proposed network-coding-based as S, R and D are involved in communication. The nodes S
probabilistic routing (NCPR) scheme. and D are not within the transmission range of each other
and thus communicate among themselves using the relay
1.1 Network coding node R. The relay node R receives the individual packets
from the nodes S and D, as shown in Fig. 1. The node R
Network coding is a technique which allows an intermediate
relays network-coded packet (s ⊕ d ) to nodes S and D
node of a network to encode the information received from its
using only one transmission. The node S decodes the data
neighbouring nodes. Let p1 , p2 , . . . , pn denote the n number
transmitted by the node D (i.e. (s ⊕ (s ⊕ d ) ¼ d )).
of sensed packets by same or different source nodes. These
Similarly, the node D can decode the data transmitted by
packets are transmitted by the source nodes and received by
the node S. In the absence of network coding, the relay
a neighbouring encoder sensor node. Let the maximum
node has to relay both the data s and d separately using
number of packets that can be encoded by an encoder node
two transmissions. Thus, network coding improves the
into one packet at a time be w. So, n/w number of encoded
bandwidth efficiency in a wireless network.
packets are generated by an encoder. A header which
The rest of this paper is organised as follows. Section 2
contains some bits for a coding vector and some bits to
presents the related work and motivation behind the work.
specify the source or generator sensor node of the packet
In Section 3, the proposed scheme has been presented and
may be associated with each encoded packet. Generalised
three network-coding procedures have been described. The
encoding and decoding processes of linear network coding
simulation results and discussions are presented in Section
[16] are described below.
4. Finally, the concluding remarks are given in Section 5.
1.1.1 Encoding process: The coded packets transmitted
through a network, based on linear network coding [16], are 2 Related work
viewed as elements in finite field GF(2s), where 2s is the This work presents a co-operative routing scheme using
size of the finite field. When a node wants to transmit XOR network coding. The information theoretic aspect of
network coded data, it first chooses a sequence of network coding was introduced by Ahlswede et al. [5]. Lun
coefficients q ¼ (q1 , q2 , . . . , qn), called encoding vector, et al. [13] have proposed their scheme for random linear
from GF(2s). A set of n packets Gi (i ¼ 1, 2, 3, 4, . . . , n)
arriving at the node are linearly encoded into a single
output packet. An output encoded packet is generated by
computing the linear combination


n
Y = qi Gi , qi [ GF(2s ) (1)
i=1 Fig. 1 XOR network coding

76 IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75– 85


& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-wss.2011.0069
www.ietdl.org
network coding that achieves packet-level capacity for both dissemination scenario in WSN using the probabilistic
single unicast and single multicast connections in wireless routing scheme. A source sensor node (S) transmits a packet,
networks. Li et al. [17] have proposed a method to say s, with probability p ¼ 1. When intermediate nodes
incorporate network coding into a non-coding-based (such as N1 , N2 and N3 in Fig. 2) first receive the packet,
localised algorithm that tries to optimise the coding gains they rebroadcasts the packet to its neighbours with
given a set of native packets and the subset of packets each probability p ¼ k (where k ¼ 0.1, 0.2, . . . , 1). The packet is
neighbour receives. Furthermore, a network coding scheme discarded with probability (1 2 p) by the intermediate nodes.
for mobile ad hoc network (MANET) using directional In case, a node receives same packet again, it is discarded.
antenna has been proposed by Yang and Wu [18]. Widmer So, each node rebroadcasts or forwards a given data packet
et al. [19] have proposed an energy-efficient broadcasting in at most once. The packet is disseminated throughout the
wireless ad hoc networks. network and ultimately reached at the destination sink node D.
A basic scheme that uses XOR of packets for practical There exists a major gap in implementing network coding
network coding has been proposed by Katti et al. [20]. for many-to-many communication paradigm in WSN. This
They have shown a significant bandwidth improvement in paper presents a probabilistic routing scheme using network
wireless mesh network. However, Keller et al. [11] have coding with XOR-based encoding – decoding procedures in
designed a communication protocol that obtains the trade- WSN. While most of the prior work assumes the analysis
off between energy efficiency and reliability in WSN. of network traffic for either many-to-one or one-to-many
Unlike existing approaches, the focus of this paper is on the communication, a many-to-many paradigm of two-way
analysis of the reliable wireless broadcast problem in a traffic flow is considered in this paper. It has also been
single-hop (intra-cluster communication) as well as multi- shown in this paper that despite the topological changes due
hop (inter-cluster communication) in WSN, where energy to link failure, the WSN performance improves.
consumption is an important constraint.
A model for single-hop communication applicable to WiMax 3 NCPR scheme
has been proposed by Nguyen et al. [15]. They employed XOR
network-coding scheme for one sender to multiple receivers. This section deals with a network model which describes the
Unlike Nguyen et al., a multiple sender and multiple receiver- communication paradigm in a cluster-based many-to-many
assisted system model is considered in this paper. Rout WSN. The proposed scheme along with the three network-
et al. [21] have proposed a recursive topology construction coding procedures are presented in this section. The issues
algorithm over a randomly deployed WSN. They have shown such as reliability, network coding gain and energy
that assigning different responsibilities to different sensor consumption of the WSN (using NCPR) are discussed. The
nodes in the recursive topology boost the network-coding symbols used in rest of this paper are given in Table 1.
benefits.
The broadcast storm problem due to simple flooding 3.1 Network model
of messages can be eradicated using probabilistic routing
protocols [3, 4, 22]. Ni et al. [22] have proposed a A WSN is represented as a graph G(V, E), where V is the set
probabilistic rebroadcasting scheme to mitigate the broadcast of nodes and E is the set of links in a network. The WSN is
storm problem in MANET. However, the simple probabilistic divided into several cluster area such as Cluster 1, Cluster
routing protocols are vulnerable to packet loss due to node 2, . . . , Cluster N (note: cluster and cluster area are having
failure or link failure. So, there is a necessity of some suitable the same meaning). The nodes, say r, which are within the
broadcast technique which can use probabilistic routing transmission range of each other form a cluster (clique).
protocols and can provide reliability against link failure or The cluster area is assumed to be circular. There are two
node failure. cases: (a) the clusters are overlapped and (b) the clusters are
A basic probabilistic routing scheme (GOSSIP1) has been not overlapped. In both the cases, nodes in neighbouring
proposed by Haas et al. [3]. Fig. 2 shows a packet clusters can communicate with each other. Moreover,
sensor nodes belonging to the boundary or edge region of a
cluster can overhear packets from neighbour clusters. Let g
be the number of sensor nodes that belong to a particular
cluster which can communicate with a neighbour cluster.
Therefore (r 2 g) number of sensor nodes cannot
communicate directly with the neighbour cluster. They can
communicate using g sensor nodes as relay nodes.
Let the probability of retransmission is denoted as p. When
the value of p ¼ 1, the probabilistic routing protocol [3]
behaves as pure flooding. Furthermore, each node can
identify its own location information by global positioning
system (GPS) [23]. However, each sensor node can also use
any other localisation schemes [24, 25] to detect whether
the node is at the edge region or at the intermediate region
of a cluster area. Furthermore, each sensor node can know
its direct neighbours by exchanging control messages.

3.2 NCPR scheme

A sensor node initialises the transmission process by


Fig. 2 Transmission method in probabilistic routing (GOSSIP1 by transmitting a native (sensed) data packet in a cluster. The
Haas et al. [3]) neighbour sensor nodes receive the native packet. The

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 77


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012
www.ietdl.org
Table 1 List of symbols used in this paper

Symbol Description of the symbol

r number of sensor nodes in a cluster (clique)


p probability of retransmission/routing probability
x native packet (sensed/non-coded packet) length in bits
g number of edge or relay sensor nodes in a cluster
Pr packet reception ratio (ratio of packets received with error to the packets received without error)
Econ energy consumption (in nAh)
Etx energy consumption for each transmission (in nAh)
et energy consumption for each reception (in nAh)
Pi ith received packet in the input (received) queue of a sensor node
Pj jth sensed packet in the input (sensed) queue of a sensor node
f number of data packets lost during transmission in the network
d transmission distance in meters
h overhead of network coding in bits
H1 number of bits present in the header of a packet in probabilistic routing
H2(i) number of header (variable length) bits in ith (encoded) packet transmission in NCPR scheme
H2 number of header (fixed length) bits in a network coded packet in NCPR scheme
T total number of transmissions in the WSN
W total number of receptions in the WSN
M1 number of bits transmitted in probabilistic routing with header H1 bits in one transmission
L1 number of bits received in probabilistic routing with header H1 bits in one reception
M2(i) number of native bits transmitted in the ith (encoded) packet transmission in NCPR scheme with header
H2(i) bits (with variable length header)
L2(i) number of native bits received in the ith (encoded) packet reception in NCPR scheme with header H2(i) bits
(with variable length header)
M2 number of native bits transmitted in NCPR scheme with fixed length header H2 bits
L2 number of native bits received in NCPR scheme with fixed length header H2 bits

neighbour sensor nodes encode (XOR) its own native packet of packets of length x are encoded from the same source or
with the received packet and retransmit the coded packet with from different sources. v is 1 for XOR network coding. To
routing probability p. Each active sensor node transmits once restrict the overhead in the packet header, the r-coded packets
before a second round of transmission starts. Let there are g need to be restricted. Since in NCPR scheme a cluster is
edge sensor nodes that act as relay nodes in a cluster. So, restricted to a clique, the value of r reduces. It helps us to
(r 2 g) number of sensor nodes share their information with design a generalised XOR coding scheme for a cluster-based
each other. In the NCPR scheme, packet loss at the receiver WSN with even and odd number of intermediate sensor nodes
sensor nodes is independent and uncorrelated (i.e. follows a which is presented later in this section.
Bernoulli trial). A receiver sensor node requests for
a packet which is not received successfully by sending a
negative acknowledgment packet. However, the receiver 3.4 Responsibility of sensor nodes
node waits till the completion of one round of
retransmissions by its neighbours. The node receives coded Inter-cluster and intra-cluster communications are shown in
packets from all other nodes and it decodes the lost packet Figs. 3a and b, respectively. Fig. 3b also shows a scenario
without requesting for a fresh retransmission. This approach of communication between two neighbouring clusters such
improves the bandwidth efficiency in the network. as Cluster 1 and Cluster 2 using edge or relay sensor node.
However, if after one round of retransmission the receiver In NCPR scheme, sensor nodes are differentiated into three
node does not get enough packets to decode, it requests the categories depending on their functionalities at various
sender to retransmit the lost packets. If a native packet at locations in a cluster: Type 1, Type 2 and Type 3 [26].
the relay node in a cluster is lost, the node requests for a Operations of all the three types of sensor nodes are shown
fresh retransmission. in Fig. 4. Type 1 (node S in Fig. 3b) sensor node
broadcasts sensed packets to the sensor nodes within its
range. Type 2 (nodes A and B in Fig. 3b) sensor nodes
3.3 Model for network coding inside a cluster perform network coding and broadcast the
encoded packets. The sensor nodes which are in its range
In NCPR scheme, the XOR network coding is used. Length receive the encoded packets. Type 1 and Type 2
of an encoded packet (i.e. A(x)) in NCPR scheme is given by intermediate sensor nodes interchange their role from time
to time inside a cluster (refer flowchart in Fig. 5). Pi is the
A(x) = vx + h (3) ith received packet that is stored in the input queue of an
intermediate sensor node (Fig. 4a). Pi is ready for encoding
where x is the length (in bits) of a sensed native packet, h is the operation. Pj is the jth sensed packet which is also ready for
length of the header (in bits) for network coding and v is a encoding operation in an intermediate sensor node
positive integer constant. The header contains an encoding (Fig. 4a). Type 3 sensor nodes (e.g. node C in Fig. 3b) are
vector as well as some bits to identify the sensor nodes whose the edge or boundary nodes (in Fig. 3a) which act as the
packets are encoded. The overhead increases as more number relay nodes for inter-cluster communication.

78 IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75– 85


& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-wss.2011.0069
www.ietdl.org

Fig. 3 Inter-cluster and intra-cluster communications


a Inter-cluster communication
b Intra-cluster communication

Fig. 4 Operations of different sensor nodes


a Type 1 and Type 2 intermediate sensor nodes
b Type 3 edge or relay sensor nodes

3.4.1 Role of packet transmitting and packet flowcharts (refer Figs. 5 – 7). As shown in Fig. 5, whenever
receiving nodes: Packet transmitting nodes are the the role of a sensor node is Type 1, the node keeps the
nodes which transmit packets in a network during a previously transmitted native packet in buffer (i.e. Pj21).
particular duration of time. Packet receiving nodes are the The current sensed packet is stored in Pj . The native packet
nodes which receive packets in a network during a Pj is transmitted if Pj is significantly different than Pj21 ,
particular duration of time. Let S transmits a native packet otherwise, the transmission is stopped. However, when the
with p ¼ 1 in the scenario as shown in Fig. 3b. The nodes role of a node is Type 2, transmission of encoded packet
A, B and C are the neighbours of S and receive the has occurred as shown in Fig. 5. The edge or relay (Type
packet. The node A encodes its sensed packet with the 3) sensor nodes receives native and encoded packets in the
received one. Node A transmits the encoded packet which cluster. The Type 3 node picks a packet and checks
is further received by the nodes S, B, C and D as shown whether it is an encoded packet or not as shown in Fig. 6.
in Fig. 3b. So, the total number of transmitting nodes is Whenever the packet Pi is a native data packet, the content
two and the total number of receiving nodes is seven after of the packet is stored in a buffer. If Pi is an encoded
two transmissions. Moreover, when the number of packet then the decoding of encoded packets takes place by
retransmissions in the network increases, the total number the edge or relay sensor node as depicted in Fig. 6. The
of nodes involved in the process of transmission and decoded packets are stored in Pk which are used for the
reception also increases (i.e. there is an increase in the encoding process at the relay node. The encoding process
number of packet transmitting and packet receiving nodes occurs by the Type 3 nodes after taking decoded packets
in the network). from Pk as shown in Fig. 7. P0 contains the received native
packet. Pc stores the encoded packet after the encoding
3.5 Working principle of NCPR process. If the number of intermediate sensor nodes (i.e. K )
is odd, the native data packet P0 is encoded (XOR) with Pc ,
The proposed encoding and decoding procedures (processes) otherwise the encoded packet is constructed without P0
for different types of sensor nodes are shown using three (refer Fig. 7).

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 79


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012
www.ietdl.org

Fig. 7 Encoding procedure (process) for an edge or relay sensor


node
Fig. 5 Network coding operation (procedure) for an intermediate
sensor node other, and each node has equal transmission radius. The
starting node S broadcasts the sensed data, say s, which is
received by sensor nodes A, B and C. The node A senses
its own data a and it broadcasts the network coded packet
(a ⊕ s) with probability p. The information (a ⊕ s) is
received by the sensor nodes S, B, C and D. Similarly, the
node B senses its own information b and it broadcasts the
network coded packet (b ⊕ s) (case i) or it can also
broadcast the network coded packet (a ⊕ b) (case ii). In the
case i, the information (b ⊕ s) is received by the nodes
which are in its range (node S, A, C, D). Furthermore, in
case ii the information (a ⊕ b) is received by the nodes
which are in its range. The received packets for various
nodes are shown in Tables 2 and 3 for case i and case ii,
respectively (Rp means received packets and Sp means
sensed packets in both the tables). These tables also contain
the received packets by an edge sensor node (D) that
belongs to a neighbour cluster.

Table 2 Received packets by the nodes in case i

Node S Node A Node B Node C Node D

Rp a⊕s s s s a⊕s
b⊕s b⊕s a⊕s a⊕s b⊕s
a⊕b⊕s a⊕b⊕s a⊕b⊕s b⊕s a⊕b⊕s
Sp s a b – –

Table 3 Received packets by the nodes in case ii

Node S Node A Node B Node C Node D


Fig. 6 Decoding procedure (process) for an edge or relay sensor node
Rp a⊕s s s s a⊕s
a⊕b a⊕b a⊕s a⊕s a⊕b
Let m, n and p be the number of Type 1, Type 2 and Type 3 a⊕b⊕s a⊕b⊕s a⊕b⊕s a⊕b a⊕b⊕s
nodes, respectively. As shown in Fig. 3b, (m ¼ 1, n ¼ 2, Sp s a b – –
p ¼ 1), the nodes S, A, B, C are within the range of each

80 IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75– 85


& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-wss.2011.0069
www.ietdl.org
Node C is an edge or boundary sensor node which receives chosen as a cluster head as all nodes have equal
the network coded packets from the nodes A and B. It also computational resources. In NCPR scheme, the edge sensor
receives native data packet from the node S. Node C nodes switched off its sensing unit and involve in relaying
decodes the received information and broadcasts the encoded packets for intra-cluster communication and inter-
encoded packet (a ⊕ b ⊕ s). Node D belongs to another cluster communication. One of the edge sensor nodes out of
cluster that receives the broadcast data from C. The node D g edge nodes takes the role of cluster head for a particular
also receives (a ⊕ s) and (b ⊕ s) in the case i. It receives duration. Another edge sensor node gets the role of cluster
(a ⊕ s) and (a ⊕ b) in case ii. These set of received packets head in the next duration. This process continues among
are sufficient to decode the data packets. There are at least edge sensor nodes in rotation basis. Furthermore, the cluster
three transmissions required so that the nodes S, A, B can head communicates between neighbouring clusters using a
share data packets within each other using network coding. single hop and communicates with the sink nodes using
With one more transmission by the node C, the node D multi-hop. After getting the data packets from the cluster
which belongs to a different cluster can able to know the heads, the sink node takes global decisions. Local decisions
information sent by the sensor nodes S, A and B. The are taken by the cluster heads as well as the other sensor
nodes inside the cluster also receive (a ⊕ b ⊕ s) which is nodes inside the cluster. Especially, the cluster heads take
broadcasted by C. Moreover, the information can be decisions locally by receiving information from its
disseminated from cluster to cluster towards the region of neighbouring clusters.
interest or the sink node R. Furthermore, in NCPR scheme,
the relay node transmits different encoded packets for even 3.7 Reliability
and odd number of intermediate sharing nodes in a cluster
as described by the flowchart given in Fig. 7. For even Reliability against information loss (due to link failure or
number of sharing nodes, the native data s is not within the collision of data packets) is an essential task in WSN [2].
coded packet. But, for the odd number case, the native data Network coding attempts to provide reliability to the
s is encoded within the network coded packet (refer network without retransmission of lost data packets [8].
Table 4). The transmission of edge sensor nodes is critical From Tables 2 and 3, it can be observed that each
for proper decode of data packets in intra-cluster as well intermediate sensor node can give protection against one
as inter-cluster communication. Hence, the probability of link failure. The upper bound of protection against link
retransmission p is set to be high (i.e. p ¼ 1) for the edge failure is (r 2 g). For n round of retransmissions inside the
sensor nodes. cluster, the cluster is protected from O(n(r 2 g)) link failures.
The routing probability p can be set for different sensor Each node can decode the received packets and gets the
nodes depending upon the denseness of the network. lost packets information. For example, there are four sensor
However, the sensor nodes may be in active state or idle nodes that belong to Cluster 1 as shown in Fig. 3b. Nodes
state. If an active sensor node wants to send a native data S, A and B can able to sustain one packet loss each without
packet which is significantly different than the previous requesting for retransmission of the lost packets (note: only
sensed data, it is transmitted. The neighbouring active one edge sensor node ( g ¼ 1) is shown in Cluster 1).
sensor nodes receive the native data and encode their
sensed data with it. The encoded packets are further 3.8 Energy consumption
transmitted with routing probability p. The routing
probability p is crucial in a WSN where a fraction of data is Energy consumption of the NCPR scheme and the
transmitted from a geographical region. The fraction of probabilistic routing scheme is calculated using the existing
information should be enough for predicting the status of model given by Shelby et al. [27]. It has been shown that
the monitoring geographical area. the NCPR scheme restricts the number of retransmissions
Edge or relay sensor nodes (e.g. node D in Fig. 3b) play an against packet loss due to link failure.
important role in the NCPR scheme. The edge sensor nodes in Let Etx be the energy consumption for each transmission
a cluster hear encoded data packets from its neighbouring and et be the energy consumption for each reception.
cluster. It decodes the encoded packets and aggregates Let the total number of transmissions occurred in a WSN is
the native packets. One native packet representing the nt and the total number of receptions is nr . So, the total
information about the neighbouring cluster is sent by the energy consumed by the network is given by
edge sensor node within its cluster. Thus the overall traffic
is minimised in the network when the data are transmitted Econ = nt Etx + nr et (4)
from one cluster area to another and towards the sink node.
The model for radio energy consumption [27] can be given by
3.6 Cluster head selection

Selection of cluster head is done using round-robin technique. Etx = etx + ex d l (5)
In a clique, all the sensor nodes are within the range of each
other which form a cluster. Any sensor node can be randomly where etx is the energy consumption of the transmitter
electronics, ex is the energy consumption of the transmit
Table 4 Relay node encoded packets
amplifier, d is the transmission distance and l is the path
loss exponent which is considered as 2.
Sharing nodes Relay node trans. Nodes involved

2 a S, A, B (relay)
3.8.1 Energy consumption against packet loss:
3 a⊕b⊕s S, A, B, C (relay)
Without network coding, the minimum number of
4 a⊕b⊕c S, A, B, C, D (relay)
transmissions (lower bound) required to share the data packets
5 a⊕b⊕c⊕d⊕s S, A, B, C, D, E (relay)
between r sensor nodes inside a cluster is r. If g nodes in the
boundary are considered as relay nodes then total number of

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 81


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012
www.ietdl.org
transmissions required is (r–g). Let f be the number of data header, the transmission coding gain can be derived from
packets lost during transmissions (i.e. f number of links have (8) as
failed). So the total number of transmissions required is
(r 2 g + f ). From (4) the energy consumed without applying M 2 T − H2 T M 2 − H2
= (9)
network coding is M 1 T − H1 T M 1 − H1

Econ = (r − g + f )Etx + nrw et (6) where H2 is the fixed length header within a network coded
packet, M2T is the total bits transmitted which contains
information symbol bits and header bits using the proposed
where nrw is the number of packet receptions without network scheme, M1T is the total bits transmitted which contains
coding. The number of link failures (i.e. f ) vary from 1 to information symbol bits and header bits in the probabilistic
(r 2 g) as per our consideration. In NCPR scheme, there is no routing scheme. Furthermore, the reception coding gain can
need of retransmissions against f packet loss. If the number of be formulated as
link failures inside a cluster is greater than (r 2 g), then there
is a requirement of retransmissions. Therefore the energy W 
i=1 L2 (i) − Wi=1 H2 (i)
consumption with network coding can be given by (10)
W (L1 − H1 )

Econ = (r − g)Etx + nrn et (7) where W is the total number of receptions in the WSN, L1 is
the number of bits received in probabilistic routing (without
network coding) with header H1 bits in one reception, L2(i)
where nrn is the number of packet receptions with network is the number of native bits received in the ith (encoded)
coding. packet reception in NCPR scheme with header H2(i) bits.
As a special case, with fixed length header, the reception
3.9 Network coding gain coding gain can be derived from (10) as
The retransmitted packets in the proposed NCPR scheme L2 W − H2 W L2 − H2
contain encoded packets. To reduce the packet header = (11)
L1 W − H1 W L1 − H1
length, only 2-coded and 3-coded packets are considered as
shown in Tables 2 and 3. When the number of sensor
nodes increases in the cluster, the r-coded packets can also where L2W is the total bits received which contains
be transmitted by the relay node as shown in Table 4. information symbol bits and header bits using the proposed
Network coding gain is defined by Nguyen et al. [15] for scheme, L1W is the total bits received which contains
single-hop communication. The definition of network information symbol bits and header bits in the probabilistic
coding gain however needs modification in view of multi- routing scheme.
hop communication with the use of coding header. Two Both the transmission coding gain and the reception coding
categories of network coding gain can be achieved: gain represent the improvement of the proposed scheme
transmission coding gain and reception coding gain. The against the probabilistic routing scheme and the pure
transmission coding gain is different than the reception flooding scheme in WSN. In probabilistic routing schemes
coding gain because for each copy of transmission of native [3, 4, 22], each message contains only one packet.
packet, multiple copies of native packets are received in However, in NCPR scheme, the intermediate nodes
WSN. In a generalised scenario of network coding, each retransmit the received packets with a probability p after
node can encode packets from its own sensed data or from encoding their own sensed data with it. Therefore a single
the received data from different neighbouring nodes or from transmission contains multiple packets which leads to high
the both. Therefore the length of the coding header network throughput.
associated in NCPR scheme is dissimilar for each
transmission. Moreover, the packet header associated with 4 Performance evaluation of NCPR scheme
each packet in network coding scheme have two parts: (a)
coding vector which is used to encode packets of a The NCPR scheme is implemented in a MATLAB-based
particular sensor node, (b) some bits that specify the source network simulator known as Prowler [28]. For the
of the encoded native packets. So, the header bits in NCPR experiments, a simple MAC layer protocol is used. In this
scheme may vary for different encoded packets. So as to protocol, a sensor node waits for a random duration of time
calculate the actual native information symbol bits before trying to transmit a packet and then waits for a
transmitted, the total header bits need to be eliminated. The random back off time if the channel is in busy state. The
transmission coding gain can be formulated as sensor node keeps trying until the transmission can be
performed. This protocol consumes less energy than the
T  more sophisticated protocols like 802.11 MAC. The radio
i=1 M2 (i) − Ti=1 H2 (i) propagation model is the free space path loss model where
(8) the signal strength decreases with increase in (d )l and with
T (M1 − H1 )
fading effect. The path loss exponent, l, is considered as
2. Results are averaged after taking 20 simulation runs: both
where T is the total number of transmissions in WSN, M1 is for sensor nodes arranged in linear topology (linear array)
the number of bits transmitted in probabilistic routing and grid topology (grid sensor or sensor nodes in grid). A
(without network coding) with header H1 bits using one simulation parameter table is given in Table 5.
transmission, M2(i) is the number of native bits transmitted Network coding gain of NCPR scheme for the linear array
in the ith (encoded) packet transmission in NCPR scheme is shown in Fig. 8a. For simplicity, all the encoded packets
with header H2(i) bits. As a special case, with fixed length are taken as two-coded packets for the linear array of 20

82 IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75– 85


& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-wss.2011.0069
www.ietdl.org
Table 5 Simulation settings Table 6 Econ per node and packet reception ratio (Pr) using
NCPR scheme
Parameter Settings
p Econ , nAh Pr
routing probabilistic routing (GOSSIP1)[3],
NCPR scheme Sensor nodes Sensor nodes Sensor nodes Sensor nodes
MAC CSMA in linear array in grid in linear array in grid
propagation free space path loss with fading
model 0.1 6.4 8.48 0.09 0.18
path loss 2 0.2 14.3 20.44 0.13 0.28
exponent 0.3 16.2 27.56 0.14 0.51
sensor nodes Berkeley motes (MICA) 0.4 24.6 34.68 0.18 0.60
packet size MICA mote [29] in Prowler 0.5 33.4 39.92 0.30 0.76
nodes (i) 20 nodes arranged in a linear topology 0.6 36.4 44.84 0.34 0.88
(ii) 100 nodes arranged in a 10 × 10 grid 0.7 42.6 46.28 0.41 1.06
data rate 40 Kbps 0.8 48.6 50.90 0.41 1.16
0.9 52.6 54.80 0.47 1.19
1 54.2 59.76 0.54 1.34
sensor nodes. The average number of neighbours is around
four in the linear topology. In the case of pure flooding (i.e.
p ¼ 1), the transmission coding gain and the reception information about the WSN (Fig. 9b). Sink node receives
coding gain are 1 each, because each transmission and 30% of the sensor data from the distributed region. The
reception carries one native packet. In the NCPR scheme, suitable routing probability using NCPR scheme is 0.3
the coding gain varies with the change of routing when the sensor nodes are arranged in the grid topology.
probability ( p). The coding gains are very low as the value The energy consumption model for MICA mote sensors
of p varies from 0.1 to 0.5. However, as the routing [29] is used in this work. As mentioned earlier, two
probability increases from 0.5 to 0.9, both the coding gains different static WSN scenarios are studied: a linear topology
increase. From 0.9 to 1, it remains constant. So, the suitable and a grid topology. Five sensor nodes form a cluster in the
routing probability to get high degree of coding gain is linear topology. In the 10 × 10 grid, nine sensor nodes
from 0.5 to 0.9. However, with higher routing probability, form a cluster. As estimated in [29], 20 nAh energy is
the number of collided messages increase (i.e. Pr increases) consumed for one packet transmission and 8 nAh energy is
which leads to more energy consumption (Table 6) [note: consumed for one packet reception. The per node energy
Pr is defined in Table 1]. Furthermore, with probability of consumptions in NCPR scheme are shown in Table 6. The
retransmission ( p) ≥ 0.7, all the 20 sensor nodes receive average energy consumption of a node in the grid topology
coded packets. This result is shown in Fig. 8b. When is more than a node in linear topology. In the grid, sensor
p ≥ 0.7, 15 sensor nodes transmit their data from the nodes receive more number of packets from more number
respective regions. The network coding gain saturates at of neighbours surrounding it and involve in more number
p ¼ 0.5 (Fig. 8a) and all the sensor nodes have received of redundant transmissions. However, in the linear array of
packets with routing probability p ¼ 0.7 (Fig. 8b). There is sensors, the neighbours are present in two directions only.
a high probability of successful decoding of data packets in
the network when the probability of retransmission is ≥0.7. 4.1 Comparison of results of NCPR with
Hence, in NCPR scheme, the suitable routing probability is probabilistic routing and pure flooding
0.7 when the sensor nodes are arranged in the linear array.
Network coding gains of NCPR scheme for a grid topology In this section, the energy consumption of NCPR scheme is
with 100 sensor nodes are shown in Fig. 9a. The average compared with probabilistic routing (GOSSIP1) [3] and
number of neighbours is eight in the grid. In this case, both pure flooding. In Figs. 10 and 11, the number of received
coding gains are saturated when p ≥ 0.3 (refer Fig. 9a). packets of the NCPR scheme and probabilistic routing
When p ≥ 0.3, almost 99% of the sensor nodes get scheme (GOSSIP1) [3] with same energy consumption are

Fig. 8 Network coding gain of NCPR scheme for the linear array
a Network coding gain against probability of retransmission ( p)
b Number of (packet transmitting and receiving) nodes against probability of retransmission ( p) in a linear array of 20 sensor nodes using NCPR scheme

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 83


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012
www.ietdl.org

Fig. 9 Network coding gains of NCPR scheme for a grid topology with 100 sensor nodes
a Network coding gain against probability of retransmission ( p)
b Number of (packet transmitting and receiving) nodes against probability of retransmission ( p) in a 10 × 10 grid WSN using NCPR scheme

the total energy consumption increases in the network. For


same number of transmissions and same energy
consumption, the number of packets received in the NCPR
scheme is more than that of the probabilistic routing. For
probability p ¼ 1, the probabilistic routing scheme becomes
pure flooding scheme (refer Figs. 10 and 11). It may also
be observed that the number of packets received in the
NCPR scheme is more than that of pure flooding for both
linear and grid topologies.

5 Conclusions
In this paper, an NCPR scheme for cluster-based WSN has
been presented. The proposed NCPR improves the
performance of the network with high network coding gain
and low energy consumption. It has been shown that the
Fig. 10 Number of received packets of the proposed NCPR scheme proposed NCPR scheme could achieve a coding gain up to
and probabilistic routing (i.e. GOSSIP1) by Haas et al. [3] with 1.9 with respect to probabilistic routing scheme. Further,
same energy consumption in a linear array of 20 sensor nodes the NCPR scheme provides better reliability in terms of
protection of data packets against link failures. Every
intermediate sensor node in a cluster protects data packets
against one link failure. The retransmission packets contain
network coded packets rather than the non-coded native
packets which utilise link bandwidth efficiently. Also, the
routing probability of retransmitting nodes can be controlled
depending upon the network topology and number of
sensor nodes in the WSN. Using network coding, both the
intra-cluster and the inter-cluster communication are
discussed. Encoding data packets distributively throughout
the network reduces the number of retransmissions in the
network. It can be observed from the simulation results that
the volume of information transmitted in the NCPR scheme
are significantly more than the probabilistic routing and
pure flooding schemes with the same number of
transmissions and energy consumption. For many-to-many
Fig. 11 Number of received packets of the NCPR scheme and
communication paradigm, suitable values of routing
probabilistic routing (i.e. GOSSIP1) by Haas et al. with same
probabilities of the proposed scheme have been identified
energy consumption in a 10 × 10 grid WSN
to minimise the number of collisions of network coded
packets. The efficacy of the proposed approach has been
shown for two cases (i) sensor nodes arranged in a linear shown through simulation results and theoretical analysis.
topology and (ii) sensor nodes arranged in a grid topology, Future work includes the performance study of NCPR in a
respectively. The total energy consumption in the network dynamic scenario with some mobile sensor nodes in a
for different routing probabilities ( p) are shown along the monitoring area. This dynamic behaviour may lead to the
X-axis (refer Figs. 10 and 11). As the probability of formation of dense and sparse regions. Further, in future,
retransmission increases, the number of transmissions and study of the NCPR scheme for delay sensitive applications
receptions of packets increases in the network. Therefore will be carried out.

84 IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75– 85


& The Institution of Engineering and Technology 2012 doi: 10.1049/iet-wss.2011.0069
www.ietdl.org
6 References 16 Li, S.-Y.R., Yeung, R.W., Cai, N.: ‘Linear network coding’, IEEE
Trans. Inf. Theory, 2003, 49, (2), pp. 371– 381
1 Yick, J., Mukherjee, B., Ghosal, D.: ‘Wireless sensor network survey’, 17 Li (Erran), L., Ramachandran, R., Buddhikot, M., Miller, S.: ‘Network
Comput. Netw., 2008, 52, (12), pp. 2292–2330 coding-based broadcast in mobile ad hoc networks’. Proc. IEEE
2 Al-Kofahi, O.M., Kamal, A.E.: ‘Network coding-based protection of INFOCOM, May 2007, pp. 1739–1747
many-to-one wireless flows’, IEEE J. Sel. Areas Commun., 2009, 27, 18 Yang, S., Wu, J.: ‘Efficient broadcasting using network coding and
(5), pp. 797– 813 directional antennas in MANETs’, IEEE Trans. Parallel Distrib. Syst.,
3 Haas, Z.J., Halpern, J.Y., Li, L.: ‘GOSSIP-based ad hoc routing’, IEEE/ 2010, 21, (2), pp. 148–161
ACM Trans. Netw., 2006, 14, (3), pp. 479– 491 19 Widmer, J., Fragouli, C., Boudec, J.-Y.L.: ‘Low-complexity energy
4 Zhang, Y., Fromherz, M.: ‘A robust and efficient flooding-based routing efficient broadcasting in wireless ad hoc networks using network
for wireless sensor networks’, J. Interconnect. Netw., 2006, 7, coding’. Proc. NetCod, Riva del Garda, Italy, April 2005
pp. 549–568 20 Katti, S., Rahul, H., Hu, W., Katabi, D., Medard, M., Crowcroft, J.:
5 Ahlswede, R., Cai, N., Li, S.-Y., Yeung, R.: ‘Network information ‘XORs in the air: practical wireless network coding’, IEEE Trans.
flow’, IEEE Trans. Inf. Theory, 2000, 46, pp. 1204–1216 Netw., 2008, 16, pp. 497–510
6 Le, T.A., Nakhai, M.R.: ‘Throughput analysis of network coding 21 Rout, R.R., Ghosh, S.K., Chakrabarti, S.: ‘Network coding-aware data
enabled wireless backhauls’, IET Commun., 2011, 5, (10), aggregation for a distributed wireless sensor network’. Proc. ICIIS
pp. 1318– 1327 2009, Sri Lanka, December 2009, pp. 32–36
7 Wang, S., Vasilakos, A., Jiang, H., et al.: ‘Energy efficient broadcasting 22 Ni, S.Y., Tseng, Y.C., Chen, Y.S., Sheu, J.P.: ‘The broadcast storm
using network coding aware protocol in wireless ad hoc network’. Proc. problem in a mobile ad hoc network’. Proc. MOBICOM ’99, Seattle
IEEE ICC 2011, Kyoto, Japan, 2011, pp. 1– 5 Washington, USA, 1999, pp. 151 –162
8 Aly, S.A., Kamal, A.E.: ‘Network protection codes against link failures 23 Karp, B., Kung, H.T.: ‘GPRS: greedy perimeter stateless routing for
using network coding’. Proc. IEEE GLOBECOM, December 2008, wireless networks’. Proc. ACM/IEEE MOBICOM, Boston, USA,
pp. 1– 6 2000, pp. 243– 254
9 Bahramgiri, H., Lahouti, F.: ‘Robust network coding against path 24 Bulusu, N., Heidemann, J., Estrin, D.: ‘GPS-less low-cost outdoor
failures’, IET Commun., 2010, 4, (3), pp. 272– 284 localization for very small devices’, IEEE Pers. Commun., 2000, 7,
10 Shabdanov, S., Rosenberg, C., Mitran, P.: ‘Joint routing, scheduling, (5), pp. 28–34
and network coding for wireless multihop networks’. Proc. Ninth Int. 25 Fekete, S.P., Kroller, A., Pfisterer, D., Fischer, S., Buschmann, C.:
Symp. on Modeling and Optimization in Mobile, Ad Hoc, and ‘Neighborhood-based topology recognition in sensor networks’. Proc.
Wireless Networks (WiOpt), May 2011, pp. 33– 40 Int. Workshop on ALGOSENSOR, Turku, 2004, pp. 123–136
11 Keller, L., Atsan, E., Argyraki, K., Fragouli, C.: ‘Sensecode: network 26 Rout, R.R., Ghosh, S.K., Chakrabarti, S.: ‘A network coding based
coding for reliable sensor networks’. EPFL Technical Report, October probabilistic routing scheme for wireless sensor network’. Proc. Sixth
2009 Int. Conf. on Wireless Communication and Sensor Networks (WCSN)
12 Chekuri, C., Fragouli, C., Soljanin, E.: ‘On average throughput and 2010, IIT, Allahabad, India, December 2010, pp. 27– 32
alphabet size in network coding’, IEEE Trans. Inf. Theory, 2006, 52, 27 Shelby, Z., Raez, C.P., Karvonen, H., Haapola, J.: ‘Energy optimization
pp. 2410– 2424 in multihop wireless embedded and sensor networks’, Int. J. Wirel. Inf.
13 Lun, D., Medard, M., Koetter, R., Effros, M.: ‘On coding for reliable Netw., 2005, 12, (1), pp. 11– 21
communication over packet networks’, Phys. Commun., 2008, 1, 28 Simon, G., Volgyesi, P., Maroti, M., Ledeczi, A.: ‘Simulation-based
pp. 3– 20 optimization of communication protocols for large-scale wireless
14 Gkantsidis, C., Rodriguez, P.: ‘Network coding for large scale content sensor networks’. Proc. Aerospace Conf., March 2003, pp. 1–8
distribution’. Proc. 24th IEEE INFOCOM, 2005, vol. 4, pp. 2235– 2245 29 Mainwaring, A., Polastre, J., Szewczyk, R., Culler, D., Anderson, J.:
15 Nguyen, D., Tran, T., Nguyen, T., Bose, B.: ‘Wireless broadcast using ‘Wireless sensor networks for habitat monitoring’. Proc. First ACM
network coding’, IEEE Trans. Veh. Technol., 2009, 58, (2), Workshop on Wireless Sensor Networks and Applications, Atlanta,
pp. 914–925 Georgia, USA, September 2002, pp. 88–97

IET Wirel. Sens. Syst., 2012, Vol. 2, Iss. 2, pp. 75–85 85


doi: 10.1049/iet-wss.2011.0069 & The Institution of Engineering and Technology 2012

You might also like