Module 3

WSN-MAC, routing, localization &QoS issues

Department of ECE, PDACE

Wireless Sensor Networks

Department of ECE, PDACE

Introduction
• Sensor networks are highly distributed networks of small, lightweight wireless nodes, deployed in
large numbers to monitor the environment or system by the measurement of physical parameters
such as temperature, pressure, or relative humidity. Building sensors has been made possible by
the recent advances in micro-electro mechanical systems (MEMS) technology.
• (MEMS devices are miniature structures fabricated on silicon substrates in a similar manner to
silicon integrated circuits. However, unlike electronic circuits, these are mechanical devices.)
• Each node of the sensor network consists of three subsystems: the sensor subsystem which
senses the environment, the processing subsystem which performs local computations on the
sensed data, and the communication subsystem which is responsible for message exchange with
neighboring sensor nodes. While individual sensors have limited sensing region, processing
power, and energy, networking a large number of sensors gives rise to a robust, reliable, and
accurate sensor network covering a wider region.

Department of ECE, PDACE

• The network is fault-tolerant because many nodes are sensing the same events. Further, the
nodes cooperate and collaborate on their data, which leads to accurate sensing of events in the
environment. The two most important operations in a sensor network are data dissemination,
that is, the propagation of data/queries throughout the network, and data gathering, that is, the
collection of observed data from the individual sensor nodes to a sink.

• Sensor networks consist of different types of sensors such as seismic, thermal, visual, and
infrared, and they monitor a variety of ambient conditions such as temperature, humidity,
pressure, and characteristics of objects and their motion. Sensor nodes can be used in military,
health, chemical processing, and disaster relief scenarios. Some of the academic and industry-
supported research programs on sensor networks include working on Smart Dust at the
University of California, Berkeley (UCB), and wireless integrated network sensor (WINS) at the
University of California, Los Angeles (UCLA).

Department of ECE, PDACE

Applications of Sensor Networks
• Sensor nodes are used in a variety of applications which require constant monitoring and
detection of specific events. The military applications of sensor nodes include battlefield
surveillance and monitoring, guidance systems of intelligent missiles, and detection of attack by
weapons of mass destruction, such as chemical, biological, or nuclear. Sensors are also used in
environmental applications such as forest fire and flood detection, and habitat exploration of
animals.
• Sensors can be extremely useful in patient diagnosis and monitoring. Patients can wear small
sensor devices that monitor their physiological data such as heart rate or blood pressure. The
data collected can be sent regularly over the network to automated monitoring systems which
are designed to alert the concerned doctor on detection of an anomaly. Such systems provide
patients a greater freedom of movement instead of their being confined to a hospital. Sensor
nodes can also be made sophisticated enough to correctly identify allergies and prevent wrong
diagnosis.

Department of ECE, PDACE

• Sensors will soon find their way into a host of commercial applications at home and in industries.
Smart sensor nodes can be built into appliances at home, such as ovens, refrigerators, and
vacuum cleaners, which enable them to interact with each other and be remote-controlled. The
home can provide a "smart environment" which adapts itself according to the user's tastes. For
instance, the lighting, music, and ambiance in the room can be automatically set according to the
user's preferences. Similar control is useful in office buildings too, where the airflow and
temperature of different parts of the building can be automatically controlled. Warehouses could
improve their inventory control system by installing sensors on the products to track their
movement. The applications of sensor networks are endless, limited only by the human
imagination.

Department of ECE, PDACE

Comparison with Ad Hoc Wireless Networks
• While both ad hoc wireless networks and sensor networks consist of wireless nodes
communicating with each other, there are certain challenges posed by sensor networks. The
number of nodes in a sensor network can be several orders of magnitude larger than the number
of nodes in an ad hoc network. Sensor nodes are more prone to failure and energy drain, and
their battery sources are usually not replaceable or rechargeable. Sensor nodes may not have
unique global identifiers, so unique addressing is not always feasible in sensor networks.

Department of ECE, PDACE

• Sensor networks are data-centric, that is, the queries in sensor networks are addressed to nodes
which have data satisfying some conditions. For instance, a query may be addressed to all nodes
"in the south-east quadrant," or to all nodes "which have recorded a temperature greater than 30
°C." On the other hand, ad hoc networks are address-centric, with queries addressed to particular
nodes specified by their unique address. Hence, sensor networks require a different mechanism
for routing and answering queries. Most routing protocols used in ad hoc networks cannot be
directly ported to sensor networks because of limitations in memory, power, and processing
capabilities in the sensor nodes and the non-scalable nature of the protocols.
• An important feature of sensor networks is data fusion/aggregation, whereby the sensor nodes
aggregate the local information before relaying. The main goals of data fusion are to reduce
bandwidth consumption, media access delay, and power consumption for communication.

Department of ECE, PDACE

Issues and Challenges in Designing a Sensor
Network
Sensor networks pose certain design challenges due to the following reasons:
• Sensor nodes are randomly deployed and hence do not fit into any regular topology. Once
deployed, they usually do not require any human intervention. Hence, the setup and
maintenance of the network should be entirely autonomous.
• Sensor networks are infrastructure-less. Therefore, all routing and maintenance algorithms need
to be distributed.
• An important bottleneck in the operation of sensor nodes is the available energy. Sensors usually
rely only on their battery for power, which in many cases cannot be recharged or replaced. Hence,
the available energy at the nodes should be considered as a major constraint while designing
protocols. For instance, it is desirable to give the user an option to trade off network lifetime for
fault tolerance or accuracy of results.
• Hardware design for sensor nodes should also consider energy efficiency as a primary
requirement. The micro-controller, operating system, and application software should be
designed to conserve power.
Department of ECE, PDACE
• Sensor nodes should be able to synchronize with each other in a completely distributed manner,
so that TDMA schedules can be imposed and temporal ordering of detected events can be
performed without ambiguity.
• A sensor network should also be capable of adapting to changing connectivity due to the failure
of nodes, or new nodes powering up. The routing protocols should be able to dynamically include
or avoid sensor nodes in their paths.
• Real-time communication over sensor networks must be supported through provision of
guarantees on maximum delay, minimum bandwidth, or other QoS parameters.
• Provisions must be made for secure communication over sensor networks, especially for military
applications which carry sensitive data.

Department of ECE, PDACE

Classification of Sensor
Networks

Department of ECE, PDACE

Department of ECE, PDACE
Sensor Network Architecture
The design of sensor networks is influenced by factors such as scalability, fault tolerance, and power
consumption. The two basic kinds of sensor network architecture are layered and clustered.

Department of ECE, PDACE

Layered Architecture

Fig. 1

Department of ECE, PDACE

• A layered architecture has a single powerful base station (BS), and the layers of sensor nodes
around it correspond to the nodes that have the same hop-count to the BS. This is depicted in
Figure 1.
• Layered architectures have been used with in-building wireless backbones, and in military sensor-
based infrastructure, such as the multi-hop infrastructure network architecture (MINA). In the in-
building scenario, the BS acts an access point to a wired network, and small nodes form a wireless
backbone to provide wireless connectivity. The users of the network have hand-held devices such
as PDAs which communicate via the small nodes to the BS.
• Similarly, in a military operation, the BS is a data-gathering and processing entity with a
communication link to a larger network. A set of wireless sensor nodes is accessed by the hand-
held devices of the soldiers. The advantage of a layered architecture is that each node is involved
only in short-distance, low-power transmissions to nodes of the neighbouring layers.
Unified Network Protocol Framework (UNPF)
• UNPF is a set of protocols for complete implementation of a layered architecture for sensor
networks. UNPF integrates three operations in its protocol structure: network initialization and
maintenance, MAC, and routing protocols.

Department of ECE, PDACE

Clustered Architecture

Fig. 2

Department of ECE, PDACE

• A clustered architecture organizes the sensor nodes into clusters, each governed by a cluster-
head. The nodes in each cluster are involved in message exchanges with their respective cluster-
heads, and these heads send messages to a BS, which is usually an access point connected to a
wired network. Figure 2 represents a clustered architecture where any message can reach the BS
in at most two hops. Clustering can be extended to greater depths hierarchically.
• Clustered architecture is specially useful for sensor networks because of its inherent suitability for
data fusion. The data gathered by all members of the cluster can be fused at the cluster-head, and
only the resulting information needs to be communicated to the BS. Sensor networks should be
self-organizing, hence the cluster formation and election of cluster-heads must be an
autonomous, distributed process. This is achieved through network layer protocols such as the
low-energy adaptive clustering hierarchy (LEACH).
Low-Energy Adaptive Clustering Hierarchy (LEACH)
• LEACH is a clustering-based protocol that minimizes energy dissipation in sensor networks. LEACH
randomly selects nodes as cluster-heads and performs periodic re-election, so that the high-
energy dissipation experienced by the cluster-heads in communicating with the BS is spread
across all nodes of the network. Each iteration of selection of cluster-heads is called a round. The
operation of LEACH is split into two phases: set-up and steady.

Department of ECE, PDACE

• During the set-up phase, each sensor node chooses a random number between 0 and 1. If this is
lower than the threshold for node n, T(n), the sensor node becomes a cluster-head. The threshold
T(n) is calculated as

where P is the desired percentage of nodes which are cluster-heads, r is the current round, and G is
the set of nodes that has not been cluster-heads in the past 1/P rounds. This ensures that all
sensor nodes eventually spend equal energy. After selection, the cluster-heads advertise their
selection to all nodes. All nodes choose their nearest cluster-head when they receive
advertisements based on the received signal strength. The cluster-heads then assign a TDMA
schedule for their cluster members.
• The steady phase is of longer duration in order to minimize the overhead of cluster formation.
During the steady phase, data transmission takes place based on the TDMA schedule, and the
cluster-heads perform data aggregation/fusion through local computation. The BS receives only
aggregated data from clusterheads, leading to energy conservation. After a certain period of time
in the steady phase, cluster-heads are selected again through the set-up phase.

Department of ECE, PDACE

Data Dissemination

Department of ECE, PDACE

• Data dissemination is the process by which queries or data are routed in the sensor network. The
data collected by sensor nodes has to be communicated to the BS or to any other node interested
in the data. The node that generates data is called a source and the information to be reported is
called an event. A node which is interested in an event and seeks information about it is called a
sink. Traffic models have been developed for sensor networks such as the data collection and data
dissemination (diffusion) models. In the data collection model, the source sends the data it
collects to a collection entity such as the BS. This could be periodic or on demand. The data is
processed in the central collection entity.
• Data diffusion, on the other hand, consists of a two-step process of interest propagation and data
propagation. An interest is a descriptor for a particular kind of data or event that a node is
interested in, such as temperature, intrusion, or presence of bio-agents. For every event that a
sink is interested in, it broadcasts its interest to its neighbors and periodically refreshes its
interest. The interest is propagated across the network, and every node maintains an interest
cache of all events to be reported.
• This is similar to a multicast tree formation, rooted at the sink. When an event is detected, it is
reported to the interested nodes after referring to the interest cache. Intermediate nodes
maintain a data cache and can aggregate the data or modify the rate of reporting data. The paths
used for data propagation are modified by preferring the shortest paths and deselecting the
weaker or longer paths. The basic idea of diffusion is made efficient and intelligent by different
algorithms for interest and data routing.

Department of ECE, PDACE

1. Flooding
In flooding, each node which receives a packet broadcasts it if the maximum hop-count of the
packet is not reached and the node itself is not the destination of the packet. This technique does
not require complex topology maintenance or route discovery algorithms. But flooding has the
following disadvantages:
• Implosion: This is the situation when duplicate messages are sent to the same node. This occurs
when a node receives copies of the same message from many of its neighbors.
• Overlap: The same event may be sensed by more than one node due to overlapping regions of
coverage. This results in their neighbors receiving duplicate reports of the same event.
• Resource blindness: The flooding protocol does not consider the available energy at the nodes
and results in many redundant transmissions. Hence, it reduces the network lifetime.
2. Gossiping
• Gossiping is a modified version of flooding, where the nodes do not broadcast a packet, but send
it to a randomly selected neighbor. This avoids the problem of implosion, but it takes a long time
for a message to propagate throughout the network. Though gossiping has considerably lower
overhead than flooding, it does not guarantee that all nodes of the network will receive the
message. It relies on the random neighbor selection to eventually propagate the message
throughout the network.

Department of ECE, PDACE

3. Rumour Routing
• Rumor routing is an agent-based path creation algorithm [6]. Agents, or "ants,“ are long-lived
entities created at random by nodes. These are basically packets which are circulated in the
network to establish shortest paths to events that they encounter. They can also perform path
optimizations at nodes that they visit. When an agent finds a node whose path to an event is
longer than its own, it updates the node's routing table. Figure 3 illustrates the working of the
rumor routing algorithm.
• In Figure 3 (a), the agent has initially recorded a path of distance 2 to event E1. Node A's table
shows that it is at a distance 3 from event E1 and a distance 2 from E2. When the agent visits
node A, it updates its own path state information to include the path to event E2. The updating is
with one hop greater distance than what it found in A, to account for the hop between any
neighbor of A that the agent will visit next, and A. It also optimizes the path to E1 recorded at
node A to the shorter path through node B. The updated status of the agent and node table is
shown in Figure 3 (b).

Department of ECE, PDACE

Fig. 3

Department of ECE, PDACE

• When a query is generated at a sink, it is sent on a random walk with the hope that it will find a
path (pre-established by an agent) leading to the required event. This is based on the high
probability of two straight lines intersecting on a planar graph, assuming the network topology is
like a planar graph, and the paths established can be approximated by straight lines owing to high
density of the nodes. If a query does not find an event path, the sink times out and uses flooding
as a last resort to propagate the query. For instance, as in Figure 3 (c), suppose a query for event
E1 is generated by node P. Through a random walk, it reaches A, where it finds the previously
established path to E1. Hence, the query is directed to E1 through node B, as indicated by A's
table.

Department of ECE, PDACE

4. Sequential Assignment Routing
• A set of algorithms which performs organization and mobility management in sensor networks is
proposed. The sequential assignment routing (SAR) algorithm creates multiple trees, where the
root of each tree is a one-hop neighbor of the sink. Each tree grows outward from the sink and
avoids nodes with low throughput or high delay. At the end of the procedure, most nodes belong
to multiple trees. An instance of tree formation is illustrated in Figure 12.5. The trees rooted at A
and B, two of the one-hop neighbors of the sink, are shown. Node C belongs to both trees, and
has path lengths of 3 and 5, respectively, to the sink, using the two trees. Each sensor node
records two parameters about each path through it: the available energy resources on the path
and an additive QoS metric such as delay.

Department of ECE, PDACE

Fig. 4

Department of ECE, PDACE

• This allows a node to choose one path from among many to relay its message to the sink. The SAR
algorithm chooses a path with high estimated energy resources, and provisions can be made to
accommodate packets of different priorities. A weighted QoS metric is used to handle prioritized
packets, which is computed as a product of priority level and delay. The routing ensures that the
same weighted QoS metric is maintained. Thus, higher priority packets take lower delay paths,
and lower priority packets have to use the paths of greater delay.
• For example, if node C generates a packet of priority 3, it follows the longer path along tree B,
and a packet of priority 5 (higher priority) will follow the shorter path along tree A, so that the
priority × delay QoS metric is maintained. SAR minimizes the average weighted QoS metric over
the lifetime of the network. The sink periodically triggers a metric update to reflect the changes in
available energy resource after some transmissions.

Department of ECE, PDACE

5. Directed Diffusion
• The directed diffusion protocol is useful in scenarios where the sensor nodes themselves generate
requests/queries for data sensed by other nodes, instead of all queries arising only from a BS.
Hence, the sink for the query could be a BS or a sensor node. The directed diffusion routing
protocol improves on data diffusion using interest gradients. Each sensor node names its data
with one or more attributes, and other nodes express their interest depending on these
attributes. Attribute-value pairs can be used to describe an interest in intrusion data as follows,
where an interest is nothing but a set of descriptors for the data in which the querying node is
interested.

Department of ECE, PDACE

• The sink has to periodically refresh its interest if it still requires the data to be reported to it. Data
is propagated along the reverse path of the interest propagation. Each path is associated with a
gradient that is formed at the time of interest propagation. While positive gradients encourage
the data flow along the path, negative gradients inhibit the distribution of data along a particular
path. The strength of the interest is different toward different neighbors, resulting in source-to-
sink paths with different gradients. The gradient corresponding to an interest is derived from the
interval/data-rate field specified in the interest. For example, if there are two paths formed with
gradients 0.8 and 0.4, the source may send twice as much data along the higher gradient path
compared to the lower gradient one. For the interest mentioned earlier, a sensor may send data
of the following kind:

Department of ECE, PDACE

• The diffusion model allows nodes to cache or locally transform (aggregate) data. This increases
the scalability of communication and reduces the number of message transmissions required.
• The concept of reinforcement is used to update a node's interest along a particular path. For
example, suppose the sink wants more frequent updates from the sensors which have detected
an event. It reinforces the path by sending an interest with a higher data-rate requirement, in
effect increasing the gradient of that path. On the other hand, if the sink needs only fewer
updates, it applies negative reinforcement by sending an interest of lower required data-rate.
• The directed diffusion model uses data naming by attributes and local data transformations to
reflect the data-centric nature of sensor network operations. The local operations of data
aggregation are application-specific. Gradients model the network-wide results of local
interactions by regulating the flow of data along different paths, depending on the expressed
interest.

Department of ECE, PDACE

6. Sensor Protocols for Information via Negotiation
• A family of protocols called sensor protocols for information via negotiation (SPIN). SPIN uses
negotiation and resource adaptation to address the deficiencies of flooding. Negotiation reduces
overlap and implosion, and a threshold-based resource-aware operation is used to prolong
network lifetime. Meta-data, or data describing data, is transmitted instead of raw data. This
requires fewer bytes and can be in an application-specific format. SPIN has three types of
messages: ADV, REQ, and DATA. A sensor node broadcasts an ADV containing meta-data
describing the actual data. If a neighbor is interested in the data, it sends a REQ for the data. Then
the sensor node sends the actual DATA to the neighbor. The neighbor again sends ADVs to its
neighbors and this process continues to disseminate the data throughout the network. This
simple version of SPIN is shown in Figure 5.

Department of ECE, PDACE

 Fig. 5

• SPIN is based on data-centric routing, where the nodes advertise the available data through an
ADV and wait for requests from interested nodes. SPIN-2 expands on SPIN, using an energy or
resource threshold to reduce participation. A node may participate in the ADV-REQ-DATA
handshake only if it has sufficient resources above a threshold.

Department of ECE, PDACE

7 Cost-Field Approach
• The cost-field approach considers the problem of setting up paths to a sink. It is a two-phase
process, the first phase being to set up the cost field, based on metrics such as delay, at all sensor
nodes, and the second being data dissemination using the costs. At each node, the cost is defined
as the minimum cost from that node to the sink, which occurs along the optimal path. Explicit
path information does not need to be maintained.
• Phase 1 sets up a cost field starting from the sink node. A sink broadcasts an ADV packet with its
own cost as 0. When a node N hears an ADV message from node M, it sets its own path cost to
min(LN, LM + CNM), where LN is the total path cost from node N to sink, LM represents the cost of
node M to sink, and CNM is the cost from node N to M. If LN was updated, the new cost is
broadcast through another ADV. This is a flooding-based implementation of the Dijkstra's
algorithm. In order to reduce the high communication costs associated with flooding, a back-off-
based approach is used. The main reason for overhead is that a node broadcasts its updated cost
immediately, whether it was the optimal cost or not. Instead, the back-off modification makes a
node defer its ADV instead of immediately broadcasting it. The time to defer is heuristically
determined as γ × CMN, where γ is a parameter of the algorithm.

Department of ECE, PDACE

• The working of the cost-field approach with back-off is illustrated in Figure 6. The numbers on the
links indicate link costs. The value of γ is assumed to be 10. Initially, nodes N and P did not have a
path to the sink and hence had their costs set to ∞. In Figure 6 (a), node M broadcasts an ADV,
which is received by nodes N and P. They tentatively fix their costs to LM +2 and LM +5,
respectively, and set their back-off timers to 20 and 50, respectively. Figure 6 (b) shows the costs
after 20 time units, when node N's back-off timer expires. Node N finalizes its cost to LM + 2 and
broadcasts an ADV, which is heard by node P. Since LN + 1 < LM + 5, node P updates its cost and
sets a new back-off timer to 10. The unnecessary ADV of node P's earlier non-optimal cost is
avoided by setting the back-off timer. Finally, at 30 time units, node P finalizes its cost to LN + 1
and broadcasts an ADV, as shown in Figure 6 (c).

Department of ECE, PDACE

Phase 2 is the data
dissemination process. Once
the cost field is established,
a source sends its message
to sink S with cost CS . The
message also contains a
cost-so-far field, initially set
to 0. Each intermediate node
forwards the packet if the
cost recorded in the packet
plus its own cost equals the
original source-to sink cost.
This ensures that the original
optimal path is used
whenever a packet is routed. Fig. 6
While forwarding, the
intermediate nodes also
update the cost-so-far field.
Department of ECE, PDACE
8. Geographic Hash Table
• Geographic hash table (GHT) is a system based on data-centric storage, inspired by Internet-scale
distributed hash table (DHT) systems such as Chord and Tapestry. GHT hashes keys into
geographic coordinates and stores a (key, value) pair at the sensor node nearest to the hash
value. The calculated hash value is mapped onto a unique node consistently, so that queries for
the data can be routed to the correct node. Stored data is replicated to ensure redundancy in
case of node failures, and a consistency protocol is used to maintain the replicated data. The data
is distributed among nodes such that it is scalable and the storage load is balanced. The routing
protocol used is greedy perimeter stateless routing (GPSR), which again uses geographical
information to route the data and queries.
• GHT is more effective in large sensor networks, where a large number of events are detected but
not all are queried. In this case, the data observed is stored in a distributed manner across all
nodes, instead of being routed to a central external storage. Queries are routed to the nearest
node which contains a copy of the relevant data. This makes the storage and traffic distribution
uniform.

Department of ECE, PDACE

9. Small Minimum Energy Communication Network
• Small minimum energy communication network (SMECN) is a protocol proposed to construct a
sub-network from a given communication network. If the entire sensor network is represented by
a graph G, the subgraph G' is constructed such that the energy usage of the network is minimized.
The number of edges in G' is less than that of G, but all nodes of G are retained in G'. The
connectivity between any two nodes is not disrupted by the subgraph. G' is constructed such that
the energy required to transmit data from a node to all its neighbors is lower in G' than in G.
• SMECN also follows the minimum energy (ME) property in its subgraph construction, that is,
there exists an ME path in subgraph G' between any two nodes that are connected in G. The
power required to transmit data between two nodes u and v is modelled as

where t is a constant, n is the path loss exponent indicating the loss of power with distance from
the transmitter, and d(u, v) is the distance between u and v. Let the power needed to receive the
data be c. Since the transmission power increases exponentially with distance, it would be more
economical to transmit data by smaller hops.

Department of ECE, PDACE

• Suppose the path between u (i.e., u0 ) and v (i.e., uk) is represented by r1= (u0 , u1 , ...uk), such that
each (ui , ui+ 1 ) is an edge in the subgraph G', then the total power consumed for the
transmission is

• The path r is the ME path if C(r) ≤ C(r') for all paths r' between u and v in the graph G. The
subgraph G' is said to have the ME property if there exists a path r in G' which is an ME path in G,
for all node pairs (u, v). SMECN uses only the ME paths from G' for data transmission, so that the
overall energy consumed is minimized.

Department of ECE, PDACE

Data Gathering
• The objective of the data-gathering problem is to transmit the sensed data from each sensor
node to a BS. One round is defined as the BS collecting data from all the sensor nodes once. The
goal of algorithms which implement data gathering is to maximize the number of rounds of
communication before the nodes die and the network becomes inoperable. This means minimum
energy should be consumed and the transmission should occur with minimum delay, which are
conflicting requirements. Hence, the energy × delay metric is used to compare algorithms, since
this metric measures speedy and energy-efficient data gathering. A few algorithms that
implement data gathering are discussed below.

Department of ECE, PDACE

1. Direct Transmission
• All sensor nodes transmit their data directly to the BS. This is extremely expensive in terms of
energy consumed, since the BS may be very far away from some nodes. Also, nodes must take
turns while transmitting to the BS to avoid collision, so the media access delay is also large.
Hence, this scheme performs poorly with respect to the energy × delay metric.
2 Power-Efficient Gathering for Sensor Information Systems
• Power-efficient gathering for sensor information systems (PEGASIS) is a data-gathering protocol
based on the assumption that all sensor nodes know the location of every other node, that is, the
topology information is available to all nodes. Also, any node has the required transmission range
to reach the BS in one hop, when it is selected as a leader. The goals of PEGASIS are as follows:
1. Minimize the distance over which each node transmits
2. Minimize the broadcasting overhead
3. Minimize the number of messages that need to be sent to the BS
4. Distribute the energy consumption equally across all nodes

Department of ECE, PDACE

• A greedy algorithm is used to construct a chain of
sensor nodes, starting from the node farthest
from the BS. At each step, the nearest neighbor
which has not been visited is added to the chain.
The chain is constructed a priori, before data
transmission begins, and is reconstructed when
nodes die out. At every node, data fusion or
aggregation is carried out, so that only one
message is passed on from one node to the next.
A node which is designated as the leader finally
transmits one message to the BS. Leadership is
transferred in sequential order, and a token is Fig. 7
passed so that the nodes know in which direction
to pass messages in order to reach the leader. A
possible chain formation is illustrated in Figure 7.
The delay involved in messages reaching the BS is
O(N), where N is the total number of nodes in the
network.

Department of ECE, PDACE

3. Binary Scheme
• This is also a chain-based scheme like PEGASIS, which classifies nodes into different levels. All
nodes which receive messages at one level rise to the next. The number of nodes is halved from
one level to the next. For instance, consider a network with eight nodes labeled s0 to s7. As Figure
8 shows, the aggregated data reaches the BS in four steps, which is O(log2N), where N is the
number of nodes in the network. This scheme is possible when nodes communicate using CDMA,
so that transmissions of each level can take place simultaneously.

Fig. 8

Department of ECE, PDACE

4 Chain-Based Three-Level Scheme
• For non-CDMA sensor nodes, a binary scheme is not applicable. The chain-based three-level
scheme addresses this situation, where again a chain is constructed as in PEGASIS. The chain is
divided into a number of groups to space out simultaneous transmissions in order to minimize
interference. Within a group, nodes transmit one at a time. One node out of each group
aggregates data from all group members and rises to the next level. The index of this leader node
is decided a priori.
• In the second level, all nodes are divided into two groups, and the third level consists of a
message exchange between one node from each group of the second level. Finally, the leader
transmits a single message to the BS. The working of this scheme is illustrated in Figure 9. The
network has 100 nodes, and the group size is ten for the first level and five for the second level.
Three levels have been found to give the optimal energy × delay through simulations.

Fig. 9

Department of ECE, PDACE

MAC Protocols for Wireless
Sensor Networks

Department of ECE, PDACE

1. Self-Organizing MAC for Sensor Networks & Eavesdrop and Register
• Self-organizing MAC for sensor (SMACS) networks and eavesdrop and register (EAR) are two
protocols which handle network initialization and mobility support, respectively. SMACS is a
distributed protocol for network initialization and link-layer organization.
• In this protocol, neighbor discovery and channel assignment take place simultaneously in a
completely distributed manner. A communication link between two nodes consists of a pair of
time slots, at a fixed frequency, which is randomly chosen at the time of establishing the link.
Such an assignment is possible in sensor networks without interference from neighboring nodes
because the available bandwidth is much larger than the data rate required for a message
transmission between two nodes. This scheme requires synchronization only between
communicating neighbors, in order to precisely define the slots to be used for their
communication. Power is conserved by turning off the transceiver during idle slots, and using a
random wake-up schedule during the network start-up phase.
• The EAR protocol enables seamless connection of nodes under mobile and stationary conditions.
This protocol makes use of certain mobile nodes, besides the existing stationary sensor nodes, to
offer service to maintain connections. Mobile nodes eavesdrop on the control signals and
maintain neighbour information. The mobile nodes assume full control over connections and can
drop connections when they move away. Mobility is hence made transparent to SMACS, since it is
independently handled by EAR.

Department of ECE, PDACE

2. Hybrid TDMA/FDMA
• This is a centrally controlled scheme which assumes that nodes communicate directly to a nearby
BS. A pure TDMA scheme minimizes the time for which a node has to be kept on, but the
associated time synchronization costs are very high. A pure FDMA scheme allots the minimum
required bandwidth for each connection. The hybrid TDMA/FDMA scheme uses an optimum
number of channels, which gives minimum overall power consumption. This is found to depend
on the ratio of power consumption of transmitter to receiver. If the transmitter consumes more
power, a TDMA scheme is favored, since it can be switched off in idle slots to save power.
• On the other hand, the scheme favors FDMA when the receiver consumes greater power. This is
because, in FDMA, the receiver need not expend power for time synchronization by receiving
during the guard band between slots, which becomes essential in a TDMA scheme.

Department of ECE, PDACE

3. CSMA based MAC Protocols
• Traditional CSMA-based schemes are more suitable for point-to-point stochastically distributed
traffic flows. On the other hand, sensor networks have variable but periodic and correlated traffic.
The sensing periods of CSMA are constant for energy efficiency, while the back-off is random to
avoid repeated collisions.
• Binary exponential back-off is used to maintain fairness in the network. An adaptive transmission
rate control (ARC) is also used, which balances originating and route-through traffic in nodes. This
ensures that nodes closer to the BS are not favored over farther nodes. ARC uses linear increase
and multiplicative decrease of originating traffic in a node. The penalty for dropping route-
through traffic is higher, since energy has already been invested in making the packets reach until
that node. ARC performs phase changes, that is, it staggers the transmission times of different
streams so that periodic streams are less likely to collide repeatedly. Hence, CSMA based MAC
protocols are contention-based and are designed mainly to increase energy efficiency and
maintain fairness.

Department of ECE, PDACE

Location Discovery
• The location information of sensors has to be considered during aggregation of sensed data. This
implies each node should know its location and couple its location information with the data in the
messages it sends. A low-power, inexpensive, and reasonably accurate mechanism is needed for
location discovery. A global positioning system (GPS) is not always feasible because it cannot reach
nodes in dense foliage or indoors. It also consumes high power and makes sensor nodes bulkier.
1 Indoor Localization
• Indoor localization techniques use a fixed infrastructure to estimate the location of sensor nodes.
Fixed beacon nodes are strategically placed in the field of observation, typically indoors, such as
within a building. The randomly distributed sensors receive beacon signals from the beacon nodes
and measure the signal strength, angle of arrival, and time difference between the arrival of different
beacon signals. Using the measurements from multiple beacons, the nodes estimate their location.
Some approaches use simple triangulation methods, while others require a priori database creation of
signal measurements. The nodes estimate distances by looking up the database instead of performing
computations. However, storage of the database may not be possible in each node, so only the BS
may carry the database.

Department of ECE, PDACE

2. Sensor Network Localization
• In situations where there is no fixed infrastructure available and prior measurements are not
possible, some of the sensor nodes themselves act as beacons. They have their location
information, using GPS, and these send periodic beacons to other nodes. In the case of
communication using RF signals, the received signal strength indicator (RSSI) can be used to
estimate the distance, but this is very sensitive to obstacles and environmental conditions.
Alternatively, the time difference between beacon arrivals from different nodes can be used to
estimate location, if RF or ultrasound signals are used for communication. This offers a lower
range of estimation than RSSI, but is of greater accuracy.
• Localization algorithms require techniques for location estimation depending on the beacon
nodes' location. These are called multi-lateration (ML) techniques.

Department of ECE, PDACE

• Some simple ML techniques are
described in what follows:
Atomic ML: If a node receives three
beacons, it can determine its position by a
mechanism similar to GPS. This is illustrated
in Figure 10. Fig. 10

Iterative ML: Some nodes may not be in the

direct range of three beacons. Once a node
estimates its location, it sends out a beacon,
which enables some other nodes to now
receive at least three beacons. Iteratively,
all nodes in the network can estimate their
location. This is shown in Figure 11. The
drawback of this multi-hop method is that
errors are propagated, hence estimation of
location may not be accurate. Fig. 11

Department of ECE, PDACE

Collaborative ML: When two or more nodes cannot receive at least three beacons each, they
collaborate with each other. As shown in Figure 12, node A and node B have three neighbors each.
Of the six participating nodes, four are beacons, whose positions are known. Hence, by solving a set
of simultaneous quadratic equations, the positions of A and B can be determined.

Fig. 12

Department of ECE, PDACE

Quality of a sensor network

Department of ECE, PDACE

1. Coverage
• Coverage is a measure of how well the network can observe or cover an event. Coverage depends
upon the range and sensitivity of the sensing nodes, and the location and density of the sensing
nodes in the given region. The worst-case coverage defines areas of breach, that is, where
coverage is the poorest. This can be used to determine if additional sensors need to be deployed
to improve the network. The best-case coverage, on the other hand, defines the areas of best
coverage. A path along the areas of best coverage is called a maximum support path or maximum
exposure path.
• The coverage problem is formally defined as follows: Given a field A with a set of sensors S = {s1 ,
s2 , ...,sn}, where for each sensor si in S, its location coordinates (xi , yi ) are known, based on
localization techniques. Areas I and F are the initial and final locations of an intruder traversing
the field. The problem is to identify PB , the maximal breach path starting in I and ending in F. PB is
defined as the locus of points p in the region A, where p is in PB if the distance from p to the
closest sensor is maximized.

Department of ECE, PDACE

• A mathematical technique to solve the
coverage problem is the Voronoi diagram. It
can be proved that the path PB will be
composed of line segments that belong to
the Voronoi diagram corresponding to the
sensor graph. In two dimensions, the Voronoi
diagram of a set of sites is a partitioning of
the plane into a set of convex polygons such
that all points inside a polygon are closest to
the site enclosed by the polygon, and the
polygons have edges equidistant from the
nearby sites. A Voronoi diagram for a sensor
network, and a breach path from I to F, are
shown in Figure 13.

Fig. 13

Department of ECE, PDACE

• The algorithm to find the breach path PB is:
1. Generate the Voronoi diagram, with the set of vertices V and the set of edges E. This is done by
drawing the perpendicular bisectors of every line segment joining two sites, and using their
points of intersection as the vertices of the convex polygons.
2. Create a weighted graph with vertices from V and edges from E, such that the weight of each
edge in the graph is the minimum distance from all sensors in S. The edge weights represent
the distance from the nearest sensor. Smaller edge weights imply better coverage along the
edge.
3. Determine the maximum cost path from I to F, using breadth-first search.
• The maximum cost implies least coverage. Hence, the required breach path is along this
maximum-cost path determined from the Voronoi diagram. The breach path shows the region of
maximum vulnerability in a sensor network, where the coverage provided by the sensors is the
weakest.

Department of ECE, PDACE

• A related problem is that of finding the best-
case coverage. The problem is formally
stated as finding the path which offers the
maximum coverage, that is, the maximum
support path PS in S, from I to F. The solution
is obtained by a mathematical technique
called Delaunay triangulation, shown in
Figure 14. This is obtained from the Voronoi
diagram by connecting the sites whose
polygons share a common edge. The best
path PS will be a set of line segments from
the Delaunay triangulation, connecting
some of the sensor nodes. The algorithm is
again similar to that used to find the
maximum breach path, replacing the
Voronoi diagram by the Delaunay
triangulation, and defining the edge costs
proportional to the line segment lengths.
The maximum support path is hence formed Fig. 14
by a set of line segments connecting some
of the sensor nodes.

Department of ECE, PDACE

2. Exposure
• Exposure is defined as the expected ability of observing a target in the sensor field. It is formally
defined as the integral of the sensing function on a path from source node Ps to destination node
Pd . The sensing power of a node s at point p is usually modeled as

where λ and k are constants, and d(s, p) is the distance of p from s. Consider a network with sensors
s1 ,s2 , ..., sn . The total intensity at point p, called the all-sensor field intensity, is given by

Department of ECE, PDACE

• The closest-sensor field intensity at p is

where smin is the closest sensor to p. The exposure during travel of an event along a path p(t) is
defined by the exposure function

where is the elemental arc length, and t1 , t2 are the time instances between which the path is
traversed. For conversion from Cartesian coordinates (x(t), y(t)),

• In the simplest case of having one sensor node at (0, 0) in a unit field, the breach path or
minimum exposure path (MEP) from (-1, -1) to (1, 1) is shown in Figure 15.

Department of ECE, PDACE

 Fig. 16
• It can also be proved that for
a single sensor s in a
polygonal field, with vertices
v1 , v2 , ..., vn , the MEP
between two vertices vi and vj Fig. 15
can be determined as follows.
The edge (vi, vi+ 1 ) is tangent
to the inscribed circle at ui .
Then the MEP consists of the
line segment from vi to ui ,
part of the inscribed circle
from ui to uj, and the line
segment from uj to vj . This is
shown in Figure 16.
Department of ECE, PDACE
• The exposure problem is still unsolved for
two points in the same corner, or for
points within the inscribed circle. For the
generic exposure problem of determining
the MEP for randomly placed sensor
nodes in the network, the network is
tessellated with grid points. An example is
shown in Figure 17.
• To construct an n × n grid of order m,
each side of a square is divided into m
equal parts, creating (m + 1) vertices on
the edge. Within each square, all vertices
are connected to obtain a grid. Higher Fig. 17
order grids have greater accuracy. For
each edge in the grid network, the
exposure function is used to determine
the edge weights, and the MEP is defined
as the shortest path, determined by
Dijkstra's algorithm.

Department of ECE, PDACE

• The mathematical concept of exposure is important for evaluating the target detection capability
of a sensor network. Sensors are deployed in a given area to detect events occurring in the field
of interest. The nodes collaborate among themselves (perform data fusion) through the exchange
of localized information, and reach a decision about the location and movement of a given event
or target.

Department of ECE, PDACE