ubi532 - wsn spring 20091 distributed clustering in ad-hoc sensor networks : a hybrid,...

UBI532 - WSN Spring 2009 1

Distributed Clustering in Ad-hoc Sensor

Networks :

A Hybrid, Energy-Efficient Approach

HEED (Hybrid Energy Efficient Distributed

Clustering)

Paper By

Ossama Younis and Sonia Fahmy

Department of Computer Sciences, Purdue University

Presentation By Deniz Özsoyeller


Contributions of HEED

HEED is a new energy-efficient approach for clustering nodes in sensor networks.

Periodically selects cluster heads according to their residual energy and a secondary parameter, such as node proximity to its neighbors or node degree.

The clustering process terminates rapidly.

The protocol incurs low overhead in terms of processing cycles and messages exchanged.

Achieves fairly uniform cluster head distribution across the network.

Considers cluster quality, e.g., load-balanced clusters or dense clusters.


Sensor Networks Overview (1)

Sensor Nodes are usually:

Typically less mobile and more densely deployed than mobile ad-hoc networks (MANETs).

Limited in processing, memory, and

communication capabilities

Constrained in battery lifetime

Left unattended e.g., in hostile environments, which makes it difficult impossible to re-charge or replace their batteries.

Sensor networks have recently emerged as an important computing platform.


Sensor Networks Overview (2)

Application - Specific, e.g.,

Monitoring physical or environmental conditions, such as temperature, sound, vibration, pressure, motion or pollutants.

Surveying military fields

Reporting radiation levels at nuclear plants

Used in many civilian application areas, including environment and habitat monitoring, healthcare applications, fire detection


Reasons Of Energy Consumption

Energy consumption in a sensor node can be due to either Energy consumption in a sensor node can be due to either

““useful”useful” or or “wasteful”“wasteful” sources. sources.

Useful energy consumption can be due to

Transmitting/ receiving data, processing query

requests, and forwarding queries/data to neighboring

nodes.

Wasteful energy consumption can be due to

Idle listening to the media, retransmitting due to

packet collisions, overhearing, and generating/handling

control packets.


How To Reduce Energy Consumption ?Due To Wasteful Sources ?Due To Wasteful Sources ? Several MAC protocols attempt to reduce.

Due To Useful Sources ?Due To Useful Sources ? A number of protocols have also been proposed to

reduce useful energy consumption.

These protocols can be classified into three classes :These protocols can be classified into three classes :

First class controls the transmission power level at each node to increase

network capacity while keeping the network connected.

Second class makes routing decisions based on power optimization

goals.

Third class decides which nodes should participate in the network

operation (be awake) and which should not (remain asleep) . (nodes

require of locations knowledge via GPS-capable antennae / message

exchange).


Topology management

Cell-based approach Cluster-based approach

observer


How To Reduce Energy Consumption ? (2)Hierarchical clustering techniques can aid in reducing useful energy consumption.

Clustering is particularly useful for :Clustering is particularly useful for :

Applications that require scalability to hundreds or thousands of nodes. (need for load balancing and efficient resource utilization.)

Applications requiring efficient data aggregation

Routing protocols

One-to-many, many-to-one or one-to-all (broadcast) communication.

For example,For example, in many-to-one communication, clustering can support data fusion and reduce communication interference.


Clustering And Clusterheads

The essential operation in sensor node clustering is to :

Select a set of cluster heads among the nodes in the network.

Cluster the rest of the nodes with these heads.

Cluster heads are responsible for : Coordination among the nodes

within their clusters (intra-cluster

coordination).

Communication with each other and/or with external observers on behalf of their clusters (inter-cluster communication).

observer

CH CH

CH


Network Lifetime

What Is Network Lifetime ?What Is Network Lifetime ?

Time until the first node / the last node in the network depletes its energy (dies).

For example,For example, in a military field where sensors are monitoring chemical activity, the lifetime of a sensor is critical for maximum field coverage.

How To Prolong Network Lifetime ?How To Prolong Network Lifetime ?

1. Reducing the number of nodes contending for channel access,

2. Summarizing network state information and updates at the cluster heads through intra-cluster coordination,

3. Routing among cluster heads, which has a relatively small network diameter.


Clustering Can Reduce The Communication Overhead

Clustering can reduce the communication overhead for

both single-hop and multi-hop

networks.

1. Sensors periodically transmit information to a remote observer (base station).

2. With clustering, nodes transmit their information to their cluster heads.

3. A cluster head aggregates the received information and forwards it over to the observer.


Primary Goals Of HEED

1. Prolonging network lifetime by distributing energy consumption

2. Terminating the clustering process within a constant number of iterations / steps,

3. Minimizing control overhead (to be linear in the number of Nodes)

4. Producing well-distributed cluster heads and compact clusters.


HEED Assumptions

A set of n sensor nodes are dispersed uniformly and independently in a

rectangular field.

Sensor nodes are

1. quasi-stationary.

2. location-unaware, i.e. not equipped with GPS capable antenna.

3. equally significant (have similar capabilities (processing /

communication)).

4. left unattended after deployment.

5. Each node has a fixed number of transmission power levels.

6. The network serves multiple mobile/stationary observers, which

implies that energy consumption is not uniform for all nodes.


No Assumptions Are Made About :

1. Homogeneity of node dispersion in the field,

2. Network density or diameter,

3. Distribution of energy consumption among sensor nodes,

4. Proximity of querying observers.


HEED Requirements

1. Each node is mapped to exactly one cluster.

2. The node can directly communicate with its cluster head (via a single hop).

3. Clustering is completely distributed. Each node independently makes its decisions based on local information.

4. Clustering terminates within a fixed number of iterations.

5. At the end of each TCP, each node is either a cluster head, or an ordinary node that belongs to exactly one cluster.

6. Clustering should be efficient in terms of processing complexity and message exchange.

7. Cluster heads are well-distributed over the sensor field.


Approach

HEED is hybrid :Clustering is based on two parameters

HEED is distributed :Every node only uses information from its 1-hop neighbors (within cluster range)

HEED is energy-efficient :Elects cluster heads that are rich in residual energy


Parameters For Electing Cluster heads

Primary parameter : residual energy (Er)

Goal : Prolong network lifetime

Used to select initial clusterheads

Secondary parameter : communication cost

Goal : Increase energy efficiency and further prolong network lifetime

Used to break ties i.e., maximize energy and minimize cost

A tie means that :

A node falls within the range of more than one cluster head or,

Two tentative cluster heads fall within the same range.

Is a function of cluster properties, such as

Cluster size

If each node is allowed to use the minimum power level to reach its cluster head or if all intra-cluster communication must use the same power level.


Communication Cost Definitions

Cost Definitions

A node joins the CH with the min.

degree

Power : Fixed for

all nodes

Goal : Load

Distribution (Balancing)

AMRP (Average Min. Reachability

Power)

Goal : For minimum intra - cluster communication

energy

A node joins the CH with the max.

degree

Power : Fixed for

all nodes

Goal : For dense clusters

(AMRP) is the mean of the minimum power levels required by all M nodes within the cluster range to reach CH “u”. (a good estimate for the communication cost)


Probability Of Becoming A Cluster Head

Before a node starts executing HEED, it sets its probability of becoming a cluster head, CHprob, as follows :

CHprob = Cprob * ( Er / Emax )

Cprob : Initial percentage of cluster heads among all N nodes (say 5%)

Eresidual : Estimated current residual energy in the Node

Emax : A reference maximum energy (corresponding to a fully charged battery), which is typically identical for all nodes.

The CHprob value of a node is not allowed to fall below a certain threshold pmin (e.g., 10−4)


HEED – Algorithm at node v

I. Initialization

II. Main Processing

(Repeat)

III. Finalization

Discover neighbors within cluster range (Snbr)

Compute and broadcast cost to Snbr

Compute the initial cluster head probabilityCHprob = max(Cprob * (Er/Emax) , pmin)

If v received some cluster head messages, choose one head with min costIf v does not have a cluster head, elect to become

a cluster head with CHprob .

CHprob = min(CHprob * 2, 1)

Repeat until CHprob reaches 1

If cluster head is found, join its clusterOtherwise, elect to be cluster head


HEED – Analysis (1)

HEED has the following lemmas :

Lemma 1 : HEED terminates in Niter = O(1) iterations Brief Proof :

The worst case : low Eresidual. Then CHprob = pmin. However, CHprob doubles in every step, and phase II of the protocol terminates one step (iteration) after CHprob reaches 1

Lemma 2 : At the end of phase III of the HEED protocol, a node is either a cluster head or a regular node that belongs to a cluster.

Lemma 3 : HEED has a worst case processing time complexity of O(N) per node, where N is the number of nodes in the network .



Lemma 4 : HEED has a worst case message exchange complexity of O(1) per node, i.e., O(N) in the network.

An ordinary node is silent until it sends one join message to a cluster head.

The number of these join messages in the network is less than N, since at least one node will decide to be a cluster head during the clustering process.

Hence, the number of messages exchanged in the network is

upper-bound by Niter × N, i.e., O(N) since Niter is constant.



Lemma 5 : The probability that two nodes within each other’s cluster range are both cluster heads is small, i.e., cluster heads are well-distributed.

Consider the following worst case scenario :

Assume that v1 and v2 are two isolated neighboring nodes, each one does not have any other neighbor in close proximity.

We compute the probability, pnbr, that at the end of phase III, both of

them are cluster heads (assume that they are fully synchronized).

Assume that neither of the two nodes decides to be a cluster head before its CHprob reaches 1. Otherwise, one of them will concede to

the other.


Inter-Cluster Communication (1)

Rt : inter-cluster transmission range

Rc : the cluster transmission range

Lemma 6 : (Blough and Santi’02) Assume n nodes are dispersed uniformly and independently in an

area R=[0,L]2 Assume that the area is divided into square cells of size (Rc / √2) ×

(Rc / √2). If Rc

2n = aL2lnL, for some a > 0, Rc << L, and n >> 1, then

limn,N→∞E (number of empty cells) = 0, so each cell contains at least one node

Rc / √2

Rc / √2 Rc

L



2.7Rc

2.7Rc

Rt

CH1

CH2

Lemma 7 : There exists at least one cluster head in any (2 + 1/√2) Rc × (2 + 1/√2) Rc area.

Lemma 8 : For any two cluster heads v1 and v2 in two neighboring areas A and B of size (2+ 1√2) Rc × (2+ 1√2) Rc, v1 and v2 can communicate if Rt ≥ 6 Rc.



Theorem 1 : HEED produces a connected multi-hop cluster head graph (structure)

Proof (by contradiction) : Assume previous 3 lemmas hold.

Assume that HEED produces two connected components (graphs) of cluster heads. G1 = (V1,E1) and G2 = (V2,E2), such that any v1 ∈ V1 can not communicate with any v2 ∈ V2.

Assume that V2 lies on the right of V1, and that a cluster head v1 ∈ V1 lies on the rightmost border of V1.

v1 is able to communicate with a cluster head v2 on its right side, since the condition in Lemma 8 holds. v2 must reside inside V2, which contradicts with the initial assumption that a Cluster head in one component cannot communicate with one in the other component. Therefore, V1 and V2 are connected.


Performance Evaluation (1)

Simulation Environment1000 nodes uniformly spread across 2000 x 2000Minimum probability for becoming a cluster head (pmin) - 0.0005

Initial CHprob = Cprob = 5% and Cluster radius – 25 to 400 m.

Residual Energy levels – 20 and # of Experiments – 100

Comparison with generic weight–based clustering protocols (GC) WCA, DCA,

etc. becauseDistributed clustering only based on local informationSelected cluster heads with the highest weights (residual energy)A node has only one cluster headNo assumptions about node dispersion fieldNumber of iterations is function of network diameterTime and message complexities are O(N) and O(1)Guaranteed that no two cluster heads are neighbor


Performance Evaluation (2)

Number of iterations to terminate

Number of iterations in HEED can be deterministically computed using Lemma 1, which is independent of the cluster radius.

For GC, the number of iterations grows quickly as the cluster radius increases. (radius implies neigbors for each node )

GC takes only 3 iterations to terminate for a cluster radius of 25. The number of iterations, however, grows to 85 for a cluster radius of 400.

HEED takes 6 iterations to terminate for all cluster ranges.

(Cprob = 5%, Eresidual is close to E max)


Clusterhead Characteristics

(1)HEED cluster heads are comparable to those selected by GC in terms of number, distribution, and energy

availability.

HEED cannot guarantee optimal head selection in terms of energy, since it uses the secondary parameter to resolve conflicts.

GC, a weight-based approach, does guarantee that the highest energy node will be the cluster head within its cluster range.


Clusterhead Characteristics (2)

If it is required to balance load on cluster heads, then it is important to have clusters with small variance in the number of nodes they cover.

The maximum degree cost type and GC show similar results.

For minimum degree cost, the standard deviation is the lowest, because ties are broken by joining the smaller degree node, thus balancing the cluster sizes.

AMRP results lie between the two extremes. Therefore, AMRP provides a compromise between load balancing and cluster density.


Clusterhead Characteristics &

Node SyncronizationHEED produces a higher percentage of non-single node clusters than GC for all cost types.

The maximum number of nodes in a cluster in HEED is on the average smaller than that of GC for all cost types

Node synchronization is not critical for the operation of HEED. selected cluster heads in both cases have comparable residual energy.


Clustering Applications

Use of HEED forEnergy efficient routing protocolsEfficient Data Aggregation

Because prolonging network lifetime is especially important for unattended networks used in environmental monitoring.

HEED vs. gen-LEACHHEED clustering improves network lifetime over gen-LEACH clustering for all cost types.HEED expends less energy in clustering than gen-LEACH.HEED prolongs network lifetime, compared to gen-LEACH and to direct communication.


Conclusion And Future Work

Authors have proposed HEED clustering.

HEED is fast and has low overhead.HEED can provide features such as load-balancing.HEED is independent of :

Homogeneity of node dispersion in the fieldNetwork density or diameterDistribution of energy consumption among nodesProximity of querying observers

Future WorkExtend the protocol to multi-level hierarchies.Cluster size constraints in HEEDIncorporate multiple external mobile observers into HEED.

ubi532 - wsn spring 20091 distributed clustering in ad-hoc sensor networks : a hybrid,...

Documents

wasteful energy consumption

residual energy

sensor nodes

clustering nodes

detectionubi532 wsn

sensor networks overview

new energyefficient

useful sources