19
Tufts Wireless Laboratory Tufts University School Of Engineering Energy-Efficient Structuralized Clustering for Sensor-based Cyber Physical Systems Jierui Cao, Huan Li

Tufts Wireless Laboratory Tufts University School Of Engineering Energy-Efficient Structuralized Clustering for Sensor-based Cyber Physical Systems Jierui

Embed Size (px)

Citation preview

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Energy-Efficient Structuralized Clustering for

Sensor-based Cyber Physical Systems

Jierui Cao, Huan Li

Tufts Wireless Laboratory

Tufts University

School Of Engineering

• Wireless sensor network is one of the most important cyber physical systems

• Clustering is an effective technique to improve the energy efficiency and network lifetime.

• The one-hop transmission from the heads to the data collection center is not scalable.

• The multi-hop transmission approach may lead to “hot spot”

Introduction

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Related Works• LEACH is an application-specific data

dissemination protocol that proposed to use clustering to save energy and prolong the network lifetime for sensor networks.– LEACH-C– DCHS– HEED– EEUC

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Structuralized Clustering• Structure a large scale sensor network with small equal

units. • The number of clusters are calculated according to the

workload of the heads and the distance to the base station.

• The cluster size should be unequally assigned so that the unit close to the base station should have more clusters.

• Adopt energy-aware head reelection strategy and relay algorithm for inter-cluster transmission.

• Demonstrate that the proposed cluster architecture and head selection algorithm have good performance in terms of scalability and energy efficiency.

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Network Model• Nodes are uniformly distributed • One fixed base station far away from the network center • All nodes have the same processing capacities, including CPU,

storage and initial energy• The power supply of nodes are limited • The nodes’ positions in the network are fixed • Assume that the whole area is divided by units with equal size, and

clusters are constructed within a given unit. • All members will report directly to the corresponding cluster head

using TDMA schedule under one-hop transmission. • For inter-cluster communication, cluster heads forward their data to

the head in the neighboring unit, and eventually to the base station.

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Network Model

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Energy Model• In sensor nodes, data transmission consumes much

more energy than sensing and data processing.• Only the energy consumption for data transmission are

considered, the overhead introduced by the routing and MAC layer are ignored

• Energy Model: transmitter dissipates energy to run the radio electronics and the power amplifier, and the receiver dissipates energy to run the radio electronics

• Assume the intra-cluster communication is running at the distance less than d0, and the inter-cluster relay distance is larger than d0

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Energy Model (continues)

• Where Eelec is the electronics energy, εfsd2 and εmpd4 are the amplifier energy and d0 is the distant threshold

• To receive this message:

• To transmit a k-bit message at distance d:

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Network Partition• To achieve even energy depletion and mitigate

“hot spot” problem.• The units that are closer to the base station

should have more clusters• Assume that heads do perfect data aggregation

in this model, where each head aggregates the received member data to a single packet

• Assume that head election algorithm and routing algorithm can guarantee each node in the unit will have equal chance to relay data

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Network Partition (continues)• Assume there are j units and the number of

nodes in the ith unit is ni. • For heads in unit_i (1<=i<j), the energy

consumed by each head:

where Ci is the cluster heads number, di is the inter-cluster transmission distance, and EDA is the cluster head aggregation energy dissipation for each bit.

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Network Partition (continues)• The energy consumed by each head in unit_j

• Each cluster head in the network should have the same energy consumption:

• Once the cluster member and heads in the first unit or the last unit in the network is given, the heads of clusters in other units can be calculated accordingly.

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Cluster operations• Base station could just randomly select the

cluster heads for all units at the initial phase• Assume the base station will ensure that all

nodes will be covered by at least one head• Base station will broadcast an initial message

containing the number of clusters and the head IDs for each unit in the whole network

• Cluster set-up phase, data relaying phase, and cluster re-selection phase

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Cluster set-up phase• Assumed nodes in different units are able

to know their home unit by their own IDs• After initial message, head broadcast a

message to notify other nodes near around for cluster set-up

• All nodes except the heads will need to choose joining one of the heads in the unit

• Use signal strength and the distance as deciding factor

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Data relaying phase• The relay node candidates are the set of

cluster heads in the next hop, or unit.

• A head forward its aggregated data to the head that has the largest amount of remaining energy in the next unit.

• The intrareporting and inter-relay process will last till the remaining energy in one of the heads decreases below a given threshold or till the end of a round

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Cluster head re-selection phase• The time interval for each round should

last at least till the farthest cluster finishes reporting data to the base station

• Each node will broadcast its remaining energy in turn

• Nodes with the highest remaining energy become heads

• System goes into the cluster set-up phase and continues a new round.

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Simulation• Using NS2• Simulate with a rectangle sensing field that is divided into

equal units• Each unit in the network is a square area with 100 nodes

deployed; the size of the unit is 62m*62m• The initial energy of each node is 2J• d0 is 87m, Eelec is 50nJ/bit• εfs is 10pJ/bit/m2, εmp is 0.0013pJ/bit/m4

• EDA is 5nJ/bit/signal• Data packet size is 525 bytes (sensing data 500 bytes

and the head overhead 25 bytes)

Tufts Wireless Laboratory

Tufts University

School Of Engineering

• At the initial stage, when the head aggregation times increases, the network lifetime increases

• At about 50 – 150, reaches the maximum value

Effect of head reelection frequency

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Head number and network lifetime

• Network is composed of four units and the head aggregation time is set as 50

• The network lifetime has the longest value when the head number is about 10% of nodes

Tufts Wireless Laboratory

Tufts University

School Of Engineering

Remaining energy