Power Optimisation on Wireless Adhoc Networks

1
Power Optimisation on Wireless Ad-hoc Networks
Department of Computing, Imperial College London

• Aris Papadopoulos
• ap7_at_doc.ic.ac.uk
• Supervised by Dr Julie A. McCann
• 2nd supervisor Dr Naranker Dulay

Transfer Presentation Monday 21 February 2005
2
Overview

• Ad-hoc, wireless and the role of energy.
• Background on Minimum Energy Broadcast Routing
  (MEBR).
• A new protocol for the MEBR problem (ABC).
• The protocol.
• Experimental results.
• A correction to a literatures result.
• Background of routing over Wireless Sensor
  Networks (WSNs).
• Design aims of a WSN.
• A new protocol for routing over WSNs (CoP).
• The protocol.
• Experimental results.
• Conclusion.
• Future work.
• Questions.

3
Ad-hoc vs. fixed infrastructure
i

• Motivation
• Emergency relief, battlefield, surveillance,
  environmental monitoring etc.
• Communication mode
• Multihop vs. direct transmission.

d
s
The energy spent on a transmission is
proportional to the square distance between the
nodes.
4
Wireless vs. wired networking

• Dynamic environment.
• Links depend on
• Distance,
• Transmission power capabilities,
• Error control,
• Interference,
• Background noise.
• Need of different modelling.

5
The role of energy

• Power awareness as the main means of securing
  network longevity.
• Critical due to the nature of the applications.
• Small mobile devices deployed for extended
  periods with no infrastructure support.
• Need of power aware algorithms to ensure
  longevity.

6
Minimum Energy Broadcast Routing (MEBR)

• Given
• Nodes can adjust their transmission power as
  appropriate.
• Every node is assigned a transmission range and
  every node within this range receives its
  message.
• Find
• An assignment of transmission ranges such that
  the total energy that is consumed across the
  network is minimum.

7
Minimum Spanning Tree (MST)

• Among the trees that span all vertices of a
  weighted graph, the one(s) with minimum total
  weight is(are) called Minimum Spanning Tree(s).
• Prims algorithm
• let T be a single vertex x
• while (T has fewer than n vertices)
• find the smallest edge connecting T to G-T
  add it to T

8
Minimum Spanning Tree (MST)
Initial graph edges Light edges forming the
MST Cuts
5
0
4
Edges connect in-range neighbours. Their weight
is proportional to the distance (shorter edge
means smaller weight).
7
3
6
9
9
Minimum Spanning Tree (MST)

• MST fails to identify an important property of
  the wireless environment.
• Establishing a link may cover multiple nodes.

A bad instance for MST Receiving nodes on a
hexagon topology around the source. Approximation
ratio 6
10
Broadcast Incremental Power (BIP)

• Introduces the circle-concept (node-based
  rather than link-based).
• Starts from the source and builds a
  minimum-energy tree.
• Adds the cheapest from the set of uncovered
  nodes.
• Either adds a new edge or increases an old one.

11
Broadcast Incremental Power (BIP)
Checks from all uncovered nodes (3, 5, 6, 7, 9)
the cheapest to connect to any of the covered (0,
4)
Checks the cheapest to connect from the source
(0)
Checks from all uncovered nodes (3, 5, 6, 9) the
cheapest to connect to any of the covered (0, 4,
7)
Step 1
Step 3
Step 2
Step 4
BIP pays for two connections 0 to 3 and 7 to
9.
5
Node 4, incremental cost c04
0
Node 7. Connecting from 4, incremental cost c47.
Connecting from 0 incremental cost c07-c04
Node 9. Connecting from 4, incremental cost c49.
Connecting from 0 incremental cost c09-c07.
Connecting from 7, incremental cost c79
4
7
3
6
Note Edges point to connected nodes, they dont
represent actual links. The Links are represented
by circles and are associated with specific costs.
9
12
Adaptive Broadcast Consumption (ABC)

• Incorporates backtracking.
• Highly adaptive.
• Uses Prims discovery order.
• Removes previous links to nodes that are
  redundantly covered due to a new nodes discovery.

13
Adaptive Broadcast Consumption (ABC)

• Main differences between ABC and BIP
• ABC uses Prims order. BIP uses next-cheapest
  strategy.
• ABC uses a complex adaptive backtracking method
  to clear redundant links. BIP can either increase
  an existing edge or add a new one.
• Main similarities shared by ABC and BIP
• Both are based on the circle concept
• Complexity O(n3).

14
Adaptive Broadcast Consumption (ABC)
After step-4 BIP pays for two connections 0-3
and 7-9. So the cost at this stage is c03c79.
5
ABC instead would be able to eliminate connection
7-9 since node-9 is covered by connection 0-3.
Therefore its cost would be just c03.
0
4
7
3
6
Note Edges point to connected nodes, they dont
represent actual links. The Links are represented
by circles and are associated with specific costs.
9
15
Adaptive Broadcast Consumption (ABC)

• How is this achieved
• ABC guarantees a valid solution by maintaining
  two invariants. Denoting with i the i-th node
  discovered at step i (source0) then
• Every node from 1 to i-1 is covered by some
  circle and
• Every node from 1 to i-1 admits an induced path
  back to the source (0).

16
Adaptive Broadcast Consumption (ABC)
At most n-1 times
Exactly n-1 times
At most n-1 comparisons
17
Adaptive Broadcast Consumption (ABC)

• For each new circle (link) from the set of the
  covered nodes find the subset that is covered now
  also by the new circle.
• For each node of this subset check from which
  circle(s) this node was covered, before the new
  circle was added.
• For each such circle check whether it links
  other nodes as well.
• If not check if deleting it, all nodes retain an
  induced path back to the source.

18
A comparison
The cost of each solution is proportional to the
surface that its circles cover. Edges can be read
as connected from.
BIP
ABC
Optimised MST
19
Experimental results

• Comparing MST, BIP and ABC.
• Random instances of various sizes ranging from 5
  to 50 nodes in a 2D area of 5x5 units.
• 500 random instances for each size.
• The metric used for the comparison is the cost of
  the resulting coverage in terms of energy.
• Some of the obtained results follow.

20
Experimental results
Total energy spent J
Figure1 7-nodes instances
Instance(x10)
21
Experimental results
Total energy spent J
Figure2 15-nodes instances
Instance(x10)
22
Experimental results
Total energy spent J
figure3 25-nodes instances
Instance(x10)
23
Experimental results
Total energy spent J
Figure4 45-nodes instances
Instance(x10)
24
Experimental results

• MST up to 2 times more expensive than BIP and
  ABC.
• ABC up to 5 more efficient than BIP.
• ABC increasing relative efficiency as instances
  grow.

25
Adaptive Broadcast Consumption (ABC)
Explaining the few instances where ABC is
outperformed
5
5
2
S
1
3
S
1
3
2
4
4
ABC
BIP
26
Adaptive Broadcast Consumption (ABC)

• The worst instance for ABC.
• Optimal cost 2(1e)2
• ABC (and BIP) cost 46e
• (approximation ratio 2)
• Optimised MST cost 66e

27
Approximation ratios

• Since ABC uses Prims discovery order, its
  approximation ratio ?(ABC) is at most ?(MST).
• Therefore (due to the instance of the previous
  slide and this last argument) 2lt?(ABC)lt?(MST).
• From literature ?(MST)lt12. (43WAN, CALINESCU,
  LI, FRIEDER. Minimum Energy Broadcast Routing in
  static, ad hoc wireless networks. Wireless
  Networks 8 (2002), 607, 617).

• Correction ?(MST)lt12.15 (proof provided in the
  paper in which we introduced ABC).

• Therefore 2lt?(ABC)lt12.15.

28
Wireless Sensor Networks (WSNs)

• Sensor board
• Sensors of different modalities (motion
  detectors, light sensors, thermometers,
  accelerometers).
• Microprocessor.
• Low power radio transceiver.

• Motivation
• Smart Dust (Battlefield, emergency relief etc).
• Ubiquitous computing (Smart environments, houses,
  hospitals etc).

29
A WSN example
30
Important Issues

• Energy efficiency longevity
• Smaller devices, prolonged life requirements.
• Scalability
• Large numbers of deployed devices.
• Mobility
• High mobility (e.g. environmental monitoring).
• Fault-tolerance
• Frequent failures due to size restrictions and
  lifetime requirements.

31
Important Issues

• Identification
• Large number of deployed devices and the
  non-end-to-end nature of the apps lead to
  avoidance of unique IDs.
• Cross layer design
• Application level design decisions directly
  influence substrate layers.
• Location-awareness
• Nodes are in some cases equipped with GPS
  receivers.

32
Application-level decisions

• Continuous
• Event-driven

• Demand-driven
• Hybrid

33
Routing for WSNs

• Directed Diffusion
• Low-Energy Adaptive Cluster Hierarchy (LEACH)
• Sensor Protocols for Information via Negotiation
  (SPIN)
• Geographical and Energy Aware Routing (GEAR)
• SPEED
• Sensor Protocol for Energy-Efficient Data routing
  (SPEED)
• Energy Aware Data Centric Routing (EAD)

34
Directed Diffusion

• Demand-driven.
• The sink floods the WSN with interests.
• The neighbour that has returned data first is
  reinforced.
• Highly adaptive as it is based on local empirical
  decisions.
• Main concern deliver fast.

35
LEACH

• Cluster organisation.
• Clusterheads are elected at each round.
• Rotation mechanism for fairness.
• Clusterheads are responsible for data aggregation
  and direct transmission to the sink.

36
LEACH

• Energy spent on the transmission of a k-byte
  message.

Transmit electronics
Receive electronics
Transmission power
Tx
Rx
(A factor a induced by the electronics)
(A factor a induced by the electronics)
(A factor b induced by the electronics)
37
LEACH

• The idea behind LEACH
• Direct transmission can be more energy efficient
  than multi-hop!
• Energy spent on the transmission of a k-bit
  message

Base station
EdETP(d)
Tx
Emh4ET4P(d/4)3ER
Key a, b the power factors, r the distance
between the intermediate nodes, n the number of
the nodes.
EhETP(d/4)ER
EhETP(d/4)ER
38
LEACH

• In reality
• This scales really poorly.
• The parameters used in the experiments conducted
  were tuned in favour of the protocol.
• More details in the report!

39
SPIN

• Meta-data are used to advertise actual data.
• Simple three stage handshaking.
• SPIN-2 allows nodes to participate in the
  protocol only if their energy level is above a
  defined threshold.

40
Connectionless Probabilistic (CoP) Routing The
idea

• Construction of a virtual infrastructure based on
  a grid.
• Starting from a point of reference p, the grid is
  constructed using the grid unit vector u.
• Each intersection of the grid lies distance uu
  from its neighbours.
• Knowing these two parameters, all nodes are aware
  of all intersections positions.

41
Connectionless Probabilistic (CoP) Routing The
idea

• A circular area of radius ds is associated to
  each intersection.
• All nodes residing inside such an area become
  members of the virtual infrastructure
  (clusterheads).
• Clusterheads are responsible for routing
  messages.
• All other nodes are responsible for sensing and
  sending data to their closest clusterheads.

42
Connectionless Probabilistic (CoP) Routing
SINK
ds
p
u
43
Connectionless Probabilistic (CoP) Routing
44
Connectionless Probabilistic (CoP) Routing

• From the literature throwing randomly n points
  in a unit square, the probability that no points
  are inside a circle of radius
• is less than or equal to
• We can therefore adjust ds so as to make
  delivery almost certain.

45
Connectionless Probabilistic (CoP) Routing The
advantages

• Makes no use of any control messages.
• Performs better in highly mobile environments!
• In static environments, enforces a periodic
  altering of parameters (eg. the grid constructor
  vector) to ensure fairness.
• Enables further energy conservation by allowing
  non-clusterheads to switch off their
  transceivers, functioning only as sensors.

46
Connectionless Probabilistic (CoP) Routing The
advantages

• Solves the problem introduced by LEACH, as the
  constructor vector can be tuned according to the
  expected network density, so that the number of
  transmissions close to the optimal occurs at each
  step.
• Enables adjustable sensing granularity, according
  to the size of the grid.
• Enables deployment of more sensors or merging
  networks while in operation.

47
Connectionless Probabilistic (CoP) Routing The
open question

• The cost of ad-hoc localisation.
• Existing systems and on going research
• Ad-hoc Positioning System (APS).
• GPS-less low-cost outdoor localisation for very
  small devices.
• Ad Hoc Localization System (AHLoS) (Medusa
  Platform, UCLA).

48
Experimental Results

• Choice of protocols to compare
• Well known algorithms.
• Assuming position knowledge for comparison
  fairness.
• Directed flooding (DF)
• Real-time routing with no use of control
  messages.
• Upon receiving a message, a node forwards it
  further, if it is closer to the destination than
  the sender.

49
Experimental Results

• Greedy forwarding (GF)
• Request-respond to obtain location information
  and adjustment of the transmission radius so as
  to reach the neighbour that is closer to the
  sink.
• Parameters tuned in favour of DF and GF
• DF Transmission radius fixed in the minimum
  value that guarantees the same delivery
  probability as CoP.
• GF Request transmission radius fixed in the
  minimum value that guarantees the same delivery
  probability as CoP.

50
Experimental Results

• Energy spent on
• Transmitter and receiver electronics 5nJ/bit.
• Energy to support the link 100pJ/bit/m2.
• Instance
• 200 random nodes.
• 5x5m2 area.
• Directed Flooding
• Fixed radius 0.6m
• Cost 6.21mJ

51
Experimental Results

• Energy spent on
• Transmitter and receiver electronics 5nJ/bit.
• Energy to support the link 100pJ/bit/m2.
• Instance
• 200 random nodes.
• 5x5m2 area.
• Greedy Forwarding
• Maximum radius 0.8m
• Cost 1.02mJ

52
Experimental Results

• Energy spent on
• Transmitter and receiver electronics 5nJ/bit.
• Energy to support the link 100pJ/bit/m2.
• Instance
• 200 random nodes.
• 5x5m2 area.
• CoP
• ds 0.25m
• Cost 0.54mJ

53
Experimental Results
nJ
Directed Flooding
Greedy Forwarding
Connectionless Probabilistic Routing
x100 nodes
54
Experimental Results
Connectionless Probabilistic Routing
Greedy Forwarding
Control messages
Data messages
55
Conclusion

• Thorough analysis of the nature of the wireless
  ad-hoc environment.
• Proposed two novel protocols for two different
  classes MANETs and WSNs.
• Presented theoretical results.
• Presented experimental results showing
  substantial improvements comparing to established
  algorithms.

56
Future directions

• Formalise a general model for energy consumption
  over ad-hoc networks.
• Applying optimisation algorithms such as genetic
  and simulated annealing.
• Ad-hoc localisation.
• Ad-hoc networking over Bluetooth.

57
Questions

• Aris Papadopoulos
• ap7_at_doc.ic.ac.uk
• Supervised by Dr Julie A. McCann
• 2nd supervisor Dr Naranker Dulay
• PhD transfer presentation
• February 21st, 2005

Department of Computing Distributed Software
Engineering