Stochastic Characterization of Mobile Ad-hoc Networks

1
INFORMS 2004
Stochastic Characterization of Mobile Ad-hoc
Networks
John P. Mullen and Timothy I. Matis Center for
Stochastic Modeling Department of Industrial
Engineering New Mexico State University
2
What Are MANETS ?

• A MANET is a mobile ad-hoc wireless communication
  network that is capable of autonomous operation
• Each node is capable of transmitting, receiving,
  and routing packets of information.
• The network has no fixed backbone
• The nodes are able to enter, leave, and move
  around the network independently and randomly

3
Mobile Ad Hoc Path Search
Y
X
4
Same MANET After a While
5
Nutshell

• MANET field performance differs greatly from
  simulations
• Field testbed performance is much poorer
• Developing MANET protocols in the field is very
  difficult
• Improving simulation fidelity increases the value
  of simulation in design.
• Higher fidelity earlier in the design process
  leads to better designs
• Research focus
• Significantly improve the fidelity of MANET
  simulations
• Without significantly increasing
• Simulation run time or
• Modeling effort.
• Research results
• Up to an order of magnitude improvement in
  fidelity
• Runtime increases are often insignificant, but
  generally less than 100
• Very little added modeling effort

6
Overview

• Multipath Fading and its impact on mobile ad hoc
  nets
• The Stochastic Model
• Objectives
• Implementation
• Validation
• Demonstrations of the Model
• Small Models
• Impact of Short Retry Limit (SRL)
• Comparing AODV and DSR
• Large Models
• AODV vs. DSR
• AODV vs. DSR using GPS data
• Impact of SRL on DSR
• Summary, Conclusions and Further Work

7
Shadowing and Fading

• Shadowing
• Is caused by objects absorbing part of the signal
• Can be estimated by looking at the Line of Sight
  (LOS) path
• Causes a random reduction in signal strength.
• Fading
• Is the result of the algebraic sum of signals
  from many paths
• Because movement of any object in the vicinity
  can change the sum
• Multipath fading is extremely difficult to model
  and predict
• Would be very time consuming to simulate exactly
• And would have little predictive value.
• This phenomena causes
• Very rapid large-scale fluctuations in signal
  strength
• Can cause the signal to be significantly lesser
  or greater than expected.

8
Main causes of signal variation
T
Shadowing
R
Multipath
9
Measured Received Signal Strength(from Neskovic
2000)
10
Stochastic Variation Model

• The Model
• Given mp(d), the expectation of power at distance
  d
• Rayleigh fading model of the instantaneous power,
  P(d)
• Pr P(d) p 1 exp-p/mp(d)
• Inverse transform of the Rayleigh fading model
• P(d) -mp(d)ln(1-r)

11
Simulated vs. Real Power
Actual Measurements
Simulated Values
12
Validation

• Simulated reported field tests and compared
  results
• K.-W. Chin, J. Judge, A. Williams, and R.
  Kermode, "Implementation experience with MANET
  routing protocols," ACM SIGCOMM Computer
  Communications Review, vol. 32, pp. 49 - 59,
  2002.
• I. D. Chakeres and E. M. Belding-Royer, "The
  Utility of Hello messages for determining link
  connectivity," Wireless Personal Multimedia
  Communications, vol. 2, pp. 504 - 508, 2003.
• D. S. J. D. Couto, D. Aguayo, J. Bicket, and R.
  Morris, "A High Throughput Path Metric for
  MultiHop Wireless Routing," presented at MobiCom
  '03, San Diego, California, USA, 2003.
• S. Desilva and S. Das, "Experimental evaluation
  of a wireless ad hoc network," 2000.
• Simulations with
• Standard non-fading model were exceedingly
  optimistic
• Proposed fading model were very much more
  realistic.

13
Impact of Multipath Fading on MANETs

• How does it affect MANETs?
• Unnecessary route searches
• Selection of false routes

14
Impact of Multipath Fading On MANETs
False Routes
OK
Dropped Packets
OK
15
Impact of Multiple Retries on MANETs
16
The MANET fading Trade-off
Increase Risk of Selecting Bad Routes
Improve Reliability On Good Routes
Protocol
MANET Nominal range is a matter of balance.
Most Wireless Nominal range is a matter of
design.
17
Demonstrations

• Small Models (validations of field tests)
• Scenario 1 Performance vs. distance.
• Used for the two cases above
• Scenario 2 Routing Test
• Focus mainly on fading effects
• Models
• Fading vs. nonfading simulations of AODV
• DSR vs. AODV with fading model
• Large models (exploration)
• Scenario 3 24 nodes.
• Also consider other effects, such as interference
• Models Fading and non-fading versions of
• AODV vs. DSR
• AODV vs. DSR using GPS data
• Impact of SRL on DSR

18
Scenario 2 Routing test(from Chin et. al., 2002)
r0 39m
10 pps
0.5 m/s
19
Sc 2 Fading vs. Nonfading AODV
Notes Default values for AODV SRL 7
20
Sc 2 AODV vs. DSR
Notes Default protocol values SRL 7
Nonfading model shows no difference
21
Scenario 3 Larger Scale Test

• Features
• More nodes (24)
• Random r-t pairs
• Interference
• Higher loads

22
Mean Throughput AODV vs. DSR

• Notes
• Default protocol values
• SRL 7

23
Mean Delay AODV vs. DSR

• Notes
• Default protocol values
• SRL 7

24
Using GPS data
2
Use GPS to block unreliable routes
1
B
r0
3
25
Impact of GPS
Without GPS
With GPS
26
Mean Throughput Impact of SRL on DSR

• Notes
• Default protocol values

27
Mean Delay Impact of SRL on DSR

• Notes
• Default protocol values

28
Execution Time in Scenario 3 (Virtually no
differences in Scenarios 1 2)
29
Summary

• Non fading model
• Overestimates field performance
• Is very insensitive to all the contrasts shown
  here and more.
• Fading model
• Provides more realistic estimates
• Better predicts impacts of protocol and parameter
  changes
• Shows promise of new techniques.
• Requires little or no additional modeling
• Has little impact on execution time
• (Alternative is a testbed or a field trial)

30
Conclusions

• Multipath Fading
• has a great impact on mobile ad hoc nets
• Including its effects in simulation
• greatly improves fidelity
• Stochastic Modeling of Multipath Fading
• Is a practical way to include the impact of
  fading
• Minor modifications to code (in OPNET, at least)
• Without great increases in
• Modeling effort or
• Execution time

31
Future Work

• More Fading Models
• Rayleigh
• Ricean
• Nakagami
• Other significant RF effects
• e.g. exponential decay factor
• Better user interface
• Allow selection of models parameters without
  need to recompile.
• Validation
• Replicating published studies
• Set up own testbed and field trials
• Better modeling of fading impacts
• Hello vs. control vs. data packet results
• Other significant measurable elements.

32
Acknowledgements

• OPNET Technologies
• Software license research grant
• Technical assistance
• Center for Stochastic Modeling
• Technical resources
• Klipsch School of Electrical and Computer
  Engineering
• Dr. Steve Horan
• Dr. Hong Huang (also CSM member)

33
Final Questions?