[PPT] - The Impact of Directional Antenna Models on Simulation Accuracy PowerPoint Presentation

SLIDE 1

The Impact of Directional Antenna Models on

Simulation Accuracy

Eric Anderson, Gary Yee, Caleb Phillips, Douglas Sicker, and Dirk Grunwald eric.anderson@colorado.edu

University of Colorado Department of Computer Science

25 June 2009

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 1 / 21

SLIDE 2

Outline

1

Introduction Physical Layer Simulation Current Models of Directivity

2

EDAM – The Effective Directivity Antenna Model Error Idea Parameters

3

Case Study Overview Metrics Results

4

Conclusions

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 2 / 21

SLIDE 3

Phy-Layer Simulation Framework

Two−ray COST 231 ITU1238

Path Loss

... ...

Fading

Rician Rayleigh

Directivity Model

EDAM Antenna Gain Only None

Environmental Factors

In/Outdoors Line of Sight Motion In/Outdoors Terrain Line of Sight

Position Direction / Orientation Temporal Variation

Antenna Properties

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 3 / 21

SLIDE 4

Directivity – Current Models

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 4 / 21

SLIDE 5

Directivity – Current Models

Fading & path loss Node a gain Node b gain Prx = Ptx ∗ X ∗ fa(θ1) ∗ fb(θ2)

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 4 / 21

SLIDE 6

Example
Eric Anderson (CU Boulder)

The Impact of Directional . . . WiOpt ’09 5 / 21

SLIDE 7

Example
fa(θ1)

fb(θ2)

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 5 / 21

SLIDE 8

Example
fa

( θ3 ) f

b

( θ

4

)

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 5 / 21

SLIDE 9

Example
Eric Anderson (CU Boulder)

The Impact of Directional . . . WiOpt ’09 5 / 21

SLIDE 10

Outline

1

Introduction Physical Layer Simulation Current Models of Directivity

2

EDAM – The Effective Directivity Antenna Model Error Idea Parameters

3

Case Study Overview Metrics Results

4

Conclusions

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 6 / 21

SLIDE 11

How Bad Is It?

25 50 75 105 140 175 210 245 280 315 350 −50 −40 −30 −20 −10 10 Patch−Panel Antenna Angle, Degrees Counterclockwise dB Relative to Peak Mean

Patch−Outdoor−B Patch−Outdoor−A Patch−Indoor−A Reference

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 7 / 21

SLIDE 12

How Bad Is It?

21 50 76 106 140 171 205 236 270 301 335 −50 −40 −30 −20 −10 10 24dBi Parabolic Dish, Indoors Angle, Degrees Counterclockwise dB Relative to Peak Mean

Parabolic−Indoor−C Parabolic−Indoor−B Parabolic−Indoor−A Reference

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 7 / 21

SLIDE 13

EDAM – The Effective Directivity Antenna Model

Key Idea:

Model offset between the expected (“pure”) antenna gain and observed effect. Offset is environment-specific and impractical to compute

(but easy to measure)

Distribution of offsets can be predicted well. Construct distributions, sample, repeat.

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 8 / 21

SLIDE 14

EDAM Distribution Parameters

Mean offset is based on antenna gain and environment type. Offset variance and packet signal variance are based on environment type. See “Modeling environmental effects on directionality in wireless networks,”

WiNMee 2009 (tomorrow).

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 9 / 21

SLIDE 15

Outline

1

Introduction Physical Layer Simulation Current Models of Directivity

2

EDAM – The Effective Directivity Antenna Model Error Idea Parameters

3

Case Study Overview Metrics Results

4

Conclusions

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 10 / 21

SLIDE 16

Case Study: Data Striping for Security

Compare simulated and real results

(S. Lakshmanan et. al, 2008)

Propagation-sensitive system:

APs beam-form to client Packet must be received from all APs to decode.

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 11 / 21

SLIDE 17

Experimental Design

Directivity Path Loss Fading EDAM Two ray Log-normal Pure ITU-1238 Ricean None/Omni Implicit Gaussian None

Directivity Null Hypotheses

1 “Pure:” Antenna gain fully describes directional effects. 2 “None/Omni:” There is no predictable directional effect. (indoor) Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 12 / 21

SLIDE 18

“Ground Truth” Measurement

5 10 15 20 25 30 5 10 15 20 25 30 35 Measurement Points x coordinate y coordinate X position 5 10 15 20 25 30 Y position 10 20 30 Signal strength, dBm −100 −80 −60 −40 −20 Signal strength from 192.168.99.2

foo

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 13 / 21

SLIDE 19

Application Performance: Vulnerability Area

Application goal: Minimize physical area of reception. Metric: Fraction of packets decodable at each location. Distribution, not scalar.

0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 1.0 CDF of packet reception probability by location, outdoor Fraction of packets received completely Cumulative fraction of nodes Observed Sim: NA, NA, NA

Actual performance, outdoor test

Complete packet reception probability

x coordinate y coordinate

10 20 30 5 10 15 20 25 30 0.0 0.2 0.4 0.6 0.8 1.0

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 14 / 21

SLIDE 20

Simulation Accuracy: Distribution Similarity

Accuracy = similarity of application performance to reality. Kolmogorov-Smirnov (KS) test: Maximum divergence between distributions. Evaluation: Factorial ANOVA on KS test across all configurations.

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

CDF of packet reception probability by location, outdoor

Fraction of packets received completely Cumulative fraction of nodes Observed Sim: pure, two ray, lognormal

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 15 / 21

SLIDE 21

Results (subset)

EDAM Pure Gain Indoor Outdoor

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 16 / 21

SLIDE 22

50% Vulnerability Region – Indoor

Directivity Model Area (points) Pure antenna 3 - 5 Measured 38 EDAM 54 - 79 Omni (no directionality) 83 (100%)

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 17 / 21

SLIDE 23

Outline

1

Introduction Physical Layer Simulation Current Models of Directivity

2

EDAM – The Effective Directivity Antenna Model Error Idea Parameters

3

Case Study Overview Metrics Results

4

Conclusions

Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 18 / 21

SLIDE 24

Highlights

EDAM enables more accurate simulation and modeling of networks with directional antennas.

Dramatic improvement indoors Marginal improvement outdoors

Directional effects are significant – even in worst case. Determinants of accuracy: (indoor):

1 Directivity model ≫ 2 Path loss model ≫ 3 Fading model

(outdoor):

1 Path loss model > 2 Fading model ≫ 3 Directivity model. Eric Anderson (CU Boulder) The Impact of Directional . . . WiOpt ’09 19 / 21

SLIDE 25

Thank you

Contact: eric.anderson@colorado.edu Simulation software (Qualnet 4.5.1 patch): http://systems.cs.colorado.edu/ Raw measurements: http://www.crawdad.org/cu/antenna

SLIDE 26

KS−Test Statistic For all Configurations and Seeds

Kolmogorov−Smirnov Test Statistic

0.2 0.4 0.6 0.8 1.0 EDAM−ITU 1238−Ricean EDAM−ITU 1238−implicit Gaussian EDAM−ITU 1238−lognormal EDAM−two ray−Ricean EDAM−two ray−implicit Gaussian EDAM−two ray−lognormal

mni−ITU 1238−Ricean
mni−ITU 1238−lognormal
mni−ITU 1238−none
mni−two ray−Ricean
mni−two ray−lognormal
mni−two ray−none

pure−ITU 1238−Ricean pure−ITU 1238−lognormal pure−ITU 1238−none pure−two ray−Ricean pure−two ray−lognormal pure−two ray−none

indoor

0.2 0.4 0.6 0.8 1.0

utdoor