Otto, Bustamante & Berry

klapdorothypondMobile - Wireless

Nov 23, 2013 (3 years and 8 months ago)

59 views

http://aqualab.cs.northwestern.edu

John Otto
,
Fabián

Bustamante & Randall Berry

EECS, Northwestern University

Otto, Bustamante & Berry

2

Size
-

and power
-
unlimited mobile network platform


Infrastructure
-
less


Mobility facilitates rapid information dissemination


Many promising applications


Traditional Internet access


Environmental sensing


Traffic advisory and driver safety


Challenging environment


Rapidly changing topology


Network density depends on vehicular density

Down the Block & Around the Corner

Otto, Bustamante & Berry

3

Live experimentation


Viable when a few nodes are enough


OK for a proof of concept


Not an option with 100’s of vehicles


Simulation
-
based experimentation and its risks


No agreed
-
upon platform


Vehicular mobility


Traces and models


Signal propagation


Trading scalability and realism

Down the Block & Around the Corner

Otto, Bustamante & Berry

4

Performance of the network stack’s physical layer
defines the boundaries of a system’s ability

… and your environment determines the performance
of the physical layer








How does this impact our applications’ performance?



Signal propagation varies
widely between open field
and urban settings

Down the Block & Around the Corner

Otto, Bustamante & Berry

5

Challenging assumptions


Kotz et al. (2004)

Opportunistic connectivity


Ott & Kutscher (2004)


Wu et al. (2005) (multi
-
hop V2V)


Bychkovsky et al. (2006)


Hadaller et al. (2007)

Varied environments


Singh et al. (2002)

DSRC 5.9 GHz band


Taliwal et al. (2004)


Cheng et al. (2007)

We focus on


Vehicle
-
to
-
vehicle (V2V)


Varied environments


Line
-
of
-
sight (LOS)
versus non
-
LOS
communication

Down the Block & Around the Corner

Otto, Bustamante & Berry

6

Deterministic models


Free space and two
-
ray ground


Ideal LOS (and ground reflection) signal strengths


Do not account for variations in environment


Empirical models


Based on measurements taken in an environment


Ray Tracing
1


Requires detailed knowledge of the environment


Incurs significant computational cost


Does not scale


Probabilistic empirical model


Two parameters used to describe the environment


Typically

a good compromise between realism, scalability

1
McKown & Hamilton. “Ray tracing as a design tool for radio networks.” 1991.

Down the Block & Around the Corner

Otto, Bustamante & Berry

7

)
,
0
(
log
10
4
log
10
2
0
10
0
10
dB
N
d
d
d
PL




























Parameters


Path Loss Exponent (
β
)
: environment decay rate


Shadowing (
σ
dB
)
: variation due to obstacles


Can complex environments be modeled
using just two parameters
?

Down the Block & Around the Corner

Free Space
path loss

Environment
path loss

Random
variations

(obstacles)

)
,
0
(
log
10
4
log
10
2
0
10
0
10
dB
N
d
d
d
PL






















Otto, Bustamante & Berry

8

Characterize signal propagation in urban settings


Pick representative environments


Measure signal propagation in


line of sight (LOS) and


non
-
LOS (Around the Corner


ATC) settings


Pick a signal propagation model, a good simulator, and a
simple application


Free
-
space, probabilistic shadowing …


ns,
GloMoSim
, JIST/SWANS …


Evaluate application
-
level impact of environment


This work appeared in Proc. of ICDCS, 2009

Down the Block & Around the Corner

Otto, Bustamante & Berry

9

Overview of radio propagation models


Experimental characterization of radio propagation in an
urban setting (Chicago)


Measurement platform


Measured environments


Data analysis


Understanding the impact of signal propagation
parameters on application performance


Conclusion

Down the Block & Around the Corner


Otto, Bustamante & Berry

10

Set of equipped vehicles with


Soekris net4801
-
60 machines,
256 MB memory, 1GB flash
storage


Garmin GPS 18 USB

for positioning


Ubiquiti Networks

2.4 GHz 802.11b/g


7 dBi 2.4 GHz omni
-
directional

antenna

Software


Linux (2.6.19 kernel)


iperf (CBR UDP stream)


tcpdump

Garmin

GPS 18 USB

Soekris net4801
running Linux

7 dBi omni
-
directional

antenna

Down the Block & Around the Corner

Otto, Bustamante & Berry

11

Measurement in representative environments & times


Open field


Provides a baseline; no
buildings or any other obstacles

Suburban


Residential area with
trees, cars and houses set back from
the road with space between them

Urban


Large and tall
buildings, very close to
the street, few gaps
between buildings, etc

Down the Block & Around the Corner

Run experiments:




Daytime (high traffic)



At night (low traffic)

Otto, Bustamante & Berry

12

Down the Block & Around the Corner

No traffic

Path loss exponent
stabilizes at 3.10

Line
-
of
-
Sight
(LOS)

Communication

Same road

Path loss exponent

Distance (meters)

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

A呃

Otto, Bustamante & Berry

13

Down the Block & Around the Corner

No traffic

Median path loss
exponent = 3.29

Around the Corner

(ATC) Communication

Perpendicular roads

Path loss exponent

Distance from intersection (meters)

Distance (meters)

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

A呃

㌮39


㌮35

Otto, Bustamante & Berry

14

Same road










Perpendicular roads

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

A呃

㌮39


㌮35

Otto, Bustamante & Berry

15

Suburban

Open field

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

㌮34


㜮78

A呃

㌮39


㌮35

Otto, Bustamante & Berry

16

Suburban

Open Field

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

㌮34


㜮78

A呃

㌮39


㌮35

㌮37


㠮84

Otto, Bustamante & Berry

17

Same road

Perpendicular roads

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

㌮34


㜮78

A呃

㌮39


㌮35

㌮37


㠮84


At 50 meters apart,
LOS and ATC
β

= 3.2


At 80 meters apart,
LOS
β

= 3.1… but
ATC
β

> 4 !

Otto, Bustamante & Berry

18

Urban

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

㌮34


㜮78

㌮37


㤮95

A呃

㌮39


㌮35

㌮37


㠮84

㐮45


㄰⸷.

non
-
LOS communication,
higher path loss exponent
due to diffraction, reflection

50 meters apart, in LOS

> 100 meters apart, no
communication possible

Otto, Bustamante & Berry

19

Suburban

Urban

Down the Block & Around the Corner

Can be
20 meters
from
intersection before
observing PLE increase

Distance of obstructions from the road:



Suburban: wide front lawns



Urban: narrow sidewalks

Immediate increase in PLE
after leaving intersection

Otto, Bustamante & Berry

20

Obstacles increase signal variability (shadowing parameter)


e.g. from
σ

= 3.23 in an open field to 9.15 in an urban setting


Vehicular traffic degrades signal strength


Overall, path
-
loss exponent is not significantly impacted


e.g. from 3.10 in an open field to 3.17 in an urban setting


Transmit range reduced by
14%


Open field: 1070 m


Urban: 915 m


(predicted with model)

Down the Block & Around the Corner

Otto, Bustamante & Berry

21

Path loss exponent varies significantly


e.g. 3.29 in an open field to 4.05 in an urban setting


Transmit range reduced by
70%


Open field: 715 m


Urban: 208 m


(predicted with model)


Non
-
LOS communication is possible


Reflection, diffraction


Gaps between buildings


Distance of obstacles from road is a significant factor

Down the Block & Around the Corner

Otto, Bustamante & Berry

22

Challenge assumption: one set of parameters is sufficient


Experiments contradict this assumption


For complex environments (suburban, urban)


LOS vs. non
-
LOS (ATC) is a key factor in communication


So, we actually need at least
two
sets of parameters:


LOS and non
-
LOS (ATC)


What is the impact at the application layer?


Use simulations to evaluate application performance under


Environments


Parameter settings (e.g. LOS, ATC)

Down the Block & Around the Corner


Otto, Bustamante & Berry

23

Pick a signal propagation model, a good simulator, and a
simple application


Signal propagation model


Log
-
normal path loss with shadowing


Sample application


Epidemic
-
based data dissemination


e.g. Communicating road (traffic) conditions


Push
-
based protocol, based on
Vahdat

& Becker (2000)

1.
Beacon

2.
Exchange digest

3.
Send messages


Application performance metric: Delivery latency


e.g. Lower latency gives fresher data and better detouring ability

Down the Block & Around the Corner


Otto, Bustamante & Berry

24

For simple environments


LOS vs. ATC does not affect performance


However… for complex environments


LOS performance much higher than ATC


Combining data sets does not give average performance


We evaluate
LOS&ATC


Switch between LOS and ATC parameters: same / different street


Gives
expected

intermediate performance


Compromise between scalability and realism

Down the Block & Around the Corner

Otto, Bustamante & Berry

25

For simulation


JiST
/SWANS++


http://www.aqualab.cs.northwestern.edu/projects/swans++/


For vehicular mobility


STRAW


Using real cities’ road maps


Lights, signals, speed limits


IDM car
-
following


MOBIL lane
-
changing


http://sourceforge.net/projects/straw/



Parameters


Map: downtown Chicago (approximate Manhattan grid), 1.76 km
2


Radio settings: match experiment configuration


26
dBm

transmit power, 7
dBi

antenna gain, 2 Mbps fixed data rate


150 vehicles


2 hour duration

Down the Block & Around the Corner

Otto, Bustamante & Berry

26

Down the Block & Around the Corner

LOS

ATC

In an open field, the locations of the communicating vehicles
(in line
-
of
-
sight or not) have no performance impact

Open field

setting

with traffic

Otto, Bustamante & Berry

27

Down the Block & Around the Corner

LOS

ATC

In urban settings, around
-
the
-
corner parameters mean
smaller transmit range, hence lower performance

Urban

setting

β

/

σ

Urban

LOS

3.17

/
9.15

ATC

4.05

/
10.74

Otto, Bustamante & Berry

28

Down the Block & Around the Corner

LOS

ATC

Combined

Averaging parameters


by combining datasets


doesn’t
yield averaged performance

Urban

setting

β

/

σ

Urban

LOS

3.17

/
9.15

ATC

4.05

/
10.74

Combined

3.43

/
11.95

Intermediate PLE, but
increased shadowing

Otto, Bustamante & Berry

29

Down the Block & Around the Corner

Urban

setting

LOS

ATC

LOS&ATC

Using two parameter sets and relative vehicle position,

select LOS or ATC parameters based on node position

Otto, Bustamante & Berry

30

Simple environments (open field)


One set of parameters is sufficient


No difference in performance between LOS and ATC parameters


Complex environments (suburban, urban)


Using one set of parameters (LOS or ATC) is
not
sufficient


Combining LOS and ATC gives
worse than expected
performance


LOS&ATC

approach gives the expected intermediate performance


Possible extensions to
LOS&ATC


Tolerance for distance from the intersection


Simulating heterogeneous environments on the same map


Utilizing LOS/ATC information at the protocol or application layers




Down the Block & Around the Corner

Otto, Bustamante & Berry

31

LOS is a major factor of signal propagation characteristics
in complex environments


Accounting for LOS versus non
-
LOS has a significant impact
on application
-
level performance


LOS&ATC
is a
computationally scalable
and
more realistic
approach for modeling complex environments


Part of
C3R
, a project on urban environmental monitoring
through
vehicular networks, working
towards


Ensuring sustainable urban growth


Participatory sensing with a mobile platform


Applications including traffic advisory, air quality and noise monitoring

Down the Block & Around the Corner

Otto, Bustamante & Berry

32

Same road










Perpendicular roads

Down the Block & Around the Corner

With traffic,

Increased β (3.31) and
σ

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

A呃

㌮39


㌮35

Otto, Bustamante & Berry

33

Urban

Open field

Down the Block & Around the Corner

β

/

σ

佰敮e䙩敬e

卵S畲扡u

啲扡b

䱏L

㌮30


㌮33

㌮34


㜮78

㌮37


㤮95

A呃

㌮39


㌮35

㌮37


㠮84

Similar to suburban:
larger variations in
path loss exponent