Modelingthe UnderWaterAcoustic (UWA

03/10/2012

1

Modeling the UnderWater Acoustic

(UWA) channel for simulation studies

Michele Zorzi, [email protected]

Collaborators: Beatrice Tomasi, James Preisig and Paolo Casari

Motivation

Channel modeling is important for

Michele Zorzi - UCOMMS 2012 - Italy

� Evaluation and comparisonof communicationtechniques and networkingprotocols

� Design and optimization of cross-layer communications and networking techniques

03/10/2012

2

Motivation

Channel modeling is important for

Simulations are


� Evaluation and comparisonof communicationtechniques and networkingprotocols

� Design and optimization of cross-layer communications and networking techniques

� Cheaper than experiments

� Based on channel modelswhich represent statistics both in time and in space

� Reliable only if models are sufficiently accurate and representative

� Therefore, accuracy vs

computational complexity

Long-term scientific objectives

IDENTIFY

� Key features of channel that most significantly affect the performance of communication systems

�


03/10/2012

3


IDENTIFY


� Key features of channel that most significantly affect the performance of networking protocols

�



IDENTIFY


� Key features of channel that most significantly affect the performance of networking protocols

� Paradigms for modeling the channel that are most suitablefor simulation studies and evaluations


03/10/2012

4

Scientific and technical approaches


� Data analysis� Time-varying channel conditions: types of dynamics, stationarity,

predictability

� Relationship between environment and channel quality dynamics




predictability


� Analytical derivations and model validation through data� Hybrid sparse/diffuse channel model

� Markov models for channel and protocol evaluations

03/10/2012

5




predictability


� Analytical derivations and model validation through data� Hybrid sparse/diffuse channel model

� Markov models for channel and protocol evaluations

� Tools and resources� Ray-tracer included in Network Simulator 2 (NS2): WOSS

� DESERT: an NS-Miracle extension to DEsign, Simulate, Emulate and Realize Test-beds for Underwater network protocols

� Making data availability to the community through a website

Key features of the UWA channel

� Statistics representing both temporal and spatial variations

� First order statistics and their temporal variations �

� Second order statistics and their temporal variations �

� First order statistics and their spatial variations �

� Second order statistics and their spatial variations �

� Complete statistical characterization in both time and space �


03/10/2012

6








� Propagation Delays (sound velocity) �


Data analysis


03/10/2012

7

Experimental data sets


KAM 11, Kauai (HI) June-July 2011

SubNet09, Pianosa (IT) May-September 2009

SPACE08, Martha’s Vineyard (MA), October 2008

Time-varying channel conditions


0 0.5 1 1.5 2 2.5 32

3

4

5

6

7

8

9

10

SN

R[d

B]

time [min]

SPACE08KAM11

0 20 40 60 80 100 1205

10

15

20

25

30

SN

R[d

B]

time [min]

SubNet09

Different SNR dynamics depend on scenario and environmentalconditions dominating propagation (KAM11 and SubNet09 water depth ~100 m -> ssp, SPACE08 water depth~15 m -> surface)

03/10/2012

8

Wide-sense stationarity of the channel

quality


0 0.5 1 1.5292

293

294

295

296

297

298

299

300

301

frequency [Hz]

time

[day

s]

0

0.5

1

1.5

2

2.5

3

3.5

4

x 10−8

0 0.5 1 1.5292

293

294

295

296

297

298

299

300

301

frequency [Hz]

time

[day

s]

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

x 10−8

Time series of Power Spectral Density (PSD) of received energy at link Tx-S3

Time series of Power Spectral Density (PSD) of received energy at link Tx-S4

Adaptation and predictability


� Time-correlateddynamics enablepredictability of the channel quality

� Long propagation delaysmake the Channel State Information (CSI) outdated

� Therefore adaptivetechniques shouldinclude a learning phaseand prediction

0 1 2 3 4 5 61.5

2

2.5

3

3.5

4

4.5

lag, τ [minutes]

Thr

ough

put,

Θ(τ

) [b

/s/H

z]

2 am00 am

Throughput as a function of feedback delay in a variable-rate communication scheme.

03/10/2012

9

Channel dynamics and environmental

conditions


0 0.5 1 1.5292

293

294

295

296

297

298

299

300

301

frequency [Hz]

time

[day

s]

0

0.5

1

1.5

2

2.5

3

3.5

4

x 10−8

Time series of Power Spectral Density (PSD) of received energy

Channel dynamics and environmental

conditions


0 0.5 1 1.5292

293

294

295

296

297

298

299

300

301

frequency [Hz]

time

[day

s]

0

0.5

1

1.5

2

2.5

3

3.5

4

x 10−8

Periodic behaviors were observed in correspondence of low wind driven surface energy during SPACE08

Time series of Power Spectral Density (PSD) of received energy

03/10/2012

10

Open issues

� Relate typical environmental conditions and channelquality dynamics

� Model the channel dynamics

� Evaluate predictive techniques with limited observability

� Identify channel dynamics that most affect coherent and non-coherent communications techniques

� Evaluate communications and networking techniques forthe modeled channel dynamics


Analytical derivations and model

validation


03/10/2012

11

Hybrid sparse-diffuse channel model


0 5 10 150

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

channel delay [ms]

|h|

Sparse componentDiffuse componentsqrt of PDP of diffuse comp.

Channel model for UWA communications via an Ultra-Wideband channel. Sparse channels have been proposed so far, however, a hybrid channel (both sparse and diffuse) is more accurate.

Hybrid sparse-diffuse channel model


−5 0 5 10 15 20 25 30

100

101

102

SNR [dB]

MS

E

G−Thres dataG−Thres modelLSSparse

0 5 10 150

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

channel delay [ms]

|h|

Sparse componentDiffuse componentsqrt of PDP of diffuse comp.

Channel model for UWA communications via an Ultra-Wideband channel. Sparse channels have been proposed so far, however, a hybrid channel (both sparse and diffuse) is more accurate.

03/10/2012

12

Markov and time-correlated models for UWA

channels


� Validated Markov and hidden Markov model to accurately represent the packet error process at the physical layer

0 10 20 30 40 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

m

Pr[P

kt. n

+m e

rrone

ous

| Pkt

. n e

rrone

ous]

, ξn,

m

measurediidMM1HMMMM2

Markov and time-correlated models for UWA

channels


� Validated Markov and hiddenMarkov model to accuratelyrepresent the packet errorprocess at the physical layer

� Validated Markov model tobe sufficiently accurate forrepresenting the performance of Hybrid Automatic Repeat reQuesttechniques

−10 −5 0 5 10 15 20 250

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Average SNR [dB]

Thro

ughp

ut [P

kt/s

lot]

T1−H1 Analysis, HARQ−IIT1−H1 Analysis, HARQ−IT1−H1 Simulation, HARQ−IIT1−H1 Simulation, HARQ−IT3−H1 Analysis, HARQ−IIT3−H1 Analysis, HARQ−IT3−H1 Simulation, HARQ−IIT3−H1 Simulation, HARQ−I

0 10 20 30 40 500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

m

Pr[P

kt. n

+m e

rrone

ous

| Pkt

. n e

rrone

ous]

, ξn,

m

measurediidMM1HMMMM2

03/10/2012

13

Open issues

� Parametric models can be used to fit the different channel dynamics and conditions measured throughout different experiments

� Model validation through experimental data may not be conclusive, but current evidence is favorable

� The gap on the relationship between environmental and channel conditions limits the generalization of such results


Tools and resources

From simulation toward experiments


03/10/2012

14

World Ocean Simulator System (WOSS)


� Ray-tracer (Bellhop) has been included in NS2, in order to get more accurate evaluations of the performance of networking protocols.

http://telecom.dei.unipd.it/ns/woss/

DESERT: DEsign, Simulate, Emulate and

Realize Test-beds


http://nautilus.dei.unipd.it/desert-underwater

03/10/2012

15

Controlled data access via website


� Experimental time series of both channel qualities and communication performance belonging to the 3 data sets are going to be public and downloadable through a login and password procedure

https://telecom.dei.unipd.it/uns/php/data.php

Open issues

� Time-varying channel conditions are not included in WOSS and are cumbersome to be recreated through a ray-tracer

� Lack of modeling able to represent how statistics of environmental conditions reflect into statistics of channel conditions (both in time and space)


03/10/2012

16

Conclusions











03/10/2012

17



















03/10/2012

18










Contributions


� Identified the most important characteristics of the channel to be included in simulators: temporal and spatial dynamics

� Made progress towards understanding the limitations in generalizing the results derived from data analysis

� Highlighted the important missing measurements and characterizations for effective simulation studies

� Provided important and useful instruments (WOSS, DESERT, and a website) to the community

03/10/2012

19

� Time- and space-varying behaviors are observed in UW channels

� Dealing with them is important and challenging

� The UW channel is very difficult to characterize, but recent results show some patterns (at least from a networking perspective)

� Understanding these effects and their relationship with environmental parameters is an emerging topic

� How to use this information will involve a mixture of adaptation, prediction, and learning

� Availability of open-source tools a key enabler for manygroups

Conclusions


� The US Office of Naval Research and ONR-Global

Thanks to…


03/10/2012

20

� The US Office of Naval Research and ONR-Global

� And to…

� The Italian Institute of Technology

� The European Commission

� The NATO Undersea Research Centre (now CMRE)

� The US National Science Foundation

Thanks to…


Further reading on our web site

� http://telecom.dei.unipd.it/underwater

� [email protected]


Documents

Modelingthe UnderWaterAcoustic (UWA