A Full-Scale Ad Hoc Thanks to Networked UAV · Jesse Himmelstein Sushant Jadhav Gerald Jones Marc Kessler Jake Nelson Phillip Nies Bill Pisano Roshan-George Thekkekunnel Institute

1

CU-Boulder

Timothy X BrownInterdisciplinary TelecommunicationsElectrical and Computer Engineering

University of Colorado

Presented at the International Symposium on Advanced Radio Technologies

March 2, 2005

A Full-Scale Ad Hoc Networked UAV

Test Bed

3/3/2005 T.X Brown (http://ece.colorado.edu/~timxb) 2

Thanks toUniversity of Colorado� Brian Argrow� Cory Dixon� Jack Elson� Sheetalkumar Doshi� Harvey Gates� Daniel Henkel� Jesse Himmelstein� Sushant Jadhav� Gerald Jones� Marc Kessler� Jake Nelson� Phillip Nies� Bill Pisano� Roshan-George Thekkekunnel

Institute for Telecommunication Sciences

� Wayde Allen� John Ewan

Fidelity-Comtech� Joe Carey

L3-Corporation� Ken Davey

Air Force Material Command


Overview

� UAVs and the test bed need

� Test bed design

� Results

� Conclusions


Unmanned Aerial Vehicles (UAVs)

� Small (10kg) Low-Cost ($10k) UAVs

� Swarming, flocking, cooperative missions

� Military, scientific, commercial applications

The key to success is robust inter-plane networking

2


Ad hoc UAV-Ground NetworksAUGNets

Scenario 1: increase ground node connectivity.

NOC

Scenario 2: increase UAV mission range.


UAVs for Forest Fires

Communication Networking

Situational Awareness

Video

IR

University of Colorado, Boulder303-492-1630


The test bed need

� There is much research in ad hoc networks� Mostly simple models or in simulation

� Test bed work is small-scale, indoor, or limited mobility

� Some commercial and military deployments

� There is little work on open, full-scale, highly mobile ad hoc networks


Wireless Test Bed Goals

� Provide a realistic environment� Outdoors� Full scale� Real time

� Collect concise, detailed and accurate test data� Embedded collection design� Provide GPS time and position of all test elements

� Terrestrial fixed and mobile� Airborne systems

� Make the test bed scalable and transportable

3


Test Bed Design


Heterogeneous Nodes

� COTS 802.11b (WiFi) Radio Components

� Linux-based Open Source Software

� Small Low-Cost UAV

241cm

Radios in UAVs

Fixed Radios

Radios on vehicles16cm

Common HW/SW


Test Bed SiteTable Mountain National Radio Quiet Zone

Experimenting to show the effect of the UAV


360° From Center Of Range

The view looking north

The view looking south

4


Monitoring Model

Monitoring Server

RemoteMonitoring

Table Mountain Ad Hoc Radio Test BedNOCUniv. of Colorado

FixedGround Nodes

Internet

Hand-held

Nodes

NormalPacket

Data

Ground Vehicle Nodes

AerialVehicle Nodes

Fixed Node/

Gateway

Monitoring Backhaul

Web Access


Node Monitoring Collection

Operating System

Radio Receive Interface

DSR Router

Radio Transmit Interface

Receive Monitor

Transmit Monitor

PacketizationUDP/IP

Encapsulation

Monitor Data

Collection

Monitoring Module

TypeSizeRouteSequenceTimestamp

Per Packet InfoMessagesGPSSequenceTimestamp

Text Messages

Rad

io N

ode


Cumulative distribution function (cdf), or the fraction of packets delivered within specified delay (sec)

Time to Database

1.0

0.8

0.6

0.4

0.2

0.00 1.0 10 100 1,000

Delay (sec)

cdf

Reliable Backhaul


Remote Monitoring

Situational Map

ControlPanel

TimeControl

Data Graphs

Web-based Java GUI accessing MySQL Database

5


Testing Metrics

Measures of Performance

Data throughputLatency (communication delay)Jitter (delay variations)Packet loss, radioPacket loss, congestionCommunication availabilityRemote connectivityRange

Measures of Effectiveness

Network self-formingNode-failure recoveryMobility impactHardware reliabilityEase of deployment/transportabilityEase of operationData, voice, Web

page communication


Testing Categories

No UAV

Node failed

Range

No UAV

Node failed

Range

Disconnected groups

Mobile node at edge

No UAVs

ThroughputConnectivityCongestionSubjective


No UAVs



Fixed

Mobile

Range

Range

RangeThroughputCongestion

Ground-ground

UAV-ground

UAV-UAV

Typical test bed ConnectivitySubjective

Baseline Networks

Satellite Connection

Scenario 1: Improved Connectivity

Scenario 2: Increased UAV Range

26 Experiments in all


AUGNet Experimentation Work Sheets

Notes: The nodes are arranged on the test bed so as to form a chain network Pings are sent from the gateway to every other node in the network sequentially . The pings are run continuously for 20secs at 1 sec intervals. Once pings from the gateway are completed, the next sequential node, (X.81) starts pings to every other node. This process is repeated till the latency data for every source-destination node pair in the network is obtained.

End Ti me: 9:05 a.m.Star t Time: 8:56 a.m.

Date: 09/08/2004Filename: seqlatwithoutuavnew

Result : Completed with r eliable data

1.1.2 Baseline Networ k M easurementFixed gr ound – No UAVLatency

Results:Table H-2 contains the average of the mi nimum, mean and maximum delay values recorded per nominal hop for the network without the UAV. The measurement error represents the expected variation in these s tatistics if the experiment were to be repeated many times. As Figure H-2 indicates we did not see a marked improvement in the latency results with the introduction of the UAV. Also Figure H-4 shows packet losses with the introduction of the UAV. We attribute these results to the loss of radio connectivity encountered due to the banking of the UAV

5.605.051.20267.9076.9524.9025

48.0027.253.7567.8853.3527.6524

0.890.130.0024.2320.1318.4043

21.137.630.0252.3526.0712.9262

16.332.500.0336.3311.448.4581

Max.Mean.Min.

Measurements Errors(ms)

MaxDelay(ms)

MeanDelay(ms)

MinDelay(ms)

Numberof Tests

NominalHop/s

Table H-2

Experiment name and reference number

Experiment time and date

Monitoring file that contains the data

Notes on the experiment setup

A description of the results including data tables and

graphs

Any issues or problems that arose during testing

Recommendations3/3/2005 T.X Brown (http://ece.colorado.edu/~timxb) 20

92 @ FS1 82

83

85

81

84.

.

Fixed Ground Sites

6


92 @ FS1 82

83

85

81

84.

.

Fixed Ground Sites with UAV

Note: UAV can not talk to node 92 inside building


Meshed Ground Throughput

UAV 2x better at 4 hops

Throughput > 1Mbps at 1 hop

Drops with more hops

UAV has better throughput at more hops


Network Latency

UAV decreases latency. 3/3/2005 T.X Brown (http://ece.colorado.edu/~timxb) 24

Network Packet Loss

0

20

40

60

80

100

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

occ

ura

nce

[%

]

Lost packets [%]

Packet Loss

with UAVwithout UAV

Packet Loss

Packet Loss (%)

Occ

urre

nce

(%

)

Fixed Ground StableNo losses 100% of time

UAV Adds VariabilityOccasional losses up to 50%

7


-46.2

-6.0

140

1090

2601160

-7.7-12.1

1080

1240

11701410

-26.5

-42.3

125

675

170

1170

WithCongestion

WithoutCongestion

Change(%)

Throughput (Kbps)Node Setup / Flow Pattern92 - 81 - 82 - 83 - 84 - 85

-7.4-64.5

250

110

270

310

-55.0

+7.1

90

300

200

280

-12.50

-75.00350

90

400

360

Sample of Meshed Flow Pattern Congestion Throughput Testing

.

Unfair!3/3/2005 T.X Brown (http://ece.colorado.edu/~timxb) 26

Web BrowsingVoice over IP

� Web browsing worked well� Can tolerate ad hoc network delays

� VoIP worked well up to 3 hops with fixed nodes� Delay and delay variations exceed 100msec.


FS2

85

..

UAV Swarm


Overhead Meshing: Throughput

Plane-Plane Plane-Ground Ground-Ground

UAVs form ~1Mbps network among themselves

8


Long Range UAV Communication

� UAV meshing over a few km worked well� Long range did not work

� 7km could not connect reliably� Throughputs zero or 0.01Mbps

� Problem� Dynamics + weaker signal = choppy links

(have since tuned routing for this)� High antenna over residential area sees many

interferers (experimenting further)S

igna

l Str

eng

th

dWeaker long distance signal more likely to be below threshold


Conclusion:

� UAV Test bed up and collecting data

� Low Delay, Good Throughput: Can do VoIP

� Multi-UAVS can effectively network

� UAV maneuvering affects performance


Next Steps:� Coupling Plane

Dynamics with Communication

� UAV Flocking

� UAV specific routing

� Security

� NSF ERC

Project website: augnet.colorado.edu

Documents

A Full-Scale Ad Hoc Thanks to Networked UAV · Jesse Himmelstein Sushant Jadhav Gerald Jones Marc Kessler Jake Nelson Phillip Nies Bill Pisano Roshan-George Thekkekunnel Institute