Research and Engineering Centerfor Unmanned Vehicles
Controlled Mobility of Helper Nodes in Delay-Tolerant
Wireless Networks
Summary of Dissertation Project
Daniel Henkel
Outline
The Network EnvironmentResearch Question/MethodologyResearch ContributionsSingle and Parallel Relay ModesEvaluation of Direct, Relay, Ferrying ModesPath Planning with Reinforcement LearningPrototype Ferrying ImplementationPublications
Network Environment - DTNs
We are dealing with Challenged Networks– Long distances (up to 100’s of kilometers)– High network latency (in sec/min/hours)– Varying and asymmetric data rates– No end-to-end paths may exist at any point in time,
i.e., no contemporaneous connectionsOur focus: Delay-Tolerant Networks (DTNs)– Special protocols enable communication (dtnrg.org)
Examples: (high path loss, long distances)– Ad hoc networks, sensor networks (smart dust),
underwater/ underground networks, space and ship-based networks
Node Mobility can be controlled: small airplanes
Three Link Modes in DTNs
GS1
UAV1UAV3
UAV2
Central Idea:Use controlled node mobility to establish best link using most appropriate mode.
Direct
Relay
FerryingA
B
How can A and B communicate in a DTN with mobility-controlled helper nodes?
Modes Defined
Direct: a direct wireless link between sender and receiver characterized by communication distance, the environment, and coding scheme.Relay: data packets traverse a series of wireless direct links from sender via one or multiple relays to the receiver. Relays can be (re-)positioned arbitrarily.Ferrying: data packets are sent wirelessly from sender to a mobile node which then stores, physically carries, and finally forwards them to the destination. Mobile nodes capable of buffering and carrying data are called “ferries”.
Research Question
Given controlled mobility of wireless network nodes, to what extent can mobile helper nodes improve network performance beyond one-hop direct communications, and which relaying and data ferrying strategies would maximize these benefits? How could ferrying be implemented in a delay-tolerant network?
Methodology
Analytical evaluation of throughput and packet delay bounds in direct, relay, and ferry communication. Detailed analysis of single and parallel transmitrelaying and variable data rate ferrying.Computer simulation of relaying and ferrying with long distances and extreme path-loss exponents.Ferry path planning using reinforcement learning to minimize average packet delay.Prototype design and implementation of reliable, delay-tolerant ferrying communication.
Research Contributions
Contributions:– Understanding of trade-offs between Direct, Relay, Ferry modes– Phase-plots show best communication mode for given application
scenario– Understanding of impact of variable communication rate on
ferrying mode– First application of Reinforcement Learning method for ferry path
planning (minimizing avg. packet delay)– OPNET Simulation of ferrying available to research community– Combined Application: Ferry forms communication hub; Relaying
and Direct maximize R, minimize τ– Prototype of reliable, delay-tolerant ferrying infrastructure can be
used by other researchers for data collection with mobile nodes (robots, planes, humans)
Direct Communication
)1(log2 SNIRWR +=
εαd
dPS =)(
αη /0WTkB=
)11(log2 ηεdWRD +=
Shannon capacity law
Signal strength
Thermal noise (normalized)
Data rate
[4]
“Single Transmit” Relay Model
Only one node transmits at any given time, a.k.a., the noise-limited case.
⎟⎠⎞
⎜⎝⎛
++=
111
kdR
kR DRS
⎟⎠⎞
⎜⎝⎛
+
+=
1
)1(
kdR
LkD
RSτ
S Rdk
t=0
Optimizing “Single Transmit”
Where is the trade-off?dk vs. # of transmissions
Optimal number of relays as function of distance, radio parameters and path loss:
),( εαCdkopt ⋅=
1−= −εε
αdckopt ε
εε ε
ccc
+⋅=+
1)1ln(with
Throughput vs. # of relays
[4]
“Parallel Transmit” Relay Model
Multiple nodes can transmit at any given time, a.k.a., the interference limited case
> Optimal distance between parallel transmissions?
{ } ⎭⎬⎫
⎩⎨⎧
+= ),,(
,1min1max ρ
ρρkdR
kR IRP
)1(log2NI
SI PP
PWR+
+=∑ ⎟⎟⎠
⎞⎜⎜⎝
⎛−
++
+=i
I iidkP εε
ε
ρρβ
)1(1
)1(1)1(
S Rρ
t=0 t=0 t=0
Optimizing “Parallel Tx”
Where is the trade-off?– interference vs. # parallel transports
One constellation of k and ρ maximizes throughputSolution using Matlab simulation
k
ρ
R
link reuse factor
100
101
102
103
1
2
3
4
5x 10
6
# of relays
Thr
ough
put [
bps]
100
101
102
103
1
2
3
4
5x 10
6
# of relays
Thr
ough
put [
bps]
100
101
102
103
1
2
3
4
5x 10
6
# of relays
Thr
ough
put [
bps]
100
101
102
103
1
2
3
4
5x 10
6
# of relays
Thr
ough
put [
bps]
Parallel TxSingle TxDirect Tx
d=2km d=4km d=8km d=16km
2km 4km 8km 16km
ε = 5 PN/α = 10-15 W
Analysis: Direct and Relay Modes
Depending on distance and environment there is one “best mode” which maximizes throughput.
[4]
Rate-Distance Phase Plot
Plot showing themode which satisfies a given throughout criterion with lowest average packet delay.
[4]
Optimal Number of Relays Parallel TX
kopt
kopt = ∞
Depending on distanceand path loss exponentan infinite number of relays might be neededto approach themaximum throughput.
Parallel Tx: Optimal Reuse Factor ρ
d=10km PN/α = 4.14·10-15 (based on 802.11)
k+1ρopt
5555565422
5555555423
5555544324
5554332225
5543222226
∞256128643216842ε
ρopt ≈ min{k+1, 5}
Optimal re-use factoras functionof path lossand numberof relays.
Starting ata certain kL,best reuseis always 5!!
[2]
Delay—Rate Phase Plot
Relay
Direct
100
101
102
103
104
105
100
101
102
103
104
data rate [bps]
dela
y[s
ec]
Unachievable
Ferry
Direct
Relay
Plot for specificdistance only.Mode resultschange withdistance.
Plot shows whichmode can achievea desired (R,τ)combination.
λAλB
AB
HF
DC λD
λC
The Ferry Path Planning Problem
What sequence of node visits minimizes average packet delay?
Example: sensor nodes A, B, C, D are generating packets with rates λi destined for collection station H.Ferry F can transport packets between nodes.
One cycle visits every nodeInsight: far-away nodes with little data to send
Visit them less often
Current algorithms based on Traveling Salesman SolutionA Bhub
fA fB
UAV
dA dB
New idea: cycle defined by visit frequencies pi
pA pB
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
hub separation from A as fraction of total distance A−B [%]
optim
al d
elay
as
frac
tion
of n
aive
del
ay [%
]
fA=1 fBfA=3 fBfA=10 fBfA=30 fBfA=100 fB
B
B
The Drawback of TSP
Simple example:
Corresponding inefficiency of TSP:
Idea: express delay in terms of pi, then minimize over set {pi}pi as probability distribution
Expected service time of any packet
Inter-service time: exponential distribution with mean T/pi
Weighted delay:
A
Chub
fA
fC
UAV
dA dB
pA pB
BfB
D
fD
pC
pDdD
dC
∑=i
ii pT τ
GoalMinimize average delay
∑ =i
ip 10≥ip
∑∑=i j i
ijj
Fpfp τ
τ
Stochastic Modeling
[1]
Probability of choosing node i for next visit:
∑=
jjj
iii f
fp
ττ/
/
Implementation: deterministic algorithm1. Set ci = 02. ci = ci + pi while max{ci} < 13. k = argmax {ci}4. Visit node k; ck = ck-15. Go to 2.
Pretty simplistic view of the world !Random selection ignores many parameters -> AI method
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
hub separation from A as fraction of total distance A−B [%]
optim
al d
elay
as
frac
tion
of n
aive
del
ay [%
]
fA=1 fBfA=3 fBfA=10 fBfA=30 fBfA=100 fB
Improvement over TSP!
Stochastic Solution and Algorithm
[1]
Agent– Performs Actions
Environment– Gives Rewards– Puts Agent in
situations called States
Goal:– Learn what to do in
a given state (Policy)
The Beauty:Learns model of environment and retains it.
Agent’s objective: maximize reward over time
New Approach: Reinforcement Learning
Learning optimal strategy from interacting with environment.
Reward Criterium:
∫ −+=1
0
)(),( 0
t
t
t dteNtasr βλ
Reward Criterion
State Space:
Action Space:
Reward:
Solution Method:
Round trips through subset of nodes, e.g., A, B, C, D, AB, AC,…DCBA
Temporal Difference Learning implemented in C++ RL framework
Tuples of accumulated traffic per node )},,,{( DCBA bbbbS =
[1]
Ferry Path Planning Results
TSP = Traveling Salesman solutionRL = Reinforcement Learning
RR = Round Robin (naive)STO = Stochastic Modeling
[1]
Sensor
Sensor
Sensor
Ferrying Implementation
TerminusGateway
SMS
GatewaySMS
SMS
Network 1 Network 2 Network 3
6.0.0.0 10.0.0.0 192.168.0.0
MANET InternetEthernet
• Custody transfer, modified UDP protocol make E2E trans-mission reliable in DTNs.• Implemented using ClickModular Router framework.• Single board computers,Atheros WiFi chipsets.
[6,7]
Publications I
[1] D. Henkel and T. X. Brown, “Towards autonomous data ferry route design through reinforcement learning,” in Proc. of IEEE WoWMoM/Autonomic and Opportunistic Communications (AOC’08), Newport Beach, CA, June 23–27, 2008.
[2] D. Henkel and T. X. Brown, “Delay-tolerant communication using mobile robotic helper nodes,” in Proc. of ICST WiOpt/Workshop on Wireless MultihopCommunications in Networked Robotics (WMCNR’08), Berlin, March 31–April 4, 2008.
[3] D. Henkel and T. X. Brown, “Route design for UAV-based data ferries in delay tolerant wireless networks,” in Proc. of AIAA Infotech@Aerospace 2007 Conference and Exhibit, Rohnert Park, CA, May 7–10, 2007.
[4] D. Henkel and T. X. Brown, “Optimizing the use of relays for link establishment in wireless networks,” in Proc. of IEEE Wireless Communications and Networking Conference (WCNC’07), Hong Kong, March 11–15, 2007.
[5] D. Henkel and T. X. Brown, “On controlled node mobility in delay-tolerant networks of un-manned aerial vehicles,” in Proc. of International Symposium on Advanced Radio Technologies, ISART ’06, Boulder, CO, March 2006, pp. 7–9.
[6] D. Henkel, C. Dixon, J. Elston, and T. X. Brown, “A reliable sensor data collection network using unmanned aircraft,” in Proc. Second International Workshop on Multi-hop Ad Hoc Networks: from theory to reality (REALMAN), Florence, Italy, May 26, 2006.
Publications II
[7] D. Henkel and T. X. Brown, “Sensor data collection through a delay-tolerant MANET of small unmanned aircraft,” in Proc. of 6th Scandinavian Workshop on Wireless Ad-hoc Networks (Adhoc’06), Johannesberg Estate, Stockholm, May 3-4, 2006.
[8] A. Jenkins, D. Henkel, and T. Brown, “Poster/demo: Reliable data collection in challenged networks using unmanned aircraft,” in Proc. of ACM MobiCom Workshop on Challenged Networks (CHANTS’07), Montreal, CAN, September, 9–14 2007.
[9] A. Jenkins, D. Henkel, and T. Brown, “Sensor data collection through unmanned aircraft gateways,” in Proc. of AIAA Infotech@Aerospace 2007 Conference and Exhibit, Rohnert Park, CA, May 7–10, 2007.
[10]A. Jenkins, D. Henkel, and T. Brown, “Sensor data collection through gateways in a highly mobile mesh network,” in Proc. of IEEE Wireless Communications and Networking Conference (WCNC’07), Hong Kong, March 11–15, 2007, pp. 2784–2789.
[11] C. Dixon, D. Henkel, E. Frew, and T. X. Brown, “Phase transitions for controlled mobility in wireless ad hoc networks,” in AIAA Guidance, Navigation, and Control Conference, Keystone, CO, August 2006.
[12] S. Jadhav, T. X. Brown, S. Doshi, D. Henkel, and R.-G. Thekkekunnel, “Lessons learned constructing a wireless ad hoc network test bed,” in Proc. of WiOpt/Workshop in Wireless Network Measurements WiNMee, Trentino, Italy, April 3–7, 2005.
Publications III
[13] E. W. Frew, T. X. Brown, C. Dixon, and D. Henkel, “Establishment and maintenance of a delay tolerant network through decentralized mobility control,” in Proc. IEEE International Conference On Networking, Sensing and Control, Ft Lauderdale, FL, Apr. 2006, pp. 23–25.
[14] T. X. Brown, S. Doshi, S. Jadhav, D. Henkel, and R.-G. Thekkekunnel, “A full scale wireless ad hoc network test bed,” in Proceedings of ISART’05, ser. NTIA Special Publications SP-05-418, March 2005, pp. 51–60.
[15] T. X. Brown, B. Argrow, C. Dixon, S. Doshi, R.-G. Thekkekunnel, and D. Henkel, “Ad hoc UAV-ground network (AUGNet),” in AIAA 3rd Unmanned Unlimited Technical Conference, Chicago, IL, September 2004.
Book Chapter[16] T. Brown, B. Argrow, E. Frew, C. Dixon, D. Henkel, J. Elston, and H. Gates, Emerging
Technologies in Wireless LAN. Cambridge University Press, 2008, ch. Experiments Using Small Unmanned Aircraft to Augment a Mobile Ad Hoc Network, pp. 695–717.