Littoral Underwater Acoustic Tracking Using
Collaborating Autonomous Ocean Craft:
Improving Communications through Dynamic Encoding
and Adaptive Vehicle Motion
T. Schneider
Advisor: H. Schmidt
0
10
20
30
Depth (m)
40
50
60
70
80
90
100
0
200
400
600
Range (m)
800
1000
Problem | Collaborative Exploration
Multiple autonomous underwater vehicles (AUVs) tasked
on a mission:
•target tracking (submarines, marine mammals)
•geological survey
•mine countermeasures
Why multiple craft?
•specialization
•redundancy
•spatial
distribution
Problem | Acoustic Target Tracking
How do we get find source information:
•heading: hydrophone array gives (left/right ambiguous)
heading through traditional beamforming. can break
ambiguity by vehicle motion.
•range: use waveguide invariant, or triangulate using
multiple craft.
•depth: modal energy distribution?
depth
range2
range1
θ2
θ1
~50-200 m
Problem | Intervehicle Communication
Multiple AUVs must share data:
•broad operational goals
•status (pose and health)
•environmental data (sound speed profile)
•tracking data (probabilistic?)
Acoustic communication (acomms) is the only practical
long range solution, as electromagnetic waves are
rapidly attenuated in water.
...010
10100
10101
01001
01010
001...
Acomms | Network
We use an acoustic network based on the WHOI
MicroModem (rate C: ~25 kHz carrier)
•application layer: encoding: various message codecs,
queuing: pAcommsHandler
•transport layer: MOOS driver iMicroModem
•link layer: MicroModem firmware
•hardware: MicroModem (allows frequency shift / phase
shift keying (FSK/PSK) from 80 to 5400 bits/sec).
MOOS Computer
pFramer
By nature of the channel,
throughput highly limited
(low bandwidth, high error
rate)
pBTRCodec
pCTDCodec
pAcommsHandler
iMicroModem
MicroModem
MOOSDB
to other nodes and
topside display
Acoustic Communications
80 bps - 5400 bps
Acomms | Dynamic Encoding
pCTDCodec - example of highly compressed encoding
scheme. need tov alsend: time,
sal,
depth,
lat,(1)lon. can
pr eci si on
pr ecitemp,
si on
=
round((v
al
−
v
al
)
∗
10
)
v alkey ,encoded = round((v
al
−
v
al
)
∗
10
)
(1)
key ,encoded
key
mi n
key
mi n
derive sound speed at receiver.
pr eci si on
pr eci si on
=
round((v
al
−
v
al
+
delta)
∗
10
)
v
al
=
round((v
al
−
v
al
+
delta)
∗
10
)
(2)
v
al
del
ta,encoded
del
ta
key
d
el
ta,encoded
del
ta
key
Ideas to save space:
(2)
•global key: preshared data comprising operational
pr eci si on
pr eci si on
sizekey
=−
ceil(log
v al))mi n )to
∗ 10
))
(3)
sizekey pi ece =(i.e.
ceil(log
almax
),∗ al
10max
(3)
pi ece
parameters
some
[v almi2n((v
]−not
be
exceeded).
2 ((v
e.g. global key for salinity can be [25, 40] (or tighter for
eci si on
eci si on
= ceil(log
∗ delta)
))
(4)
sizedel2ta((2
∗ delta)
∗ 102pr((2
)) ∗ 10pr(4)
sizeregion).
pi ece
del ta pi ece = ceil(log
specific
•key frame: within a packet, key all frames after the first
to the values in the first (assumes data samples are
related).
•dissolve byte boundaries.
•discard unused precision (decimal places). e.g. 12.200
becomes 12.2
Acomms | Dynamic Encoding
pCTDCodec - encoding scheme
Data (Sal, Temp, Depth)
CTD Packet
(32, 64, 256 bytes)
CCL Type (0x20)
Decode
MOOS Routing (1 byte)
Global Key
_
Key Frame (variable bits)
Packet Key
Delta Frame (variable bits)
Packet Key
_
Delta Frame
+
Delta Frame
Global Key
si on Frame
Encode
v alkey ,encod ed = round((v alkey
− v almi n ) ∗ 10pr eciDelta
)
(1)
pr eci si on
v alkey ,encoded = round((v
al
−
v
al
)
∗
10
)
(1)
...
key
mi
n
First sample of packet
Remaining samples
eci si on
Data (Sal, Temp, Depth)
v alkey ,encoded = round((v alkey − v almi n ) pr
∗ eci
10sipron
)
(1)
) (2)
v ald el ta,encod ed = round((v aldel ta − v alkey + delta) ∗ 10
pr eci si on
si on
v alkey ,encoded
= round((v
alkey
) ∗ 10 ∗ 10pr eci
) (Similar
aldel ta
− v−alvkeyalmi
+ndelta)
) (1)
(2) to Excess N scheme)
v aldel ta,encod
ed = round((v
eci si on ∗ 10pr eci si on )
= round((v al −tav−
v al
+prdelta)
v al
d elpita,encoded
size
almi
∗ 10
))
(3)
key
ece = ceil(log2 ((v almaxdel
n )key
pr eci
onsi on
pr si
eci
sizekey
=
ceil(log
((v
al
−
v
al
)
∗
10
)) ) (2)
(3)
pi
ece
max
mi
n
2
=
round((v
al
−
v
al
+
delta)
∗
10
v aldel ta,encod
ed
del ta
key
(2)
pr eci si on
si on
= ceil(log
almax∗−10vpraleci
))
=ececeil(log
delta)
(4)
sized elsize
mi n ) ∗))10
ta pikey
ece pi
2 ((v
2 ((2 ∗
pr eci si on
=
ceil(log
((2
∗
delta)
∗
10
(4)
size
d
el
ta
pi
ece
2
sizekey pi ece = ceil(log2((v almax − v almi n ) ∗ 10pr eci si on))))
(3)
(3)
pr eci si on
Acomms | Dynamic Encoding
pCTDCodec: performance at GLINT08 sea trials (Pianosa,
Italy):
Used in three vehicles (Bluefin21, OEX, Folaga).
Given operational parameters:
•32 byte (FSK) message: 3 CTD samples (85.3 bits / sample)
•64 byte (PSK) message: 7 CTD samples (73.1 bits / sample)
•256 byte (PSK) message: 31 CTD samples (66.1 bits / sample)
Compare to WHOI CCL format: 128 bits/sample
Qualitative observation at GLINT08: dropped messages
seemed highly linked to vehicle depth. can we explain /
improve?
Ray Tracing (Intuition)
Plane wave in homogeneous fluid:
c
z
Plane wave in linearly varying sound speed profile:
c
z
Ray tracing follows path of rays launched at a fan from
a point source. High number of rays passing through the
receiver point corresponds to low transmission loss and
vice versa.
Ray Tracing (Mathematics)
Ray equations (derived from the Helmholtz equation):
dr1 dc
dξ
1 dc
dr
dξ
=
cξ(s),
=
−
= cξ(s),
=− 2
dsc dr
ds
c 2 dr
ds
ds
dr
dξ
1 dcdr
dξ
1 dc
=dz
cξ(s),
=dζ
− 2 dz1 =dccξ(s), dζ = − 12 dc
ds
ds
c=dr
ds = cζ(s), ds = − c dr
= cζ(s),
−
ds
c 2 dr
dsc 2 dr
ds
ds
dr
1 dc
dz
dζ
1 dc dr = cξ(s), dξ = − 1 dc
dz
dζ
=
ds
2
=
cζ(s),
=
−
=
cζ(s),
=
−
ds
ds
c dr
z(s))
- trajectory
of ray
along
arclength
s ray along arclength
ds (r (s),
ds- trajectory
c 2 drdsof
ds
cs 2 dr
dξ dz =1
dz
dζdr
1 dc
)) - tangentc(ξ(s),
vector ζ(s))
to ray- tangent vector to ray
=−
= cζ(s),
==
−cξ(s),
jectory of ray
along
arclength
s
(r (s),
z(s))
- trajectory
of ray along arclength s
ds ds c
ds
dsds c 2 dr
angentwith
vector
to rayζ(s))
c(ξ(s),
- tangent
vector to ray(r
these
initial
conditions
for
a fan
of rays
each
with
(r (s),
z(s))
dζ of ra
1
dz
(s), z(s))
- trajectory
of ray along
arclength
s - trajectory
=
−
=
cζ(s),
cos
θ
cos
θ
ds ζ(s)) - tangent
ds vectc
c(ξ(s),
ζ(s)) - tangent vector to ray c(ξ(s),
angle θ rto
source:
r
=
r
,
ξ
=
= rthe
,
ξ
=
s
s
c(0) (r (s), z(s)) - trajectory of ray along arclength s
c(0)
cos θ
cos θ
r = rs , ξ =
cos θ to ray
- tangent vector
θ c(ξ(s), ζ(s))
sin θ r = rs , ξ = sin
r
r
=
r
,
ξ
=
c(0)
s
c(0)
z = zs , ζ =
z = zs , ζ =
c(0)
c(0)
c(0)
sin θ
sin θ
cos θ z
sin θ
r
=
r
,
ξ
=
z
=
z
,
ζ
=
z = zs , ζ =
z = zs , ζ =
s
s
c(0)
c(0)
(rs ,zs ) - sourcec(0)
position
rce position
c(0)
osition
(rs ,zs ) - source position
(rs ,zs ) - source position
(rs ,zs ) - source positionsin θ
z = zs , ζ =
c(0)
(rsComputational
,zs ) - source position
Reference: F.B. Jensen, W.A. Kuperman, M.B. Porter, H. Schmidt.
Ocean Acoustics.
dq
= cp(s),
ds
dp
1 d 2c
= 2
q(s)
2
ds
c (s) dn
Ray Tracing (BELLHOP model)
n - direction normal to ray path
BELLHOP uses finite element rays:
1
p(0) =
2
c(0)
dq
dp
1
2d c
dq = cp(s), dp = 1 d c q(s)
2
= cp(s),
= 2c 2(s) dn
q(s)
ds
ds
2
c (s) dn
ds
ds
�1/2
�
�path
� c(s) cos(θ)
1normal
n
direction
to
ray
� by the dynamic ray
�
n - direction
normal
to ray path
Amplitude of the ray
(A0(s) )=is
computed
�
4π r c(0)q(s) �
q(0) = 0,
equations:
dq
= cp(s),
ds
1
2
2
1
dq
dp
1
c
d
q(0)
=
0,
p(0)
=
dp
1 d c
0, p(0) = c(0)
= =2 cp(s),2 q(s) = 2 q(0) 2=q(s)
c(0)
ds
c (s) dn
ds ds c (s) dn
- direction
ection normaln to
ray pathnormal to ray path
�1/2
�1/2
�1 �� c(s) cos(θ)
�
�
�
�
�
c(s)
cos(θ)
1
(s)
=
A
0
�
�
�
�
A0(s) = 4π
r
c(0)q(s)
�
�
4π r c(0)q(s)
1
1
q(0) = 0, p(0) =q(0) = 0, p(0) =
c(0)
For high frequency problems,
ray c(0)
tracing is fast
compared to normal modes and sufficiently accurate.
�1/2
�1/2 ��
�
�
�
�
c(s)Tracing,
cos(θ)Theoretical
1 �Ray
c(s)
cos(θ)
References: M.B. Porter,1Y-C
Liu.
Finite
Element
and Computational Acoustics
�
�
�
(s)
=
A
(s)
=
A
0
� Computational Ocean Acoustics.
� r c(0)q(s)
F.B.0 Jensen, W.A.
Kuperman,�M.B.
H. Schmidt.
4πPorter,
4π � r c(0)q(s)
Ray Tracing Setup | GLINT08
Experimental setup:
•single AUV performing a variety of missions that we can
consider here to place it at arbitrary locations (with range
< 2km) relative to communications buoy (30 m depth).
•water depth ~100 m (assume constant).
•silt bottom: c = 1600 m/s, ρ = 1.8 g/cm^3, attenuation:
0.5 dB/λ
Can we correlate modeled transmission loss to quality of
received messages?
Ray Tracing Setup | GLINT08
dots: AUV sampled profile (Jul 31 2008 12:16-19:41 UTC)
yellow line: ship cast profile (Jul 31 2008 16:10 UTC)
Single ship cast acceptable approximation
Ray Tracing Results | GLINT08
Incoherent transmission loss for R/V Alliance CTD profile (Jul 31 2008
16:10 UTC) from a 30 m source (buoy) in a 100 m homogenous (in range)
waveguide with silt bottom.
Incoherent TL (dB)
buoy
Ray Tracing Results | GLINT08
Incoherent TL (dB)
successfully decoded:
dropped message: low error
high error
Ray Tracing Results | GLINT08
mean squared error (dB)
Glint Jul 31 08 CTD 16:10 UTC | Messages: 12:16−19:41 UTC
0
successful messages
linear fit
−5
−10
−15
−20
30
35
40
45
50
calculated TL (dB)
55
60
65
70
Proposed Applications
Back to target tracking example:
•initial research suggests it is impractical to adjust depth
to improve beamforming on target.¹
•therefore, it may be advantageous to make a depthvarying communications optimizing behavior (leaving
heading / speed to behaviors responsible for present
mission)
Also, it may be useful to choose a message rate to satisfy
a desired probability of reception (as higher rates tolerate
less error).
¹ K. Cockrell, H. Schmidt. Depth dependence of plane wave beamformed data with a horizontal array
in a waveguide. Presentation at 153rd ASA meeting, New Orleans.
Acomms Improvement Behavior
Scenario:
•AUV undergoing mission that requires certain path in xyplane but does not mandate any particular depth (e.g.
target tracking)
•AUV has a certain (known) maximum dive/surface rate.
•want to minimize depth maneuvers to
keep array straight and save power.
AUV
closest point
of approach
range
comms buoy
Acomms Improvement Behavior
current
position
“reachable” space
given current heading
possible best path
(and stay at this
depth?)
AUV
closest point
of approach
range
comms buoy
xy (top down) view
Given sound velocity profile and position of receiving vehicle (or buoy),
change depth to optimize placement in sound field.
currently being implemented, then test in simulation.
Incoherent TL (dB)
buoy
Hypothesis Testing
choose H if
Conditional Probability
choose H0 if
0
choose H0 if
fL|H (l |H0 )
>η
fL|H (l |H1 )
η = P1 /P0 to minimize error.
η = P1 /P0 to minimize error.
choose H1
ηchoose
= P1 /H
P01 to minimize error.
fL|H (l|H0 )
choose H1
fL|H (l|H1 )
l (ray tracing calculated TL)
Want to “decide” before sending a message whether it will
be received
•hypothesis 0: message will be received
•hypothesis 1: message will be dropped
choose H0 if
l (ray tracing calculated TL)
η = P1 /P0 to minimize erro
Conditional Probability
choose H1
fL|H (l|H0 )
fL|H (l |H0 )
>
f
(
l
|
H
)
fL|H (l|H1 )
1
L|H
0.35
choose H0 if
0.3
0.25
0.15
fL|H (l|H1 )
chooseη H=0 Pif1 /P0 to minimize error.
0.2
fL|H (l|H0 )
Conditional Probability
l (ray tracing calculated TL)
Hypothesis Testing | GLINT Jul 31 08
choose H0 if
choose
TL) H0 if
choose Hl 1(ray tracing calculated
fL|H (l |H0 )
>η
fL|H (l |H1 )
Conditional Probability
η = P1 /P0 to minimize error.
η
=
P
/
P
to
minimize
error.
1
0
0.1
fL|H (l|H0 )
ηchoose
= P1 /H
P01 to minimize error.
choose H1
0.05
fL|H (l|H1 )
choose H1
0
20
30
40
50
60
70
80
Given choice of hypothesis:
l (ray tracing calculated TL)
•transit to depth within
choose ‘H0’ ifregime or
Conditional Probability
•choose different rate (which will have a different set of fL|H (l |H
fL|H (l |H
fL|H (l|H0 )
conditional probability densities)
Proposed Multivehicle Experiment
•small environmental AUV takes sound velocity profile
with depth yoyo.
•small AUV transmits profile acoustically to two (or more)
tracking AUVs with arrays.
•array AUVs localize and track source, varying depth to
improve communication.
Acknowledgments
•H. Schmidt
•K. Cockrell, S. Petillo, C. Pelekanakis
•Ocean Acoustics Library authors (Acoustics Toolbox)
•WHOI Acoustic Communications group
•Crew of CRV Leonardo, NRV Alliance