canary - royal observatory, edinburgh€¦ · canary phases • 2010 – phase a – 3 ngs in 2.5...
TRANSCRIPT
CANARY
Tomography workshop
Edinburgh, 25‐26 march 2014
CANARY
• CANARY is a technical demonstrator for MOAO
• Installed on the William Herschel Telescope (La Palma, Canaries)
• Born in 2007 as a « fast track project » for the need of the phase‐A of EAGLE, a MOS proposal on the EELT
• Works on quadruplets (= 3+1) of stars
– 3 off‐axis stars for tomography
– 1 central star for making an image and diagnos\c purposes
– 4 lasers guide stars
• First success in Sept. 2010 : MOAO demonstrated on‐sky
• Recent sucessful a`empts for astrophysics (merging galaxy cores)
CANARY phases
• 2010 – PHASE A
– 3 NGS in 2.5 arcmin diameter
– tomography + open loop
• 2012 – PHASE B1
– PHASE A config +
– 1 Rayleigh LGS on‐axis in open‐loop
• 2013 – PHASE B2
– PHASE A +
– 4 LGS Rayleigh on a square, 23’’ off‐axis,
at 21 km al\tude
• 2014 – PHASE C1 : LTAO
• 2015 – PHASE C2 : 2‐stage MOAO
Jupiter #53
Acq. cam
d=49’
’
8.3 < magnitudes R < 11.2
25’’ < Dist. from center < 65’’
2.5’ field of view
Observed asterisms
CANARY TOMOGRAPHY
CANARY MOAO control algorithm
• Learn and Apply : op\mal sta\c approach (MMSE)
– get data from turbulence
– learn, from the wavefront sensors data, what are the best parameters for the
tomographic reconstructor
– introduce turbulence knowledge + a priori (kolmo, devia\ons)
– introduce calibrated system command matrix (truth DM)
– get the final sta\c reconstructor
– Vidal et al., « A tomography approach for MOAO », JOSA A, 27, 253
• Temporal op\miza\on : op\mal temporal filtering
– op\mize the gain of an integra\ng filter versus
• turbulence speed
• noise propagated amer tomographic reconstruc\on
MMSE tomographic reconstructor
• MMSE = minimum mean‐square error
– between R . measur
– and phase
– on average
• Minimizes 〈 |phase – R . measur|2 〉
• Expression is R = 〈 phase . measurt 〉 . 〈 measur . measurt 〉‐1
contains all informa\on
about what measurements
are made of
‐ noise
‐ geometry
‐ turbulence profile
‐ LGS, NGS ...
contains all informa\on
about how measurements
are related to phase
Today’s method (Learn & Apply)
• Determina\on of covariance model
– Based on the WFS data acquired by the instrument itself
mixed WFS data
L G S +
N G S +
T T
Experimental
covar. matrix
Theore\cal
covar. matrix
Model
fipng
proc.
MMSE
tomo reconst
0 10 20 30 40 50 60
0.0
0.2
0.4
0.6
0.8
1.0
1.2
10+4
Cn2(h) profile, r0
outer scale L0
others (LQG,..)
measured WFS centroid gain,
telescope tracking
INPUTS OUTPUTS
Instrument
calibra\ons :
pupil shims
pupil magnif.
star separatn
instrum. calibra\ons may
either be inputs or outputs
of the Learn process
CANARY ON‐SKY RESULTS
green: GLAO4L3N
red: MOAO4L3N
black: SCAO
script292
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
Average profileduring the scriptProfile identified on−sky
Script script292
−100 −50 0 50 100
0
5
10
15
20
25
30
Relative Strength (%)
Altitude (
km
)
Relative Strength (%)
Altitude (
km
)
MOAO vs
GLAO
• Using full info 3N + 4L
• ast. A53, 17/9/13
2h30, script 292
r0=16.1cm, v=3.3 m/s
r0=26.3cm, v=7.7 m/s
L0= 6 m
Total
BWFit
Tomo
OL
Noise
Alias Static
NCPA
Black:SCAO
Red:MOAO4L3N
Green:GLAO4L3N
0 2 4 6 8 10 0
100
200
300
400
500
Wav
efr
on
t err
or
(nm
rm
s)
140 nm
108 nm
191 nm
LGS≈190 nm
SCAO
MOAO 4L3N
GLAO 4L3N
Average profileduring the scriptProfile identified on−sky
Script script274
−100 −50 0 50 100
0
5
10
15
20
25
30
Relative Strength (%)
Altitude (
km
)
Relative Strength (%)
Altitude (
km
)
green: GLAO4L3N
red: MOAO4L3N
black: SCAO
script274
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
MOAO vs
GLAO
• Using full info 3N + 4L
• ast. A47, 17/9/13
21h56, script 274
r0=15.3cm, v=2.9 m/s
r0=25.4cm, v=5.6 m/s
L0= 8.5 m
Total
BWFit
Tomo
OL
Noise
Alias
Static
NCPA
Black:SCAO
Red:MOAO4L3N
Green:GLAO4L3N
0 2 4 6 8 10 0
100
200
300
400
Wav
efr
on
t err
or
(nm
rm
s)
90 nm
56 nm
100 nm
LGS≈220 nm
SCAO
MOAO 4L3N
GLAO 4L3N
red: MOAO4L3N
magenta: MOAO4L3T
black: SCAO
script291
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
Average profileduring the scriptProfile identified on−sky
Script script291
−100 −50 0 50 100
0
5
10
15
20
25
30
Relative Strength (%)
Altitude (
km
)
Relative Strength (%)
Altitude (
km
)
4L3T vs
4L3N
• Using full info 3N + 4L
• ast. A53, 17/9/13 2h15,
script 291
r0=16.7cm, v=3.3 m/s
r0=22.4cm, v=7.8 m/s
L0= 7 m
Total
BWFit
Tomo
OL
Noise
AliasStatic
NCPA
Black:SCAO
Red:MOAO4L3T
Blue:MOAO4L3N
0 2 4 6 8 10 0
100
200
300
400
Wav
efr
on
t err
or
(nm
rm
s)
100 nm
64 nm
140 nm
LGS≈200 nm
SCAO
MOAO 4L+ 3N
MOAO 4L + NGS=TT 104 nm
Average profileduring the scriptProfile identified on−sky
Script script293
−100 −50 0 50 100
0
5
10
15
20
25
30
Relative Strength (%)
Altitude (
km
)
Relative Strength (%)
Altitude (
km
)
blue: MOAO3N
red: MOAO4L3N
black: SCAO
script293
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
3N vs
4L3N
• Using full info 3N + 4L
• ast. A53, 17/9/13 2h39,
script 293
Total
BW Fit
Tomo
OL
Noise
Alias
Static
NCPA
Black:SCAO
Red:MOAO4L3N
Blue:MOAO3N
0 2 4 6 8 10 0
100
200
300
400
Wav
efr
on
t err
or
(nm
rm
s)
92 nm
59 nm
132 nm
LGS≈180 nm
SCAO
MOAO 4L+ 3N
MOAO 3N 93 nm
r0=18.8cm, v=2.6 m/s
r0=30.2 cm, v=6.1 m/s
L0= 6 m
Average profileduring the scriptProfile identified on−sky
Script script275
−100 −50 0 50 100
0
5
10
15
20
25
30
Relative Strength (%)
Altitude (
km
)
Relative Strength (%)
Altitude (
km
)
blue: MOAO3N
red: MOAO4L3N
black: SCAO
script275
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
3N vs
4L3N
• Using full info 3N + 4L
• ast. A47, 17/9/13 22h03,
script 275
r0=16.3cm, v=2.6 m/s
r0=27.2cm, v=5.5 m/s
L0= 11 m
Total
BWFit
Tomo
OL
Noise
Alias
Static
NCPA
Black:SCAO
Red:MOAO4L3N
Blue:MOAO3N
0 2 4 6 8 10 0
100
200
300
Wav
efr
on
t err
or
(nm
rm
s)
97 nm 55 nm
125 nm
LGS≈230 nm
SCAO
MOAO 4L+ 3N
MOAO 3N
86 nm
The « Pluto mode »
• Originally intended to observe Pluto (that has, actually, never been
observed)
• 4 LGS in open loop + 1 TT star (=pluto) in closed loop on truth sensor
• Allows to test tomography purely on LGS
• Comparison between MOAO and GLAO
only \p‐\lt
is sensed on this
star
blue: MOAO3N
red: MOAO4L3N
black: SCAO
script275
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
green: GLAO4L3N
red: MOAO4L3N
black: SCAO
script274
1 2 3 4 50.0
0.1
0.2
0.3
0.4
0.5
# Acquisition
Mea
sure
d S
treh
l in
H b
and
Tomography with 4L
• With, and without lasers
• ast. A47, 17/9/13 22h03,
• script 274, 275 + pluto mode
• Lasers are actually doing tomography
SR IR cam (H)
script 275
3N vs 4L3N script 274
GLAO vs MOAO
SCAO
MOAO 4L
GLAO 4L
MOAO 4L3N
MOAO 3N
SCAO SCAO
MOAO 4L3N
MOAO 4L3N
Overview of results on A47, Sept 2013
Ground Wind speed < 5 m/s
0.6 0.8 1.0 1.2 1.40.0
0.1
0.2
0.3
0.4
0.5
Seeing (arcsec)
Mea
sure
d S
treh
lM
easu
red S
R (
H b
and)
seeing (‘’)
• MOAO performs well
• most of the \me
between GLAO and
SCAO
• Bad seeings : ground
layer dominates
– GLAO ≈ MOAO
• Performance fluctuates
quite a lot
CANARY : astrophysical result
NGC 6240
Galaxies in interac\on
1 guide star
38’’ off‐axis
mR ≈ 13
HST image
NGC6240
Détails
Résolu\on ≈ 2x0.090’’
Pb = sta\c aberra\ons
Total observa\on \me ≈ 2.5 h
TOMOGRAPHIC ERROR
Ver\cal error distribu\on
• Common wrong ideas :
– MMSE tomographic reconstructor designed for 2 layers compensa\on will perfectly compensate the 2 layers
– MMSE tomographic reconstructor will perfectly compensate the ground layer
anyway (because it’s easy ?)
– If all WFS see the same pa`ern ground layer only perfect compensa\on
• Error superimposi\on principle
Reconst meas1 error1
Reconst meas2 error2
Reconst
a.meas1 +
b.meas2
a.error1 +
b. error2
Reconst
a.meas1(t) +
b.meas2(t)
a2 σ21 +
b2 σ22
for independent processes 1 and 2
Ver\cal error distribu\on
• Summing profiles is allowed (encouraged ?) :
Reconst σ21
0 20 40 60
0.0
0.5
1.0
1.5
10+4
0 100 200 300 400 500
0.0
0.5
1.0
1.5
10+4
0 10 20 30 40 50 60
0.0
0.5
1.0
1.5
10+4
Reconst σ22
Reconst σ21 + σ22
profile 1
profile 2
both
• Summing layers of unitary strengh strengh (Cn2=1, r0=D, seeing=1’’...)
0 5 10 15 20
0.0
0.5
1.0
1.5
10+4
0 10 20 30 40 50 60
0.0
0.5
1.0
1.5
10+4
ε2(h) Reconst
Ver\cal error distribu\on (VED)
Reconst Cn
2
h
! (h) !2(h)
Valid for any kind of
tomo reconstructor,
not only MMSE !
ver\cal error distribu\on:
ε2(h) is associated to a given reconstructor
r0!5/3
h
" (h) !2(h)note : can be more convenient
Building the VED
• Simulate covariance matrices of single layers, at each al\tude
– every ≈ 500 m ?
• with unitary Cn2(h)
• derive the error on each
〈err.errt〉 = 〈phase. phaset〉 – 〈phase. meast〉.Rt
– R.〈phase. meast〉t + R.〈meas.meast〉.Rt
• ε²(h) allows to
– compute the tomographic error
– an\cipate the impact of unexpected layers
– assess the impact of al\tude change
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
−60 −40 −20 0 20 40 60
−60
−40
−20
0
20
40
60
0 20 40 60
0.0
0.5
1.0
1.5
10+4
GLAO
MOAO
VED example
• Same example, cont’d :
– SCAO case : R=1 Iden\ty matrix (grows like H5/3)
– error on the 3 layers increase w
al\tude, although declared w
same strengh
– GLAO beats MOAO for
0<H<2500m
– error at ground layer is not 0 for
MOAO (beware of sta\c
aberra\ons..)
– « notch » holes : width is in
(Øsubap/α)=2500m here
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
−60 −40 −20 0 20 40 60
−60
−40
−20
0
20
40
60
0 20 40 60
0.0
0.5
1.0
1.5
10+4
GLAO
MOAO
SCAO
VED example
• Varying the star separa\on :
– Zoom on an asterism
– « notch » holes : width is in
(Øsubap/α)
– (Øsubap/33’’)=3500m here
– « notch » holes are present un\l
pupils do not overlap enough (or
any more ..)
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
altitude (m)
ε2(h
)
altitude (m)
ε2(h
)
−60 −40 −20 0 20 40 60
−60
−40
−20
0
20
40
60
0 20 40 60
0.0
0.5
1.0
1.5
10+4
33’’
66’’
140’’
VED example
• Splipng 1 single layer in 2
– Separated by 1500 m
– makes ≈ no difference
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
altitude (m)
ε2(h
)
−60 −40 −20 0 20 40 60
−60
−40
−20
0
20
40
60
0 20 40 60
0.0
0.5
1.0
1.5
10+4
VED on an EELT ?
• Telescope diameter kept
constant
• Increasing the number of
subapertures creates sharper
« notch holes » in the VED
sensi\vity/resolu\on in al\tude
is higher
• An increased resolu\on of the
profile is required
7x7
10x10
15x15
30x30
Lengh of red dash
propro\onnal to Øssp / α
Impact of mul\ple thin layers
• Example on a CANARY profile
• No\ce how introducing small layers allow to reduce the error
• Can extra « fake » layers be inserted « in case they pop‐up » ?
GLAO
MOAO
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
Impact of mul\ple thin layers
• Example on a CANARY profile
• No\ce how introducing small layers allow to reduce the error
• Can extra « fake » layers be inserted « in case they pop‐up » ?
– yes, but ...
– you then pay most of the price on the ground layer
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
0 200 400 600 800 1000 1200
0.0
0.5
1.0
1.5
2.0
2.5
10−4
altitude (m)
ε2(h
)
Impact of ground layer strengh
• Same example on the same CANARY profile
• 3 MMSE reconstructors
– different Ground Layer strengh
• Safe tomography : protect yourself, consider doubling the layer.
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
0 500 1000 1500 2000
0.
1.
2.
3.
4.
5.
6.
10−4
altitude (m)
ε2(h
)
Scidar data analysis
• Scidar
– Operated by James Osborn on JKT 1 m telescope
– many data, all night long
• Match pre`y well with CANARY data
• Ground layer / low layers need
adjustment
• Many « weak » layers appear/disappear
all the \me
• However they’re not crucial to op\mize
the reconstructor
−2 0 2 4 60.0
0.5
1.0
1.5
2.0
10+4
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
2.0
10−13
22h21
VED of a raw reconstructor
• Raw reconstructor is :
Rraw = 〈 measurTS. measurt 〉 . 〈 measur . measurt 〉‐1
• Built with NO knowledge on the profile at all
• Offset wrt « learned MMSE » – temporal convergence ? (hyper‐adapted to the learning sequence)
– lack of robustness
• Naturally immunized against 10km layers
0.0 0.5 1.0 1.5 2.0
10+4
0.0
0.5
1.0
1.5
10−3
altitude (m)
ε2(h
)
0
50
0
50
computed on 10000 frames (≈1 mn)
Conclusion
• Tomography demonstrated by CANARY
• More important : simula\ons validated
– provided an average ≈100 nm model error is added
• Science demonstra\on also done
– although CANARY is not suited for that
• VED : new tools can help understand tomography for EELT ...