following the photons… empirical, pixel-based corrections for cte jay anderson stsci october 12,...
TRANSCRIPT
Following the Photons…
Empirical, Pixel-Based Corrections for CTE
Jay AndersonSTScIOctober 12, 2011
Back-Tracking the Electrons
30s, 47 Tuc Outer field
Shuffle
Plan for the Talk
• Introduce the CTE problem• Brief history• Version 1.0: My initial solution• Version 2.0: Soon-to-be-pipelined solution• WFC3/UVIS• Version 3.0: Additional needed improvements
Steadily increasing problem for:– STIS, ACS’s WFC, … WFC3?– Is also bad for archival
WFPC2, HRC
The Problem:CTE/CTI
readout
CTE=Charge-TransferEfficiency
CTI = Charge-TransferInefficiency
readoutreadoutreadoutreadout
observedSymptoms:
–Loss of flux in source–Increase of flux in trails
Cause:–Traps within silicon pixels that delay individual electrons–Number of traps increases over time
A Brief HistoryMany Approaches
• Laboratory work: – 55Fe 1620 e events ; FPR ; EPER two trap species– Also computer modeling of distribution within pixel– Limited array of tools used, incomplete picture
• Post-hoc corrections– Common wisdom: CTE worst for faint sources on low background– Empirical photometric corrections (Riess, Mack, Ciaberge, Goudfrooij, … )
• Problem 1D: Observed flux + sky, time, location initial flux• What about astrometry? Shape?
• Pixel-based corrections/reconstructions– The holy grail– STIS: Bristow, Alexeev– WFPC2: Riess– ACS: Massey et al 2010 on COSMOS data
• Limited focus (medium/high backgrounds)• Proof of concept: generated renewed excitement at ST
EPER
paralleloverscan
image pixels (flat)
readoutFPR
up-shifted flat
charge grabbed charge
let go
My Model 1.0• Previous:
– Bristow: Sources– Riess: CRs – Massey: WPs in science frames
• Trail data from lab tests• Assume mini-channel from manufacturing expectations• Modeled specific representative trap locations
• Model 1.0: WPs in dark images– Explore lower backgrounds than GOODS sky (50 e)– Purely empirical: Just look-up tables
• Trap density: (q) traps per marginal electron• Trap release: (n;q) short + long trails
– Trap and release assumptions• Trapping deterministic• Release probabilistic• Keep track of state of each trap during transfer
– Modeling the pixel array: • continuum of fractional traps in each pixel• code economizing for speed: 2048 steps 1 to 5 steps• iterate for to get input distribution (like Massey)
= 0.01% chance
OBS’N
MODEL
INPUT
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
Animationof Model
Parametersof Model:
Trap density: (q)Trail profile: (t;q)
One Raw Dark, post SM4
Stack of 168 Post-SM4 Darks
CR Tail Measurement
Empirical TrailsFaint
Bright
No “notch”channel apparent!
consistentwith commonwisdom that CTE worse forfaint sources
Corrected WP Trail
Residuals
Faint
Bright
Adjust by handthe model parameters1) density: (q)2) profile: (n;q)
Corrected WP Deep
The tests…
1) Aesthetic test: trails gone?
2) Photometry: flux back?
3) Astrometry: flux in right place?
4) Shape: flux really in the right place?
339s, 47 Tuc Outer field
339s, 47 Tuc Outer field
30s, 47 Tuc Outer field
30s, 47 Tuc Outer field
30s, 47 Tuc Outer field
30s, 47 Tuc Outer field
The tests…
1) Aesthetic test: trails gone?
2) Photometry: flux back?
3) Astrometry: flux in right place?
4) Shape: flux really in the right place?
• Limitations of PB approach– Read-noise problem– S/N loss
• Limitations of v1.0– Time dependence (assumed linear)– Temperature dependence of trails– Even darks not dark– Need to explore lowest packets (10 e)
Not the end of the story…
OBS’N
MODEL
INPUT
+RN
Model 2.0• Will soon be released as standard pipeline product• Compare long darks + short darks
– Can see the 10 e WP events– Absolute handle on losses
Known WPs!
Same WPs?
1000s,1000 e-
100s,100 e-
Creeping CTE
TOP OF CHIP
BRIGHTER WPFAINT WP BOT OF CHIP
Model 2.0
crosssection
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
(q)
(DN)
(DN
)
• Will soon be released as standard pipeline product• Compare long darks + short darks
– Can see the 10 e WP events– Absolute handle on losses– Truly pathological losses…
15 e < 1 e
• Aging fast• Why?
– SBC, mini-channel, etc?! • Maybe, but useless….
– Solar cycle– Different observing regime
• Lower background• Narrow filters, UV, low dark current, few WPs
• True mitigation available– Charge-injection: every 10, 17, or 25 lines– Benefit, but limited…
WFC3/UVIS
WFC3/UVIS• Aging fast• Why?
– SBC, mini-channel, etc?! • Maybe, but useless….
– Solar cycle– Different observing regime
• Lower background• Narrow filters, UV, low dark current, few WPs
• True mitigation available– Charge-injection: every 10, 17, or 25 lines– Benefit, but limited… True mitigation,
but add noisemodel dependence (get better model)
Model 3.0• Realization that dark current important
– Readout ~ 90s, but many WPs…– Even bias frames have 15 e at top!– Do the correction on raw frames
• Study everything– Column by column dependence– WPs in all exposures over time– EPER parallel overscan over time– Pin-down UVIS model, using charge-injection– Explore UVIS CI mitigation
• Goddard exploring possible injection mitigation…
THE END