1 NIR wavefront sensing Df A Garching 200910

1 NIR wavefront sensing Df. A Garching 200910 -14 Low-Noise IR Wavefront Sensing with a Teledyne Hx. RG David Hale Gustavo Rahmer & Roger Smith Caltech

2 Why a Natural Guide Star for Laser AO? NIR wavefront sensing Df. A Garching 200910 -14 • Wavefront tilt is not seen by a laser guide star, since the laser light retraces its outward path, so… • The TMT/IRIS OIWFS will use three natural guide stars to measure tilt, rotation and scale changes. • The brightest guide star will pass through a 2× 2 Shack. Hartmann sensor to measure focus and astigmatism. – Each of three probes can be reconfigured on the fly to be either TT only, or TTFA

3 Why the NIR ? NIR wavefront sensing Df. A Garching 200910 -14 Goal: increase fraction of sky over which adequate AO performance is achieved. • Need to guide on AO corrected image to close the tilt loop. – Need NIR to get good Strehl and thus adequate tilt sensitivity. • The most common stars are brightest in the NIR. • Diffraction core of 30 m telescope is so small that background per pixel is negligible in J+H band – Although Strehl is better for K band, sky is much brighter and diffraction core is larger

4 Some Detector Options Considered NIR wavefront sensing Df. A Garching 200910 -14 • Intevac: electron bombarded CCD with In. Ga. As photocathode. – Dark current way too high and uncontrolled – >100 Hz frame rates not available (until CCD upgraded to CMOS imager) due to ROI overheads – Current format not ideal (1024× 256) • Hg. Cd. Te APD arrays: – Attractive promises, but not ready enough yet • Hx. RG – – Well understood Large format High QE. Noise on recent devices good enough after multiple sampling.

5 Format (not to scale) NIR wavefront sensing Df. A Garching 200910 -14 • H 2 RG – Size of capture region is set by seeing and probe positioning accuracy 2048 Capture region ~ 1 k × 1 k 14× 14 recapture window • Spot will be small when high order correction is turned on, moving on scale of seeing profile until low order loop closed. • Final guide window size may be as large as 14× 14 pixels to handle impulse perturbations. 4× 4 guide window 2 mas/pix 2 arcsec 2048 Current baseline is ~ 1 K× 1 K operable region within H 2 RG rather than H 1 RG, since Teledyne advises this will be no more expensive and that the H 1 RG may be discontinued.

6 Zoom to Capture NIR wavefront sensing Df. A Garching 200910 -14 • Start with seeing limited image (>1/4 detector area) – Big, fuzzy, low contrast. – Move to center by adjusting probe position or telescope pointing. • Turn on high order correction. – Tiny spot scribbles lines over the seeing profile. Not much change seen in long exposure. • Window down on seeing limited image. – Faster frame rate makes wiggly line shorter. • Close loop at low gain to drive centroid towards center of window. – Steadily reduce window size to increase frame rate and loop gain. – Servo keeps spot within shrinking window. • Zoom in to 4× 4 window – avoid bad or noisy pixels

Why Such a Big Detector? 7 (… just to replace a quad cell !) NIR wavefront sensing Df. A Garching 200910 -14 • Big telescope aperture makes tiny diffraction core: FWHM = /D = 0. 008 arcsec at 1. 2µm • Quad cell requires ~0. 004 arcsec/pixel • Prefer 0. 002 arcsec/pixel so that positioning on pixel boundary not required to maximize centroiding sensitivity. • 2 arcsec field of view needed to capture seeing profile. We could go larger to aid acquisition. Thus need >1 K 2 • Freebie: science image can be acquired around guide star, since H 2 RG allows nested windows and independent reset.

8 Maximizing Frame Rate NIR wavefront sensing Df. A Garching 200910 -14 • The spot is compact throughout the zoom, since laser guide star sensor is blind to tip tilt, but it is smeared by image motion. – Our problem is to locate it (short frame time) and – Measure tilt accurately (low noise) to feed back to servo. • Flux per pixel depends on image motion rather than exposure time, so maximize frame rate. – Pixel time has been minimized. – For window >64 pixels: 32 ch readout, skipping unwanted lines. – For window <64 pixels: single channel, window mode. • By 64× 64, frame rate = 50 Hz … tilt error already << window size. – Readout time = (5. 16*N 2 +10. 33*N + 5. 28) µs = 21. 8 ms • For fainter guide stars final rate ~100 Hz. For frames <44× 44, can use multiple sampling to reduce the read noise below the ~11 e- for CDS. • For brighter guide stars final rate ~800 Hz, with less noise averaging.

9 Noise & Frame Rate During Zoom NIR wavefront sensing Df. A Garching 200910 -14 Begin multiple sampling here Bright stars Faint stars Start zoom Big frames, CDS noise 4 x 4 1024 x 1024

Let’s review common readout timing options…. 12 Correlated Double Sampling NIR wavefront sensing Df. A Garching 200910 -14 e = 1 = number of exposures to do …. not shown here r = number of reset scans between exposures m = 1 = number of scans to coadd then store. p = 10 = number of dummy scans between coadded groups k = 2 = number of store cycles per exposure Frame time Exposure time Reset while idling Ignore p scans Initial scan Final scan At least one reset between frames • Exposure delay = p dummy reads for constant self heating • Subtract first frame from last frame • Equivalent to Fowler sampling with m = 1

13 Fowler “m” NIR wavefront sensing Df. A Garching 200910 -14 e = 1 = number of exposures to do …. not shown here r = number of reset scans between exposures m = 3 = number of scans to coadd then store. p = 6 = number of dummy scans between coadded groups k = 2 = number of store cycles per exposure Duty cycle < 1 Frame time Exposure time Coadd m Ignore p scans Coadd m • Exposure delay is in units of full scan ties but need not be multiple of m. • Subtract mean of first group from mean of last group.

14 Sample up the ramp. NIR wavefront sensing e = 1 = number of exposures to do …. not shown here r = number of reset scans between exposures m = 1 = number of scans to coadd then store. p = 0 = number of dummy scans between coadded groups k = 12 = number of stores per exposure Df. A Garching 200910 -14 • Store every scan (no real time coadd) • Use post facto least squares fit to measure slope with best S/N; • Effective exposure duty cycle due to weighting of shot noise by least squares ~ 90%; reduce this to include effect of the reset overhead. • Equivalent Multi. Accumulate with m=1.

15 Multi-Accumulate (JWST terminology) NIR wavefront sensing e r=2 m=3 p=0 k=4 Df. A Garching 200910 -14 = number of exposures to do …. not shown here = number of reset scans between exposures = number of scans to coadd then store. = number of dummy scans between coadded groups = number of stores per exposure Coadd m Reset r scans Coadd m • Coadd in real time, store every m scans, total exposure duration is multiple of m scan times. • Least squares fit of stored (coadded) scans is used to estimate noise. • Advantage of coadd over single samples with gaps is lower noise and better cosmic ray detection ( which appears as jump in ramp). • One or more reset scans between exposures.

Proposed mode for OIWFS (used in the noise tests presented here) 16 Differential Multi-Accumulate NIR wavefront sensing Df. A Garching 200910 -14 Exposure time Coadd m Difference = frame 1 = frame 2 = frame 3 Reset Coadd m Occasional Difference = frame 4 gap ! • Using previous frame as baseline for next frame (without reset) makes duty cycle ~100%, except for a gap when reset occurs. • Gaps at time of reset can be reduced in duration by usingle scan or significantly fewer scans to establish first baseline instead of averaging m scans. This will produce a noisier result instead of a missing result. Which is better? • The reset scan and initial read scan be combined so the reset time is hidden.

17 Nested Windows NIR wavefront sensing Df. A Garching 200910 -14 • TMT operates at such a fine plate scale that there is concern over loss of lock of the tip tilt servo due to imperfections in the M 3 bearing. • Nested windows: multiple sample small guide window, then CDS read surrounding window during the exposure delay for the small window. • Thus if the spot jumps out of the guide window we can get a snapshot just half a frame time later…

18 Nested Windows NIR wavefront sensing coadd subtract Df. A Garching 2009 -10 -14 subtract coadd subtract • Read a 4 x 4 window multiple times and coadd to beat down noise. • Read surrounding, larger window once, then revert to small window. • Calculate differences of (coadds of) small windows and differences of large windows separately. • Exposure times for each window size are same (though offset by half an exposure time). Noise is lower for the central 4 x 4 window since it is sampled more often. • Size of the large window depends on frame rate and fraction of time allocated to big window as opposed to beating down the noise in the 2 x 2 window. • If 50% of time is allocated to the larger window at 100 Hz, it can be 31 pixels across with 11 e- read noise, or a 14 pixel window can be read five times reducing read noise to ~5 e-.

19 Noise vs. Frame Rate measured for various frame sizes NIR wavefront sensing Df. A Garching 2009 -10 -14 Latest low noise 2. 5µm recipe Higher CDS noise Why this bump? • Mean noise for all pixels in window. • Negligible dark current at 80 K. Why these turn-ups? Desired 100 Hz operating point gives <3 e- read noise for 4× 4 window.

20 “Dark Signal” (…mentioned in Roger’s talk this morning) NIR wavefront sensing Df. A Garching 200910 -14 Slope depends on number of reads per pixel, not time 0. 0034 e- / read for 6µs pixel

23 Noise vs. Frame Rate measured for various frame sizes NIR wavefront sensing Latest low noise 2. 5µm recipe Df. A Garching 200910 -14

24 “Dark Signal” Effect on Measured Noise NIR wavefront sensing Latest low noise 2. 5µm recipe Still have these… Df. A Garching 200910 -14 • The small turn-ups are caused by the “dark signal” • Subtracting mean effect leaves us with 1/f noise But… • still have the “bump” and the unexplained large turn-up at very low frequencies Just 1/f noise

25 2× 2 Window NIR wavefront sensing Latest low noise 2. 5µm recipe CDS noise is poor predictor of final noise floor ! Df. A Garching 200910 -14 Individual pixels in the 2× 2 window … mean was contaminated by noisy pixels

26 64× 64 Window NIR wavefront sensing Df. A Garching 200910 -14 Latest low noise 2. 5µm recipe Hot pixels

27 Tip-Tilt Wavefront Error NIR wavefront sensing Df. A Garching 200910 -14 Read noise contours • Median WFE 26. 3 nm • Zero read noise device needs QE ≥ 65% to be as good as H 2 RG Demonstrated: 2. 8 e- @ 80 Hz, 4× 4 window >80% QE • Neither QE nor noise improvements offer significant WFE reduction in this regime

28 Sky Coverage Analysis NIR wavefront sensing Df. A Garching 2009 -10 -14 Requirements: • 2 mas jitter @ 50% Exceeded, ~90% sky coverage 2 mas

29 The End NIR wavefront sensing Df. A Garching 2009 -10 -14 s ’ t Le t r pa ! y