- Gemini Home
- Telescopes and Sites
- Science Visitors at Gemini
- Observing With Gemini
- Retired Instruments
- Interface Specs for VI
- Visiting Instrument Policy
- DSSI Speckle Camera
- TEXES (North)
- Integration Time Calculators
- Magnitudes and Fluxes
- Near-IR Resources
- Mid-IR Resources
- Observing Condition Constraints
- Performance Monitoring
- SV/Demo Science
- Future Instrumentation & Current Development
- Queue and Schedules
- Data and Results
Change page style:
- Read-out Modes
- Dark Current and Bad Pixels
- Cosmic Rays
- Linearity Correction
- Saturation and Persistence
The GSAOI detector is formed by four Rockwell HAWAII-2RG (H2RG) arrays with 2048 x 2048 18 μm pixels each. The arrays are mounted in a 2 x 2 mosaic that create a 4080 x 4080 pixel focal plane. There is a gap of ~2.5 mm between the arrays, corresponding to ~2.4" on sky. The arrays have an additional four rows and columns around the outer edges that read out as reference pixels and are not illuminated. The HAWAII-2RG arrays are sensitive to light out of 2.6 μm and use a HgCdTe array layer. Each array is readout through four amplifiers that simultaneously read out 512 x 2048 pixels in ~ 5.3 seconds. GSAOI arrays are normally read out using the Fowler sampling technique (see below for details) to reduce read noise. Each array also contains a programmable On-Detector Guide Window (ODGW). These ODGWs provide tip-tilt information for up to four natural guide stars to GeMS. See On-Detector Guide Windows page for details. The table below summarizes the basic GSAOI array properties.
|Type||Rockwell HAWAII-2RG HgCdTe
|Array sizes||2048 x 2048 pixels each (2040 x 2040 active)|
| Detector area
||4080 x 4080 pixels (~ 85" x 85")
|Physical Pixel size
||0.9 μm to 2.6 μm (data / plot)
||~ 2.4 e-/ADU|
|Dark current||~ 0.01 e-/s/pix (~12 e- in the maximum integration time of 20 minutes)|
|Saturation||~ 50,000 ADU|
|Well Depth||~ 120,000 e-|
|On-Detector Guide Windows (ODGW)||One programmable ODGW per detector|
GSAOI arrays are normally read out using a technique known as ''Fowler sampling'' (Fowler & Gatley 1990, ApJL, 353, L33) to reduce read noise. In this mode, a set of multiple non-destructive reads are made at the beginning of an exposure and at the end of the exposure. The two sets of non-destructive reads are averaged and the signal is formed from the difference between these two averages. Each array is continuously read out and reset in “idle” mode between science exposures. This prevents the arrays saturating, and maintains the readout electronics at a stable operating temperature. When a science exposure is initiated, the current idle mode read is interrupted and the science exposure is started.
The table below summarizes the three read-out modes implemented in GSAOI for Bright, Faint and Very Faint objects. Note that the readout noise can be reduced by optimizing the readout method. For example, tests performed in the laboratory and confirmed during the commissioning, show that the read noise can be reduced to ~10 e- if we use 8-8 Fowler sampling (16 NDRs). However, the penalty in the read noise reduction is the increase in the read out time by a factor of two, from 26.2sec to 47.7sec.
|Mode||Fowler Samples (FS)||Non-Destructive Reads (NDR)||Read Noise
||Read Out time
||Min. Exp. time
|Bright Object||1||2||28 e-||10.0 sec||5.3 sec|
|Faint Object||4||8||13 e-||26.2 sec||21.1 sec|
|Very Faint Object||8||16||10 e-||47.7 sec||42.2 sec|
The table below shows the gain and read noise values for each GSAOI array.
|Bright Object||Faint Object||Very Faint Object|
|Array||Gain (e-/ADU)||Read Noise (e-)||Read Noise (e-)||Read Noise (e-)|
Dark current and Bad Pixels
The dark current in the GSAOI arrays is very low (~0.01 e-/s/pix) and stable. This produce a dark charge of ~ 12 e- in the maximum integration time of 20 minutes (set by cosmic ray hits, see below). However, there are a large number of "hot" pixels. The "hot" pixels appear as a single high dark current pixels with adjacent moderately high dark current pixels ("cross" shape). An example is given below, where the "cross-shaped" hot pixels are clearly seen in the 300 sec dark current image. The "cross-shaped" hot pixels appears to be due to high leakage current in the central pixel, which affect its neighbors through capacitive coupling between adjacent pixels.
Dark current images show spatial structure within each array. The level of the structure can get up to 10 ADU above the mean of the full array, but it is of the order or smaller than the standard deviation of the pixel values around the mean. This means that in any dark subtracted image, the dark structures will be within the noise. The existing spatial structures show temporal variations in time scale of seconds to hours, however the variations are smaller than the standard deviation around the mean values. The dark current also shows temporal variation in amplitude within each array, in the same time scale, of the order of 2 to 6 ADUs. This amplitude variation is smaller than the standard deviation from the mean. All these effects are very small and dark subtraction in the form of dark frames are not required for the different types of science and calibration frames. The normal near-IR observing techniques will be enough to take care of any dark current counts/structure.
The GSAOI arrays have imperfect cosmetic quality. This is due to the existence of 'dead' pixels (debonded), light emitting-diode pixels, "killed" spots, and excessive dark current areas. Bad pixel masks are necessary to construct to correct by the imperfect cosmetic and hot pixels. The procedure to derive a bad pixel masks can be found here.
The GSAOI arrays are sensitive to the cosmic rays. Cosmic rays interact with the CdZnTe substrate above the HgCdTe array layer and appear to forward scatter photons or electrons that strike the detector in a diffuse pattern extending over a diameter of ~ 20 pixels. Generally, these events add only a few hundred electrons to the pixel signal. However, their diffuse nature makes cosmic ray detection difficult in real time.
The cosmic rays removal appears to be more effective if multiple science exposure, at the same location on the sky, are recorder and then median of these frames are used to remove these event. However, this requires a reduction of the integration time to a point where cosmic ray hits contaminate only a small fraction of the detector area.The test results show that this integration time limit is around 15 minutes. This is supported by the results obtained during the GSAOI commissioning.
The GSAOI arrays are intrinsically non-linear due to their source-follower-per-detector architecture. The linearity correction per detector has been derived during the commissioning, and is implemented in the standard GSAOI IRAF reduction package. Once corrected by non-linearity, the data are found to have less than 2% departure from linearity at 96% of the saturation limit (see Saturation and Persistence). For count levels below 80% of saturation, the linearity correction is better than 1%.
The implementation takes the form of a quadratic polynomial fit:
Linearity Correction Factor = a+b*ADU+c*ADU*ADU
The data used to derive the parameters a,b,c above was a set of well-illuminated H and Z-band flats, and the resulting values are shown in the table below. Sigma(res) corresponds to the residual dispersion of all points in the fit.
Saturation and Persistence
Saturation of the GSAOI arrays occurs at about 50,000 ADU, which also results in a strong persistence. The persistence causes saturated images to appear as faint artefacts in subsequent exposures. Read out methods using more non-destructive readouts result in lower noise that reveals residual images for several hours so care should be taken to avoid saturation. Note that while not exposing, the arrays are continually reset pixel-by-pixel at 5s per frame. This will normally prevent the arrays saturating on the exposed field between integrations.
The table below shows the GSAOI array saturation values and count levels (in percents of the saturation level) where the data diverges >5% from linear relation before the linearity correction is applied. The table also shows the level where the linearity corrected value diverges >2% from the linear relation.
Because the arrays become >5% non-linear at ~50,000, it is recommended to keep the counts below ∼40,000 ADU (~96,000 e-) per co-add. Saturation can be checked using the ITC.
|Array||Saturation (ADU)||5% (without lin. correction)||2% (with lin. correction)|