Both LUCI instruments are now equipped with Teledyne HAWAII-2RG (H2RG) HgCdTe detectors. These detectors are 2048 x 2048 pixels2. There is a border of 4 pixels around the edge that are used as reference pixels and are not illuminated. The detector is read out by 32 amplifiers in parallel to keep the minimum DIT relatively low, although it still takes ~2.51 seconds to read out the full detector in LIR mode reading 100 kilo-pixels per second per amplifier. Each amplifier channel is 64 pixels high and 2048 pixels long.
On sky, the detectors are oriented in standard astronomical fashion, with north up and east to the left when the position angle is zero. On the Aladdin display, the amplifier channels are horizontal, read out from left to right. Images of the spectroscopic slits are vertical on the detector, thus for spectroscopy the dispersion direction is horizontal. Wavelengths increase to the right.
The QE of both detectors is quite similar and relatively constant, about 83% from the dewar’s dichroic window cut-on to the red end of the K filter. This is a significant increase in QE (~60%) over the original first-generation Hawaii-2 detectors used for LUCI1. The dark current and read noise are low, allowing for background-limited observations in most imaging and spectroscopic configurations. There is significant persistence and nonlinearity with these detectors.
|Dark Current (e-/sec/pix)||0.03||0.006|
|Read Noise (e-)||LIR||9.6||9.2|
|Full Well Capacity (e-)||TBD||122,000|
|Linearity (% at 40k ADU)||11.6%||11.1%|
|Crosstalk at Saturation||TBD||≤0.2%|
|Persistence (% after 5min)||TBD||0.03%|
|Minimum DIT (sec)||LIR||2.503||2.503|
LUCI1 read noise and dark current re-determined Aug 2021 (using gain=2.21) Keep in mind that these are global mean values…each pixel is electrically independent and on a pixel-to-pixel level there are variations in all these values. For example, while the distribution for dark current peaks at 0.03e-/s/pixel, there is a tail to higher values. 96% of the usable pixels have dark current <0.1e-/s/pixel, and only ~0.02% have a dark current above 10e-/s/pixel (the latter would normally be flagged as “hot pixels” during data reduction and thus ignored). For imaging mode, where the data are typically taken in LIR/INTEGRATED mode with NDIT>>1, the read noise comes out a little higher…about 10.3e-. Normally you are quite thoroughly background limited in this case and the read noise is negligible, but researchers trying to push LUCI to the limit of what is possible should keep this in mind.
The LUCI instruments support three different detector readout modes appropriate for various observing applications/modes. These are: line-interlaced read (LIR), multiple endpoint read (MER), and sample-up-the-ramp (SUR).
There are different regimes in observing where one of these modes is preferred over the other and are discussed in more detail below.
LIR mode is most useful when your observations reach a background-limited state quickly. Use this mode if the shot noise from the sky and/or source is much higher than any contribution from the intrinsic read noise of the detector. This is most commonly encountered when imaging through the N3.75 camera or in any mode (imaging or spectroscopy) if your target is bright enough that you would saturate with longer exposures, such as telluric star spectroscopic calibrations.
In LIR mode, as you clock through the 64 pixels in each line of an amplifier channel, the pixels are read out for the second read of the previous exposure, reset, then read again for the first read of the next exposure. So the previous and next exposures are interlaced, and there is very little dead time between exposures. The line-interlaced readout incurs a single ~2.51s readout overhead independent of the number of NDITs.
MER mode, also known as Fowler sampling, is most useful when it might otherwise be difficult to reach background-limited conditions in a reasonable exposure time. Use this mode when the background from the sky between the OH emission lines is low (<1.65 μm), you are working at higher spectroscopic resolutions (spreading the background light over more pixels), or using narrower slits (not letting as much background light into the instrument).
In MER mode, the detector is reset, then read out non-destructively multiple times as fast as possible at the start of the exposure, then again the same number of multiple reads at the end of the exposure. For LUCI the number of multiple samples is fixed to five, a trade-off between larger overheads and increasing amplifier glow and other effects that limit the effective reduction of the read noise. The file saved to disk is the average of the five equivalent DCR exposures constructed from the data, so an MER mode readout has the read noise reduced by ~sqrt(5) compared to a single LIR mode readout. This may be seen as MER10 elsewhere in LUCI documentation as that is how it was originally called (5+5 readouts for MER10).
DITs shorter than 60 seconds are not likely to need MER mode readouts. Generally, most imaging and bright object spectroscopy (e.g. telluric standards) are background limited even using the LIR readout mode.
NOTE: This mode is still being developed and is not released for general use.
SUR mode is being implemented as part of a means to actively control the internal flexure in LUCI, which can affect any longer exposure observations. Use of this mode is only needed when active flexure compensation (AFC) is needed. This mode is primarily applied with low background spectroscopy or narrowband imaging, especially when using the N30 camera for AO observations.
In SUR mode, the detector is read out N times at equal intervals during the integration. The image saved to disk is derived from a linear fit through the measurements up the ramp for each pixel. Because N readouts are involved, the effective read noise is also reduced by a factor of sqrt(N/2) with respect to LIR readouts (If N=10, the read noise for SUR mode should be about the same as that of MER mode readouts). Normally N will be automatically set by the needs of the AFC algorithm, not by the user.
Saturation is around 55kADU in both LUCIs, though it is a function of both source brightness and readout mode. We recommend you stay below half-well (<27k ADU) for peak counts on your science targets to be safe. If data are linearized, it is possible to work nearly up to ~80% of saturation.
All near-infrared detectors are non-linear. Both LUCIs are about 11% non-linear at 40k ADU for LIR-mode readouts. Even at 10% full-well (about 5500 ADU) they are already ~1.5% nonlinear. Thus, if you work with high-dynamic range data or you care about doing higher precision photometry, it will be very important to linearize your data.
Linearization should be applied to crosstalk-corrected (see below) LUCI data. The equations are:
ADU_lin = ADU_raw + 2.767×10-6(ADU_raw)2 for LUCI1
ADU_lin = ADU_raw + 2.898×10-6(ADU_raw)2 for LUCI2
Using a global second-order constant like this should reduce the photometric uncertainty to 1.2% if one stays below 80% of the full well capacity. However, there is some structure in the linearization fits across the detector, with an outer frame and the keyhole showing slightly different corrections from the remainder of the detector. Thus for particularly demanding observations, where data spans the full dynamic range of the detector or there are particularly tight requirements on the overall photometric accuracy, linearization using a full 2D pixel map will be necessary. Formally, there is a readout overhead correction term needed (1+e) as well, but for LIR mode data this is completely negligible.
Since any reduction of the data will include subtraction of a dark or temporally adjacent image, zero order constant term from the linearization fit is not included. Strictly speaking, these equations are only valid for LIR-mode data. The linearization coefficient for the MER readout mode has not been explicitly measured. It should have the same form as in the equations above, but the LIR results suggest the constant would be different. Proceed with normal reduction algorithms using the linearized data.
The H2RG detector shows relatively high persistence. There is also some intrinsic structure to the persistence pattern on the detector. You will see this most strongly during a spectroscopic acquisition sequence and in the first few spectra taken on a given source. You can see this in the image below.
At the speed the detector is read out, there is some crosstalk visible from very bright sources (well beyond saturated in one DIT) between the amplifier channels. Because of the architecture of the detector (32 parallel amplifier channels that are 64×2048 pixels in size) the crosstalk signals show up in a vertical line, spaced every 64 pixels. There is no correction algorithm currently available. Correction of the crosstalk signal should be the first step in the data reduction process when the algorithm becomes available.
The LUCI2 detector has had one amplifier channel (the second one down from the top) that is showing a different gain than the other channels. This channel is not “dead”, although it often shows up as a black bar across the top of the image display (see image). As this channel is outside the region covered in spectroscopic mode with the N1.8 camera and near the edge of the imaging field of view, priority to investigate and potentially fix this is low. Any attempted repair will likely require sending the detector back to Teledyne.
You can mostly recover the data in this channel by multiplying all pixels by a factor of 2.77 (in iraf the region is [*,1921:1984], the factor may need to be adjusted slightly up or down to match the data in the adjacent amplifier channels).