MOIRCS Detector Information


All Information below are for NEW detectors (2016-)

Information for old detector can be found at the end fo the page.

Appearance of the data

Hawaii-2 RG detector has 2040 x 2040 sensitive pixels, with the additional 4 rows and columns of "reference pixels" on each side. The reference pixels track voltage drifts caused by temperature or bias voltage variations. They enable an immediate compensation during readout as well as a detailed offline analysis using them. We provide the CDS science data (MEF primary data) after applying the reference-pixels correction.

Examples of raw image (Ks band dome flat data and 5-sec dark image) can be downloaded from below. They are the result of median-combine of 10 individual frames.

Ks band Dome Flat
CHIP 1 [jpg, fits]CHIP 2 [jpg, fits]



Dark (5 sec)
CHIP 1 [jpg, fits]CHIP 2 [jpg, fits]
Note that there are some latent flux on the data.

The Quantum Effeciency provided by Teledyne data sheet is below.

WavelengthH2RG Sci#16926 (channel 1)Sci#16843 (Channel 2)
800nm82%80%
1000nm89%78%
1230nm81%80%
2000nm82%81%


Linearity

Figures below is the result of the on-instrument test for non-linearity, under the current detector control parameter setting (Apr 2016). Note that the vertical axis is in ADU in single readout data (not a Correlated Double Sampling (CDS) result). Note that the pedestal level is each 7k and 8k (ADU) for ch1 and 2, respectively.

The ~5% non-linearity level is reached at about 45k and 60k ADU for ch1 and 2 in single readout, respectively.

Non-linearity measurement (Apr 2016)

CHIP 1 (H2RG#16926)CHIP 2 (H2RG#16843)

In the science data after CDS calculation, the non-linearity level would be stronger for very strong input flux, due to a large pedestal level the data would have. It would be safer to keep the peak count level to less than 25000 ADU for high-accuracy photometry (< 2 % non-linearity level under mild input flux).

In future we have a plan to do the non-linearity correction "on the fly", during the data collection and assembly.



Latent Flux

Though the "latent" or "residual flux" in Hawaii-2 RG is often claimed as smaller than the old H2 SCA, we have noticed that there is a recognizable amout of latent flux for channel-2 detector (as seen in the dark data above). The latent from completely saturated stars will remain on one or two data after the event. The observers shoud be aware of such spurious signals near the saturated stars.

For channel-1 detector we see quite negligible amount of latent flux.



Cross Talk

If there is a heavily-saturated star etc in a part of the detector, the artificial signal pattern might apper at the same position in slow-read direction of each readout section. This is called the (inter-channel) crosstalk signal.

We tuned the detector readout parameters so that the crosstalk become much less evident. As a result, we do not see any recognizable crosstalk signal from both detectors in single exposure level. But if you have any heavily-saturated stars (with hole in the center, for example) in the same field of view, you should always check the possibility of having such spurious object by crosstalk to make sure. If the Y coordinate (in raw image format) of the suspicious object is close to that heavily saturated star, it might be the crosstalk.



Bias Drift and Additional Noise Pattern

We often notice that there are a lot of recognizable horizontal line pattern, as is shown in the example below (a dark data with 20sec & NDR=10). This is a random patern, and so is likely be the outside origin (electric noise). The detail is still under investigation. As the amplitude of the pattern is relatively low, they are mostly not recognizable for the broadband imaging data. However they might be visible on the data with NB imaging or spectroscopy.

When we have no large objects in the data, it is relatively easy to remove them by the line-averaging with object masks.

And also, there are the other patterns in vetical direction too. We have 32 sub-regions which corresponds to the readout section in vertical direction, and in the example image below we can see that the 2nd sub-region from left is clearly darker in the middle. The cause of this "inter-channel bias drift" is not likely due to the error in the bias-level correction by reference pixels. It is still under investigation.

Again, we can reduce the pattern by subtracting it section by section when there is no large objects in the data.

Overall performance about the stability of the data seems superior for channel-1 detector. Thus we recommend to put the high-priority object to channel-1 side.

* Note that the star-like signal is caused by the Alpha-particle radiation from the AR coating on the last lens, which was already eliminated by our particle block window installed in March 2016.



Multi-Sampling Data Acquisition

Mutil-sampling readout is the effective way to reduce the detector readout noise. In ope file we use NDR parameter for multi-sampling. In theory, the readnoise under the multi-sampling NDR could reduce the readout noise 1/sqrt(NDR) of the original single CDS readnoise. For example, we can reduce the readout noise to a third of original CDS readout noise using 10-times multi-sampling (i.e., 15/sqrt(10) =~ 5 e-).

However, in reality the measured readnoise is higher than expected. The figure below is the measured readnoise as a function of the NDR number. The measurements (purple open circles) always exceeds the expected value (solid orange line) at larger NDR. With the additional noise proportional with NDR to the orange curve we can explain the data (solid blue line). The readnoise seems the minimum of ~4 e- at around NDR=20.

    A Note for the Background-noise Operation

    Values in the right vertical axis on the figure above shows the sky count level (in ADU) which corresponds to the comparable sky noise level to the readnoise (sky photon noise = 1 x readnoise). The background-noise-limited performance could be achieved with the minimum of the value, though we recommend to set the goal of 5-10 times larger sky values than this, in order to achieve the 2-3 times larger sky background noise than the readnoise. For spectroscopy, the sky level between night emission lines (with the scattered light component) can be used for the estimate of NDR number, with 1 to 2 times readnoise setting for sky. For NB imaging, a larger multiple factor is recommended.

      (Example 1)
      Your NB imaging data shows ~300 ADU sky level for a 100 sec exposure. In this case, the recommended number of the NDR value are NDR = 3 (300/5 = 60 ADU level: the sky noise is ~2 times of the r.o. noise) to NDR=20 (300/10 = 30 ADU level: the sky noise is ~3 times of the r.o. noise). We recommend to use the maximum NDR value between 3 to 20, considering the acceptable overhead time. Or, you can just increase the exposure time to 300 sec (sky would be 900ADU) and use NDR=1 (i.e., the sky noise can be ~3 times of the r.o. noise without multisampling). This will give your a high observing efficiency with enough sky count level, while such long exposure might make the median-sky subtraction difficult (due to a rapid change of the sky). We recommend to use a shorter exposure time within your acceptable efficiency range.

      (Example 2)
      Your VPH-J spectroscopy data shows only ~20 ADU sky level for a 200 sec exposure. In this case, you have to increase the unit exposure time at least longer than 300 sec (corresponding to ~30 ADU sky level between lines) then use NDR=20 to 25 in order to achieve at least 1 x readnoise for the darkest part of the spectra. As the darkest spectral regions are relatively wide for the VPH data, even a longer unit exposure could be beneficial (in terms of both better S/N and the observing efficiency), while for the R500 grisms 1x readonoise is enough because most of the spectra is dominated by the sky emission. Keeping the unit exposure short could especially give you a benefit for R500 grisms,as it will reduce the affection of the quick change of the sky emission.

    Note that the Multi-sampling observation will add a "1.48 * (NDR-1)" sec to the overhead time. The minimum exposure time will also increase to "1.48*NDR" sec.

    Also note that the multi-sampling data file will be a big data size, because a NDR=10 exposure will generate 10 "pedestal" and "signal" images each in addition to a CDS result (they are all stored in a single MEF FITS file: see "Data Format" below). For example, the data size by a NDR=10 exposure will be 8MB x (10+10) + 32 = 192MB.



"Window" Mode and "Up-the-Ramp" Readout Method

As of January 2016, the "Window" mode and "Up-the-Ramp" readout methods are not yet available.

"Window" mode is similar to the Partial Read mode for old sysytem. It will offer the function of fast readout of small sub-windows in a detector area. "Up-the-Ramp" readout method is the technique similar to the Multi-sampling readout method. Instead of getting the CDS data multiple times (multisampling), up-the-ramp method continuously read the detector and evaluate the ramp rate of each pixels many times, thus effectively reduce the readnoise while keeping the overhead small. This readout method is effective for spectroscopic observation.

Instrument Division team is trying to inplement these modes now. We will announce to users whenever they are ready to users.



COADD and NEXPOSURE

The "COADD" data acquisition method, which we offered for the old detectors (before June 2015), was decomissioned by the change of the detector readout system.

We will instead use the NEXPOSURE parameter in GETOBJECT command as a substitute of COADDing. For example, acquiring the data with NEXPOSURE=3 will take 3 exposures at the same dither position, and we will get 3 FITS files for each dither position. If we add up these data during the post-processing, it will be the same as COADD=3 observation.

The advantage of using NEXPOSURE is that we can reduce the overhead by the telescope dithering by the use of NEXPOSURE. We can also expect a better PSF size by finely registering the frames taken by a NEXPOSURE set using the objects on the image. The cosmic-ray events can be minimized by applying a rejection of deviant pixels during the combining process.



Bad Pixel Map

The bad pixel masks after the detector upgrade (Oct 2016; in the IRAF .pl format): md5)

    * Note that the above data is for imaging data. The shadow region by the beam splitter in each channel is masked.

Bad pixel map files for old detectors can be found from here(2011-04-16: md5).



Data Format

We started to provide the FITS data as the Multiple Extensions FITS (MEF) format since 2016.

Under the CDS data acquisition, we first read the "reset (pedestal) level" image immediately after the detector reset. Then after some exposure time, we read the "signal level" data. The "raw" CDS data is the result of "signal level" data subtracted by "pedestal level" data.

In the current fits format, the extension [0] is the CDS result (double data format: 32MB), and is usually used for science. The extension [1] and [2] are the reset level image and signal level image, respectively (short data format: 8MB each).

If you want to apply the non-linearity correction to the raw data, you will need to use the data in extensions. The distribution of the non-linearity correction software is still TBD.

For the multi-sampling data, you will have the same number of the reset level and signal level images as the number of the multi-sampling.



FITS header keywords

The header information has moved to here.



Miscellaneous:

Upgraded MOIRCS: Subaru Web Article, Test report, etc


Old MOIRCS Detector Page

The old detector information page can be accessed here for records and for the archive data users.




Please note that all data on these pages are subject to change as the evaluation of the performance of MOIRCS progresses.

Updated 2017-10-24


Copyright © 2000-2017 Subaru Telescope, NAOJ. All rights reserved.