NIRCam Detector Readout Patterns
As charge accumulates during a NIRCam integration, the detectors are read out multiple times, non-destructively, sampling the data while conserving the charge in each pixel. This MULTIACCUM technique enables “up-the-ramp” fitting to determine the count rate from multiple data samples obtained over time. Up-the-ramp fitting facilitates cosmic ray rejection, reduces the effective readout noise (approximately by the square root of the number of samples), and increases the dynamic range of the final image (sampling bright sources before they saturate).
The voltage of each pixel is sampled and converted to a 16-bit (2 Bytes) integer in 10 µs. This data rate could originate 540 GB/day (accounting for overheads) when using all 10 detectors with four simultaneous outputs each. Such a data volume would overfill the onboard solid state recorder, which can store about 57 GB of data for science, downloaded twice daily.
To reduce the data rate and enable longer exposures, integrations must adopt predetermined readout patterns (see Figure 1). Each readout pattern produces a ramp containing up to 20 data points for up-the-ramp fitting. A data point is obtained by averaging as many as eight individual samples, or frames (Nframes), followed by other samples that are discarded (Nskip). The combination of averaged and skipped samples, Nsamples=Nframes+Nskip, is a group. A readout pattern is thus made by up to 20 groups, or Ngroup=20.
Groups containing a larger number of averaged frames yield lower readout noise but also allow more time for a potential cosmic ray impact, in which case the entire group, and possibly all those who follow, must be discarded. Choosing the best readout pattern thus involves a tradeoff. Initial estimates suggest that averaging more frames generally yields slightly higher signal to noise, for a given group size and Nsamples (Robberto 2009, 2010).
Integrations are terminated by a reset, which clears accumulated charge from the pixels. Multiple integrations can be sequentially executed without interruption to produce an exposure. The exposure time can therefore be regarded as the photon collection duration at each dither position.
Available readout patterns
NIRCam has nine readout patterns (see Table 1 and Figure 2). Their names encode their group size (Nsamples) followed by the number of averaged samples (Nframes). There are five available group sizes (Nsamples = 1, 2, 5, 10, and 20) named according to their potential applications (RAPID, BRIGHT, SHALLOW, MEDIUM,and DEEP, respectively). So for example, a RAPID group consists of a single sample whereas a DEEP8 group contains 20 samples (Nsamples = 20), eight of which are averaged (Nframes = 8) and 12 of which are skipped (Nskip = 12). With the exception of RAPID, the name of the readout patterns end with a number signifying Nframes, the number of averaged frames in each group. SHALLOW2 and SHALLOW4 both have groups of 5 samples, but Nframes=2 and Nframes=4 in the two cases.
Each readout pattern may contain up to 10 groups (Ngroups = 10) or up to Ngroups = 20 for the DEEP2 and DEEP8 patterns, in most observing modes. Also, in most cases up to 10 integrations are allowed per exposure, as long as the total exposure time remains within allowed limits. Tighter restrictions are placed on the faster RAPID pattern, which is designed to save all frames: when all 10 full detectors are being read, exposures are limited to Ngroups = 1 or 2.
Each detector readout takes 10.737 s for the full frame (2048 × 2048 pixels using four outputs) or as little as 49.4 ms for the smallest supported science subarray (64 × 64 pixels). Since pixels are read out sequentially, the integration start and end time varies slightly from one pixel to the next. The total integration time, however, is identical for all pixels.
The tables and diagrams below illustrate the nine readout patterns available for NIRCam observations. Tables 2 and 3 give total integration times achievable with multiple groups.
Table 1. Available NIRCam MULTIACCUM readout patterns
|Readout pattern||Samples per group |
Nframes + Nskip)
|Frames averaged |
in each group
For all readout patterns that involve averaging frames into groups (i.e., all patterns except RAPID or BRIGHT1), the initial frame will always be saved and is termed “frame 0”. It is saved as a separate extension in the data file, and its purpose is to increase the dynamic range of the data. If the first averaged group of an integration is saturated, then the photon count rate cannot be determined; however, frame 0 may not have reached full well and could therefore be used to estimate the count rate (see Figure 3). Similarly, if an integration has been contaminated by a cosmic ray that hits within the first group (and after the first frame), frame 0 may still be trusted even though the group has to be discarded.
Because the count rate in such cases would be determined by calculating the slope using frame 0 only, it is important that the bias of the detectors is very well characterized. Any uncertainty on the signal level at the very beginning of an integration (the “reset” or “bias” level) due to, for example, global electronic offsets or pixel-dependent kTC (thermal) noise, would have a significant effect on slopes determined using only frame 0 (Rest 2018). If the count rate measured in frame 0 is very high, the associated Poisson noise can dominate the uncertainty of the reset level, making the use of the single frame 0 data point entirely appropriate.
Tables 2 and 3 give integration times for groups of reads of the full frame detector (2048 × 2048 pixels) using four outputs.
Integration time = Tframe × [(Nsamples +Nskip)*(Ngroups -1) + Nsamples] = Tframe × (Ngroups × Nsamples – Nskip)
having defined Nsamples = Nframes + Nskip .
Tframe = 10.73677 s for the full detector that's read out through four outputs. Note that skipped reads at the end of an integration are not executed.
For example, three groups of SHALLOW4 consist of two groups of five reads plus a final group of four reads. The 14 total reads of the full detector take 150.3 s. Each pixel collects photons for this amount of time.
Table 2. Pixel integration times (s) for groups of short readout patterns for the full detector
Table 3. Pixel integration times (s) for groups of long readout patterns for the full detector
Rest, A., 2018, JWST-STScI-006203
Frame 0 Analysis for NIRCam Integrations
Robberto, M., 2009, JWST-STScI-001721
NIRCAM Optimal Readout Modes
Robberto, M., 2010, JWST-STScI-002100
NIRCAM Optimal Readout II: General Case (Including Photon Noise)