10.3 Instrument Overheads

10.3.1 Exposure Overheads

The instrument-specific overhead times for WFC3 exposures are dominated by the time to move the filter wheels, to read out the detector, and especially to transfer and store the data. Although in a Phase II proposal the overheads computed with APT may prove to be smaller than the values given in this section, it is nevertheless important to plan your Phase I proposal using the conservative values given here in order to ensure the award of time is adequate to attain your scientific goals.

Several kinds of overhead activities are associated with each exposure, and the specific activities depend on whether the exposure is a new one, or part of a series of associated ones. Associated exposures are defined as exposures of the same target, using the same filter and aperture, executed with or without dithering. Identical exposures are undithered associated exposures. Dithering is the preferred method of acquiring more than one exposure in a given observing mode as it improves the PSF sampling as well as the photometry (averages over flatfield errors as well as UVIS or IR channel image artifacts).

For UVIS (i.e. CCD) exposures, dithered associated readouts are generated when the observer creates images within an APT Pattern container or when applying appropriate POS TARGs to individual exposure specifications. Nearly all observers use dithering as it improves the removal of hot and bad pixels, improves PSF sampling, and improves photometry (averages over flat-fielding errors). In rare situations, identical exposures can be generated if the observer specifies a CR-SPLIT number greater than 1 or specifies a Number of Iterations greater than 1.

For IR MULTIACCUM exposures, CR-splitting is not used, although identical exposures can still be generated with Number of Iterations greater than 1. Dithered associated exposures can be generated as well.  The full set of readouts generated by a Sample Sequence is considered to be one exposure.

At the end of each UVIS or IR exposure, images are read into WFC3's internal buffer memory, where they are stored until they are transferred to HST's solid-state data recorder. The time needed to read a UVIS CCD image is 98 seconds. The time needed for a single read of an IR image is 3 seconds, leading to a total of 48 seconds for a full 16-read exposure. These times are included in the overhead times for the first and subsequent exposures presented in Table 10.2 below.

Dumping of the WFC3 Buffer

The WFC3 buffer provides temporary storage of images read from the WFC3 detectors before they are dumped through the HST science data formatter (SDF) to the solid state recorder (SSR). The buffer can be dumped either between exposures (a "serial" dump), or during an exposure (a "parallel" dump), but cannot overlap any commands executed in WFC3, including the commands at the beginning or at the end of an exposure. The buffer may be dumped during pointing maneuvers, but not during acquisition of guide stars. The buffer may be dumped during target occultation, which does not deduct from the target visibility time. Switching channels from IR to UVIS and vice-versa does not require dumping the buffer (but will require 1 min of overhead time as noted below). Observers will generally prefer to use parallel dumps, in order to more fully utilize the time when a target is visible for science exposures. Although buffer dumps are typically forced by science data volume, a buffer dump will also be forced whenever the buffer holds 304 image headers, regardless of the size of the images themselves. The 304-file limit is unlikely to be reached under typical conditions.

The rules for dumping the buffer in parallel with UVIS exposures differ in some respects from those for dumping in parallel with IR exposures. The two channels are considered separately in the following paragraphs.

UVIS Parallel Buffer Dumps

The buffer can hold up to two full-frame UVIS images. A single full-frame image can be dumped in parallel with a UVIS exposure greater than 347 seconds, and two full-frame images can be dumped in parallel with a UVIS exposure greater than 663 seconds. When the buffer is dumped, all stored images must be dumped. Consequently, a sequence of 348-second (or longer) exposures can effectively "hide" a buffer dump overhead within the exposure time of the following image, thereby maximizing exposure time during the orbit. Whether a sequence comprised of short exposures (less than 348 seconds) and long exposures (greater than 347 seconds) will require serial buffer dumps will depend upon the order of the long and short exposures and the duration of the long exposures. Dumping the buffer during a sequence of short and long exposures will be more efficient if the long exposures are 664 seconds (or longer). For example, an orbit with exposures with exposure times in the sequence 10-348-10-664-10-664-10 will incur no serial dump overhead penalty. The observer will plan such sequences with APT in Phase II. Sequences of full-frame, un-binned exposures less than 348 seconds will require the overhead of serial buffer dumps. For short exposures, using subarrays or binning can help reduce the overhead of serial buffer dumps. The time to dump small subarrays or a binned exposure scales approximately with the number of pixels stored in the buffer.

IR Parallel Buffer Dumps

The buffer can hold up to two full-frame, 15-sample IR images (which, along with the zeroth read, result in FITS files containing 16 reads). To dump one such image in parallel with an IR exposure requires that exposure to be longer than 348 seconds. To dump two such images requires an exposure longer than 646 seconds. The rules for dumping IR exposures are somewhat more efficient than those for dumping UVIS exposures as each readout sample is treated individually.  That is, all samples in the buffer are not required to be dumped together and samples can be dumped during the non-destructive read of the IR detector. Sequences of full-frame, 15-sample exposures shorter than 349 seconds will require serial dumps after the second and subsequent exposures. Sequences of longer-exposure (i.e., greater than 348 seconds), full-frame, 15-sample exposures will incur no overhead for dumping the buffer. Sequences comprised of short (less than 349 seconds) full-frame, 15-sample exposures and long exposures (greater than 348 seconds) may incur overhead for serial dumps, depending upon the sequence of exposures and the duration of the long exposures. The observer will plan such sequences with APT in Phase II. The time to dump an n-sample, full-frame exposure is approximately 39 + 19 × (n + 1) seconds. Subarrays may also be used to reduce the overhead of serial buffer dumps.

Filter and Channel Select Overheads

Both the UVIS and IR channels may be used during a single orbit, although not simultaneously. Only one channel switch (from UVIS to IR, or IR to UVIS) is permitted per orbit, and the time required to reconfigure between the two channels is 1 minute. If the buffer is full when switching channels, then time must also be taken to dump it before the exposure can begin with the other channel. Because the centers of the fields of view of the UVIS and IR channels are the same, acquisition of new guide stars is not required when changing channels to observe the same target.

The overhead for each exposure includes an allowance for the time required to position the filter or grism; however, selecting a UVIS quad filter requires an additional 1 minute of overhead to re-position the telescope, as indicated in Table 10.1.

Table 10.2 summarizes all of the instrument overheads described in this subsection.

Table 10.2: WFC3 Instrument Overhead Times.

Action

Overhead Time

   (minutes)

Reconfiguration between UVIS & IR channels during a single orbit1.0
UVIS ACCUM Mode
Single exposure or first exposure in a set of associated exposures (e.g., the first sub-exposure of a Pattern or CR-SPLIT set)2.6
Subsequent exposures in set of associated exposures (e.g., subsequent exposures in a Pattern or CR-SPLIT set), per exposure2.1
Buffer dump if exposure is not last one in an orbit, or if next exposure is less than 348 seconds5.8
Post-flash - overhead depends on flash level required to reach the recommended background value0.1-0.2
IR MULTIACCUM Mode
Each exposure1.0
Buffer dump if 16-read exposure is not last one in an orbit, or if next exposure is less than 346 seconds5.8

10.3.2 Reducing Overheads with Subarrays and Binning

If your science program is such that a field of view smaller than the full detector size is adequate and you require many short exposures, then one way to reduce the frequency of buffer dumps, and hence their associated overheads, is to use a WFC3 subarray. Subarrays are described for the UVIS channel in Section 6.4.4, and for the IR channel in Section 7.4.4

When subarrays are used, only a small region of the detector is read out and stored in WFC3's buffer. The reduced data volume permits a larger number of exposures to be stored in the buffer before the memory fills and it becomes necessary to transfer them to the telescope’s solid-state recorder. Use of subarrays reduces the amount of time spent dumping the buffer, and also usually reduces detector readout time. An additional bonus of using subarrays is the ability to place the target close to the readout amp, thereby providing some CTE loss mitigation (see Section 6.9).  Note, however, that the full-quadrant UVIS 2K2 and UVIS-QUAD-SUB apertures have somewhat longer readout times than the full- detector apertures because of the way that the readout is performed. A dump is still required if the 304-file limit is reached before buffer memory is filled.

Table 10.3 illustrates the advantage in orbit packing to be gained by using UVIS subarray apertures. We consider the case of a sequence of 5-second exposures without FLASH that fill a 3200 sec orbit as fully as possible. The table lists three subarray apertures of different sizes and a full detector aperture. The subarray apertures have been defined such that they can be read out by one amplifier. The quadrants of the full detector aperture are read out by the four amplifiers simultaneously.

Table 10.3: Orbit structure for an orbit filled with 5-second exposures

Aperture

size (pixels

#exp

Location of Buffer Dumps

UVIS2-C512C-SUB

513 × 512

44

In occultation

UVIS2-M1K1C-SUB

1024 × 1024

28

In occultation

UVIS2-2K2C-SUB

2047 × 2050

12

After 8 exposures & in occultation

UVIS-CENTER

4 × 2048 × 2051

7

After pairs of exposures & in occultation


The areas (ASA) of the supported UVIS subarrays are 1/4, 1/16, or 1/64 of the area (AFF) of a full-frame image. The areas of the IR subarrays are 1/4, 1/16, 1/64, or 1/256 of the area of a full-frame image. The number of subarray exposures that may be stored in the buffer, limited by image data volume, is n = 2 (AFF/ASA). For example, eight 1/4-area exposures may be stored in the buffer, which would allow eight 4-minute exposures to be taken and stored before having to dump the buffer. If the exposures were full-frame, the buffer would have to be dumped after each pair of observations, thus leading to very low observing efficiency.

The 304-file limit must also be considered in optimizing buffer dumps. For UVIS exposures, the limit will almost never be encountered. For IR exposures, each read (not each exposure) counts against the limit. The number of IR exposures that can be stored before a buffer dump is forced is therefore n = 304/(NSAMP +1), or 19 exposures for NSAMP = 15.

In the IR channel, certain combinations of subarrays and sample sequences give rise to images containing a sudden low-level jump in the overall background level of the image. This jump artifact can be prevented by sequencing the IR images within an orbit from large to small (see Section 7.4.4).

Data volume and overhead time can also be reduced for UVIS images by using on-chip binning of adjacent pixels, as described in Section 6.4.4. By using 2 × 2 pixel binning, the data volume is reduced by a factor of 4, although the readout time is only reduced by about 2 to 50 sec. For 3 × 3 pixel binning, the data volume is reduced by a factor of 9, and the readout time by 4 to 23 s. Note, however, that post-flash diminishes the data volume gains of binned mode as the post-flash level required to mitigate CTE losses (see Section 6.9) is the same level as that required for unbinned images. Thus, for example, an image requiring 20 e- post-flash for CTE mitigation will result in 80 e- or 180 e- in a 2x2 or 3x3 binned pixel, respectively, along with the concomitant reduction in signal to noise. IR readouts cannot be binned, but data volume may be reduced by taking less than the default 15 samples during an exposure.