Spectroscopy: Slits

Introduction

In spectroscopic mode UIST will only be used with the 0.12"/pixel plate scale. The instrument has a 16-slot slit wheel which contains field stops, polarimetry masks and slits for spectroscopy. The latter are listed below (see the Engineering pages for the full slit-wheel populations).

Which slits with which grism?

Please note that, because of focus limitations, new slits have been installed in UIST for use with ONLY the IJ and JH grisms. In the OT these are called 2_pix_wide_120s and 4_pix_wide_120s. The standard slits (1_pix_wide_120, 2_pix_wide_120, 4_pix_wide_120) should NOT be used with the IJ or JH grisms; likewise the 2_pix_wide_120s and 4_pix_wide_120s should NOT be used with the other grisms. Users will find that an incorrect combination can not be selected in the OT.

Spectral Resolution

The spectral resolution with each grism is given on the next page. Generally, the 2_pix_wide_120 slit gives roughly twice the resolution of the 4_pix_wide_120 slit. This is not the case with the IJ and JH grisms, however, where only moderate improvement (by ~30%) in spectral resolution is achieved with the narrower 2_pix_wide_120s slit.

The 1 and 2-pixel slits both give essentially 2-pixel resolution. See e.g. these portions of HK spectra: 1-pix spec, 2-pix spec.

Spectro-Polarimetry

The two polarimetry masks, pol_mask_spec_2 and pol_mask_spec_5, each have two aligned, 20"-long slits, separated by about 25". These are used for spectro-polarimetry of an object and adjacent blank sky (see the Polarimetry web pages for further details.).

Recommendation on Slit Position Angles:

If a specific slit position angle is not required, we recommend using an east-west slit (i.e. p.a. set to -90 degrees in the ukirt-ot). When acquiring the source, the "acquisition image" will then appear with the same orientation on the Movie display as a normal "imaging-mode" image (i.e. N to the left and E up). Any tracking errors will also move the source "up and down" the slit, rather than "off" the slit (though such errors are extremely unlikely!).

When a specific position angle is needed, for angles greater than 90 degrees we recommend using NEGATIVE VALUES for the position angle, i.e. for an angle of 110 degrees E of N, enter -70 degrees in the OT. This will help the TSS with source acquisition.

With a position angle of -90 degrees, the TOP of the slit will be to the EAST; 0 degrees puts the top to the south. A table showing the relationship between position angle and slit orientation on the sky is available in another section.

	Spectroscopy: Current Grism Set

Long-slit spectroscopy

UIST has two 9-slot grism wheels which contain the polarimetry prisms and the grisms for spectroscopy. 1, 2, 4, 5 and 7-pixel slits are available for use with each grisms (the exceptions being the IJ and JH grisms, which have their own 2 and 4-pixel slits - see previous page). The table below lists the grisms currently available.

Note that with the wider slits the spectral resolution is reduced, though with the narrower slits it is usually improved, roughly by the ratio of the slit widths. Unfortunately this is not the case with the IJ and JH grisms, where only a ~30% improvement in spectral resolution is seen when the 2-pixel slit is used instead of the 4-pixel slit.

The spectral resolution with the IFU is roughly equivalent to a 2-pixel slit, so is double that given below.

Click on the name of a grism below... to see the relative transmission across the passband of that grism. Each plot shows a spectrum of a bright standard star. The spectrum has been "normalized" via division of an appropriate black-body function, thus giving the transmission (though note that the absolute scale on the y-axis is arbitrary). Absorption due to the atmosphere plus the telescope and instrument optics (especially the grism and spectral blocking filter) all contribute to the overall shape of each plot. Photospheric absorption lines associated with the standard have not been removed. Note that UIST's throughput drops quite considerably towards the I-band, and that the long-wavelength end of the JH grism is suppressed by the blocking filter (see below).

A comparison of IJ, JH and HK spectra obtained through a 4-pix slit of HIP 87895 (G2V, V=6.3), after division by an A0V star (HIP 85382, 5.9 mag) for telluric correction, is given here (wavelength scales are approximate).

Note also that in RAW data frames the wavelength increases to the LEFT ; the pipeline software will, however, display the wavelength increasing to the right (further details are given in the pages on Data format).

Low versus Moderate-Resolution Grisms

Should I use a moderate-resolution or a lower-resolution grism? The answer depends on your needs. The relative transmissions are similar. However, with most of the higher-resolution grisms background-limited performance is essentially impossible, so (read)noise on the array can be a dominating factor. The higher-resolution grisms work well if one is trying to detect line emission superimposed on continuum emission, and obviously they offer higher spectral resolution.

Figure 1: Comparison of short-K (left) and HK (right) spectral images of the same target, using the same (60 second) integration time. The data are flat-fielded and sky-subtracted, though not corrected for telluric absorption (i.e. no division by a standard star). In addition to continuum from the star, faint H2 emission at 2.122 microns is detected along the slit.

To compare the performance of the HK and short-K grisms, a young star with a line-emission jet was observed with both grisms, using the 4-pixel slit and 60 sec exposures in both cases. In both datasets, continuum from the star and weak line emission from the jet were detected (Fig.1). However, in extracted spectra from the jet (Fig.2 - left) a low-frequency "ripple" is evident in the short-K data which isn't apparent in the HK data: this is produced by a "chevron" readnoise pattern across the array, which though variable can dominate the noise at very low flux levels.

Figure 2: Comparison of short-K (red) and HK (yellow) spectra extracted from along the jet (left) and from the star itself (right). Rows 795-805 were extracted and coadded for the left-hand spectra; optimal extraction was used for the right-hand spectra (for display purposes 3000 counts were added to the short-K spectrum at right). The spectra have not been smoothed.

The short-K spectrum extracted from the star fairs much better (Fig.2 - right): in this case the main source of noise is the continuum from the star. At the higher spectral resolution of the short-K grism, the (spectrally-unresolved) line-emission towards the star is more prominent in the short-K data than it is in the HK spectrum.

	Spectroscopy: Sensitivity Tables

These sensitivities are based on data taken with the 4 pixel slit in seeing conditions of 0.5-0.7 arcseconds (i.e. average). They depend closely on the slit width chosen and the delivered image quality. Under similar conditions, a 2-pixel slit would yield 1/sqrt(2) times the signal-to-noise ratio (due to halving the slit width [the increased resolution won't affect the S/N because the flux per pixel doesn't decrease, i.e. narrower slits don't give smaller wavelength coverage...]). The sensitivities also assume nodding the source up and down the slit; for extended objects which require nodding to blank sky, sensitivities will be 0.4 mag (or a factor of 1.5) poorer, i.e. subtract 0.4mag from the numbers below.

Those preparing telescope proposals should also consider the overheads associated with spectroscopic observations. These can be considerable for short exposure times (see the section on read speed and efficiency for details). For brighter science targets, two minutes per source, e.g. one "quad" (one object-sky-sky-object group) comprising four 30sec exposures, is probably a sensible "minimum time" to spend on each target (provided they don't saturate, of course).

The sensitivity figures are measured near the center of the wavelength range covered by the grism, or at a wavelength where the transmission is good. Sensitivity may decrease towards the edges of the wavelength range, or between atmospheric windows. Click on the grism name to get a rough idea of how the transmission changes across the wavelength range. See also the discussion on obtaining background limited performance.

Which values should I use...?

• For Point, continuum sources - Use Table 1
• For Point, line-emission sources - Use Table 1 (see notes)
• For Extended, continuum sources - Use Table 2, column 3
• For Extended, line-emission sources; line spectrally RESOLVED - Use Table 2, column 5
• For Extended, line-emission sources; line spectrally UNRESOLVED - Use Table 2, column 6

• The above values were derived from the signal-to-noise measured on an optimally-extracted spectrum of a faint standard star.
• Values apply to the wavelength specified in column 2. The S/N will of course degrade towards the edges of the atmospheric window associated with each grism (as noted above).
• Values are per spectral pixel, so this is what you'd get with unsmoothed, though coadded continuum source data. The spectral resolution is ~4 pixels, so sensitivities per resolution element (e.g. for comparison with CGS4 figures) will be 0.8mag better (add 0.8mag to the above numbers). To get the "per resn element" sensitivity, observers would have to bin their spectra over 4 pixels.
• For spectrally unresolved emission-line point sources, remember that the line flux will be spread over 4 spectral pixels (with the 4-pix slit), so for your source you must use a line magnitude per spectral pixel (add 1.5mag with the 4-pixel slit) before comparing your line flux with the above table: e.g. for a 15.6mag line spread over 4 pixels, will effectively get 17.1mag per spectral pixel and so would expect 3sigma in 30mins with the HK grism. (Smoothing spectra will again improve sensitivity for fainter lines of course.)
• In the non-thermal, the tabulated performance was achieved with long (240sec) exposure times; in the J-band (IJ grism), 600 second exposures were used. Shorter exposure times will probably give poorer performance (discussed here).
• At thermal wavelengths extracted spectra were divided by a standard star to remove telluric absorption features. Sensitivities at these wavelengths (particularly at short_L and M) will be dependent on stable atmospheric conditions and good sky-subtraction.

• The surface brightness sensitivities in columns 4-6 were derived from per resolution element point source values (i.e. the values in column 3, plus 0.8mags for the 4-pixel slit/resolution). They are therefore appropriate for spectrally unresolved emission-line sources (i.e. where the quoted flux is spread over 4-pixel resolution element).
• To get the mag/arcsec sensitivities in column 3 and 4, we have assumed flux from point source calibrator spread over 10 rows, or 40 pixels with the 4-pixel slit (so first ADDED 2.5 log(sqrt[40])); the pixel scale is 0.12 arcsec (so then SUBTRACTED 2.5 log(1/0.12")).
• For weak emission lines superimposed onto bright continuum sources the S/N will be worse than the 3sigma prediction (because of the additional poisson noise). Here's one possible way of doing it.
• IMPORTANT: All extended source sensitivities assume nodding the source up and down the slit; for extended objects which require nodding to blank sky, sensitivities will be 0.4 mag (or a factor of 1.5) poorer, i.e. subtract 0.4mag from the numbers below.

	Read Modes and Noise

The information below pertains to the new ARC (formerly SDSU) controller commissioned with UIST in December 2006. For numbers specific to the old Edict system, please contact the instrument scientist.

Subarrays

In imaging mode, readout areas of 1024x1024, 512x512 or 256x256 are available, each with either the 0.12 arcsec or 0.06 arcsec pixel scale. Subarrays are centered on the centre of the full array (unlike UFTI, there is no speed advantage when using an off-centre quadrant).

In spectroscopy mode, UIST is always used with the full array readout and the 0.12 arcsec pixel scale, giving a long-slit spectroscopy slit length of ~2 arcminutes. Subarrays are - strictly speaking - not available in spectroscopy mode.

NDSTARE (NDR) and Digital Averages

NDSTARE is UIST's non-destructive readout mode. It is the default mode of operation for all spectroscopy, all IFU, and all non-thermal imaging. To minimise noise, as many reads as possible (one per second) will be fit into each NDSTARE exposure. In other words, the array is reset then read N times where N is at least 2 (for a 1sec exposure). The minimum exposure time (full array) with NDSTARE is 1.0 seconds.

To reduce extraneous noise each pixel is also sampled multiple times on readout and the result is "digitally averaged". As a compromise between reducing noise and adding overheads, with NDSTARE the number of digital averages is always (automatically) set to 4.

With NDSTARE a reverse bias of 600 mV is used.

Correlated Double Sample (CDS)

CDS, or Correlated Double Sampling, is available in imaging mode with each full and sub-array. In this mode the array is reset then read out just twice, at the beginning and end of the exposure (more detail is given here). 4 digital averages are taken per read.

CDS should only be used if a full-array readout with a very short (less than 1 second) exposure time is required, or if linearity is likely to be a problem. CDS might be useful on very bright targets where the full array is needed. The minimum exposure time (full array) with CDS is 0.623 seconds, which can of course be used with a number of coadds.

With this mode a reverse bias of 600 mV is used.

Thermal Imaging

Two readout modes are available for thermal imaging, THERMAL CDS and THERMAL ND. As the names suggest, CDS and ND are correlated double-sample and non-destructive readouts, respectively. Both modes utilise a higher reverse bias (900mV) which gives increased full well depth (saturation at a higher count level). However, these modes do no digital averaging (multi-samples = 1), facilitating faster readout.

Minimum exposure times (full array) are 0.172 sec and 0.20 sec for Thermal CDS and Thermal ND, respectively. Shorter exposure times are possible with sub-arrays.

Imaging Polarimetry

Imaging polarimetry is unusual in that long and very short exposures are often required; the former for sky flat fields, and the latter for bright polarisation standards (which are often 7-9th mag). A read mode called IRPOL CDS is available for these circumstances. It is essentially the same as the THERMAL CDS read mode, except that the NULL read (see below) that precedes the first read of the CDS has a minimum exposure time (the dwell time on the NULL is set to 0.001 sec).

NOTE - IRPOL CDS is only available with the full 1024x1024 pixel array.

The following is NOT required reading for observers

Noise Reduction with Digital Averaging

Data in the top two and bottom two quadrants of the array are presented separately. Increasing the number of averages clearly reduces the noise. However, more digital averaging leads to increased overheads. A value of 4 has therefore been adopted for non-thermal readout (NDSTARE and CDS).

Array tests, which use the MEASURE_READNOISE DR recipe, should be run at the beginning of each night of UIST observing. This will give a measure of the current readnoise on the array in a 1sec (4 digital averages) exposure. Past values are stored in a text file in /ukirt_sw/logs.

Noise with NDSTARE - Multiple Reads and Digital Averaging

Read noise decreases with increasing exposure time, as shown in the plot below. In these data the digital averaging has again been set to 4; the number of NDSTARE reads increases with increasing exposure (one per second). The readnoise drops to around 8 electrons with long (greater than 60 sec) NDSTARE exposures; long exposures are clearly desirable with short-wavelength imaging and all IFU and long-slit spectroscopy of faint targets. (Data courtesy of David Atkinson, UKATC.)

Null Reads

All UIST exposures are preceded by a NULL read, i.e. a read that is thrown away. The duration of this NULL read is the same as the total exposure time in CDS, or the read interval in NDSTARE (dT = 1 sec for full-array NDSTARE readout). Read intervals for the various sub-arrays and read modes are tabulated on the next page. (The only exception is the IRPOL CDS mode which has a fixed, very short NULL read - see above.)

As an example, in a 10 second NDSTARE exposure (full array), the controller returns 12 reads, the null read, plus 11 reads separated in time by 1 second. A 10 second CDS exposure would return only three frames, the NULL plus two further reads separated by 10 seconds.

The array is reset between the NULL read and the first read of the exposure.

Idling

With the ARC controller, the array is continuously RESET between exposures (as was the case with the Edict controller); the settling delay between resets is kept short at 0.01 sec.

Idling is enabled as soon as an application (exposure with a given read mode) finishes executing. However, note that downloads to the controller do disable Idling - until the application has finished executing.

Periods of Idling can reduce the signal on the subsequent frame by 10-20 counts; at the telescope this may manifest itself as a dark with slightly negative counts. We did experiment with Idling that consisted of a RESET+READ. However, we found that when filter wheels were moved while idling (even when using a fast, 0.1 sec read), light was getting onto the array as an open or thermal filter passed over the array. This resulted in the first frame having high counts; in some cases this latent signal was hundreds of counts in a 60sec exposure.

	Read Speed, Exposure Times and Efficiency

The information below pertains to the new ARC (formerly SDSU) controller commissioned with UIST in December 2006. For numbers specific to the old Edict system, please contact the instrument scientist.

Maximum Exposure Time

IMPORTANT: Currently, because of memory limitations on the acquisition machine, the MAXIMUM exposure time possible with UIST is 240 seconds.

Read Speeds and Minimum Exposure Times

The array can be addressed or "clocked" at various speeds. However, the faster readout rates lead to output coupling, where a fraction of the signal on one pixel is picked up on a pixel 8 columns away. Read speeds are therefore limited by a desire to minimise this effect.

With NDSTARE readout of the full array (used in non-thermal imaging and all spectroscopy modes), a read time of 622 millisecond has been adopted; with an additional dwell time of 378 secs, this means that the array is read out every second. With a 10 second NDSTARE exposure, the array is read 11 times (12 including the NULL read), with each read being sampled four times (the digital averaging). Faster readout is possible with sub-arrays, and of course the thermal readout modes (used only with thermal imaging), run with faster readout clocks leading to shorter read times, as shown below.

The READ, DWELL and TOTAL times in the above table represent the following:

• The READ time is time it takes to physically read out the whole array.

• The DWELL is effectively a pause after the READ:
  • The DWELL is fixed with NDSTARE - longer exposures are created by increasing the number of reads. For example, a 0.6sec NDSTARE 512 exposure would consist of a Null read, a 0.184sec read, a 0.016sec dwell, a 0.184sec read, a 0.016sec dwell, a 0.184sec read, a 0.016sec dwell, and finally a 0.184sec read.
  • The DWELL is variable with CDS - longer exposures are created by increasing the dwell time. For example, a 2sec CDS 512 would consist of a Null read, a 0.184sec read, a 1.816sec dwell, then the final 0.184sec read.

• The TOTAL time listed in the above table is READ + DWELL, and therefore represents the minimum exposure time with each readout mode: your exposure time can be (much) longer than this.

Readout Overheads and Efficiency

As described earlier, digital averaging is used to beat down the read noise in each exposure. The read noise also decreases with longer exposure times, reaching an optimum above about 60 seconds. Long exposures are also by far the most efficient. Only saturation on source or non-linearity, or the desire to collect data before the sky background changes appreciably, should limit the exposure time used.

If short exposure times must be used, then it may be desirable to combine these with a number of co-adds. Although this makes each individual observe less efficient, overall this may be more efficient when considering the time taken to nod the telescope between "source" and "sky". Increasing coadds also brings down the read noise (a little).

Optimum exposure times

The optimum exposure time is dependent on many factors, such as resolution, wavelength, object brightness and weather conditions. Consequently, although clicking on "Default" in the instrument configuration in an OT sequence will yield a reasonable exposure time for a specific source magnitude, you may need to fine-tune this value. Note, however, that generally the maximum possible exposure time is the optimum exposure time, as overheads are reduced to a minimum.

Spectroscopy tests during commissioning, where the same source was observed for the same total on-source period of time, but using 10sec, then 30sec, then 120sec and then 240sec exposures clearly showed that the better detection was obtained with a few long exposures as compared to many short exposures. Indeed, it seems that 200-240 sec exposure times are optimum for non-thermal spectroscopy of faint sources, or half of this if the OH sky lines aren't being subtracted off too well. In imaging mode, a five or nine-point jitter pattern should probably be acquired within 5-10 minutes, so that a suitable flat field frame can be created from the median average of the jittered target images.

	Gain and Linearity

Much of this is NOT required reading for observers; a linearity correction is currently being applied in the ORAC-DR pipeline.

The information below pertains to the new ARC (formerly SDSU) controller commissioned with UIST in December 2006. For numbers specific to the old Edict system, please contact the instrument scientist.

Bias Voltages and Saturation

The detector reverse bias is automatically selected by the low-level software. With NDSTARE readout (with full and sub-array imaging and all spectroscopy modes) 600 mV is used; for thermal imaging (THERMAL ND and THERMAL CDS) 900 mV is used. The latter gives increased well depth and better linearity, which are important with the high background flux encountered at L and M.

The array goes into hard saturation at around 21,000 counts (~135,000 electrons) with NDSTARE; in the thermal saturation occurs at 33,200 counts (~210,000 electrons). See below for details...

Dark Current

At 600 mV reverse bias the dark current is ~0.1 e-/pixel/second. At 900 mV a value of 0.4 e-/pixel/second has been measured.

Pixel Settling Time and Output Coupling

Performance in the thermal is improved by increasing the reverse bias, which increase the full well depth, but also by reducing the readout speed. In thermal readout modes video processing has been minimised so that the pixel processing time is dominated by the pixel settling time. 2.5 microseconds is the minimum pixel settling time (limited by the fibre optics speed). However, as the settling time is reduced, output coupling increases. That is, the fraction of signal on an output channel that remains when the channel is next sampled. This produces a ghost image 8 columns along the detector. The output coupling resulting from a 2.5 microsecond pixel settling time is shown here.

For non-thermal readout a pixel settling time of 7.1 microseconds has been adopted. This leads to an output coupling factor of only 0.44%. In the thermal, a settling time of 4.0 microseconds is used, which gives output coupling of 1.16%.

Gain and Linearity

Tests with the ARC controller suggest that there is no single value for the system gain. Rather, this seems to vary with flux level, from around 5.75 electrons per data number (per count) at ~10% full-well, to ~6.65 electrons per DN at ~90% full-well with 600 mV reverse bias (non-thermal readout). Similar behaviour is seen with 900 mV reverse bias (the gain is slightly lower because of reduced detector capacitance).

Linearity curves for the 600 mV (non-thermal) and 900 mV (thermal) bias settings are shown below. The blue lines give the linearity in data numbers; the pink lines give linearity after correction by the flux-dependent system gain level noted above. This correction results in a linear plot.

In the thermal, it is recommended that the same exposure time (and/or counts on the source and sky) is used with the target and the photometric standard star.

Click for an expanded plot

Linearity - 600 mV Non-thermal readout	Linearity - 900 mV Thermal readout
Residuals - 600 mV Non-thermal readout	residuals - 900 mV Thermal readout

NOTE: use the right-hand (900 mV) plots ONLY with thermal imaging modes. If you're unsure about your data, check the DET_MODE fits header: 900mV is used with modes that end in 'T', e.g. ND1T, CDS1T, etc.

Interpixel Capacitance

Finally, there is very little interpixel capacitance coupling on the UIST array. Approximately 10% is transmitted to the surrounding 4 pixels, either side in a row and top and bottom in a column (with both 600 mV and 900 mV reverse bias).

Data and analysis courtesy of David Atkinson, UKATC (Winter 2007).

	Spectroscopy: Saturation and Sky Counts

IMPORTANT UPDATE: A new, more efficient, controller was commissioned in May 2007. This means that currently, because of memory limitations on the acquisition machine, the MAXIMUM exposure time possible with UIST is 240 seconds.

On the Quest for Background Limited Performance (without saturating on source)...

It is very difficult to obtain background limited data with the IJ, short-J, long-J, short-H, and long-H grisms. With the 4-pixel (0.48") wide slit the required exposure times are 10 minutes or greater. A target line that happens to coincide with a sky line could be "background limited", though this of course depends on the strength of the OH line.

The table below gives sky counts and backgound limited exposures for the longer wavelength grisms and the 4-pixel wide slit. Note, however, that these change with wavelength. With the HK grism (and the 1-2.5 micron short-/long- grisms) the "longest possible" exposure time is recommended for best performance (for recommendations on exposure times please have a look at the section on preparing observations for UIST spectroscopy).

In the thermal, the "maximum possible exposure times" simply give the time to saturation on sky lines or the thermal background - they are not recommended exposure times . For other slits scale the sky counts and exposure times by the slit width (for smaller slits decrease the sky counts and increase the exposure time).

In the above the time to reach background limited performance is equal to RN*RN/(sky counts * G) . RN is the readnoise (we assume 15 electrons -- expected for a longish exposure time) and G is the gain (to convert counts to electrons); the sky counts used is that given in column 2.

Note regarding the IFU: Per-pixel on the array, the IFU is equivalent to a 2-pixel slit, so the numbers in the last column in the above table should be doubled!

Saturation on Source

Saturation on source is only potentially a problem for very bright targets with the low-resolution gratings (and widest slits); e.g. for the HK grism saturation occurs in the shortest exposure time (1 second) for targets with K < 3.

Below we list the brightest point source that can be observed in a given exposure time. The figures were derived from observations of bright standards in median seeing. Obviously, wider slits and/or better seeing will affect the magnitude limit; a 7-pixel slit will potentially transmit twice the signal from the source (and the sky), so the magnitudes below would be increased by 0.75mags.

In summary, point sources that are fainter than 8th magnitude can be observed with 240sec exposures with all grisms except those in the L and M-bands.