The first part of this chapter described various detectors (IPCS and CCDs) available at the AAT for optical spectroscopy. How does an observer decide which detector is most suitable for a particular observation?
NOTE: This section of the manual was written in 1991, when the AAO's primary detector was the IPCS &endash; ie. the available CCD detectors were only competitive in a few applications (like direct imaging) where their high read noise was not a factor. Since then the introduction of the thinned TEK chip, and in particular, its XTRASLOW readout mode, has reversed this situation. It is now the IPCS which is only useful in a limited number of applications &endash; so much so, that it is likely to be de-commissioned in the near future. The following discussion has not yet been re-written to reflect this new state, though the tables have been updated to include the TEK figures.
One of the most important criteria in a detector is the time taken to reach the required signal-to-noise ratio. In a typical observation there are three main sources of noise:
These three terms add in quadrature to give a signal-to-noise ratio (SNR):
The signal-to-noise ratio obtained therefore depends on both the efficiency of the detector and the level of the detector noise. The IPCS has poor efficiency , though it does have essentially zero detector noise. The TEK CCD has both high QE and (for almost all applications) essentially negligible read-noise. This has resulted in its being by far the most popular detector on the AAT. Quantum efficiencies for the IPCS and the current CCD detectors are listed in Table 6.7, and plotted in Figure 6.1.
|Differential Quantum Efficiency|
The principle sources of detector noise for the IPCS and CCDs are (for NORMAL CCD readout speed if not specified):
|IPCS||3 counts/pix/hour||Dark Noise|
For the purposes of the SNR calculation, however, we require the effective detector noise per resolution element. This depends on the number of pixels in each resolution element, npix, and the number of separate exposures being added. For nexp exposures of equal duration, it can be simply shown that the effective detector noise per resolution element (DN) is:
where N is the readout noise per pixel given above. In an ideal case, on-chip binning with the CCD will enable the observer to set npix=4 (the limit imposed by sampling theory). To prevent an unacceptable build-up of cosmic ray events, long exposures on the CCD must be broken up into a series of shorter ones; giving nexp~t/1 hour for the GEC chips, and nexp~t/5-20 minutes for the RCA, Thomson and TEK chips.
We have also calculated approximate zenithal count rates for a B=15 [ie. F(4500Å)=4.3 mJy] A0V star observed in a dark (B=22.5 mag/sq.arcsec) sky with the RGO spectrograph + 25cm camera. Typical slit losses ( 1" slit and 2" seeing) are assumed and the count rates apply to a resolution element along the slit. For the UCL echelle spectrograph, these count rates should be divided by . The sky background will be approximately ten times brighter at first/last quarter and one hundred times brighter at full moon.
N.B. These figures are for illustrative purposes only. Accurate estimates of exposure times will require a more detailed calculation of the instrumental throughput (including grating efficiencies, realistic zenith distances etc.).
Using these numbers, we can now inspect the SNR expression above to see which detector will be the most efficient in a variety of different conditions.
The most straightforward case is that of sky-limited observations, when the noise from the sky background dominates that from the detector. This is usually the case for low resolution spectroscopy. This condition is satisfied when:
For practical purposes the IPCS always satisfies this condition, since the dark noise is negligible at all but the very highest resolution, and all IPCS exposures are sky limited.
For CCD detectors and a typical exposure time of 1800 seconds, the condition is satisfied for >> 0.2Å for the IPCS, >> 0.1Å on the TEK chip with Xtraslow readout, >> 0.4Å on the TEK chip with normal readout, >> 1Å on the Thomson, >> 4Å on the GEC and >> 30Å on the RCA in a dark sky. The low read-noise of the TEK chip in XTRASLOW readout means that even for very high resolutions there is no noise advantage to using the IPCS.
For sky-limited observations, the expression for SNR reduces to:
and the `best' detector is simply the one with the highest quantum efficiency, ie. the TEK.
This condition arises for high signal-to-noise observations of bright objects. For the Thomson chip and the TEK chip in NORMAL readout, this condition is fulfilled when:
for the GEC chip when:
and for the TEK chip in XTRASLOW readout when:
assuming `typical' exposure times of 30 minutes. At these count rates the IPCS will be saturated. As in the previous case, the most suitable detector will be the one with the highest DQE. For all wavelengths this will be the TEK CCD.
Finally we consider the case when the detector noise is of the same order as the Poisson noise from the source but the sky background can be ignored, for example in high-resolution spectroscopy. Clearly, detector noise must be minimised, making the only possible choice that between the IPCS and TEK CCD in XTRASLOW readout. As a general rule, the behaviours of the CCD and IPCS at high count rates (see above) makes the CCD the preferred detector for short exposures (< 1 hour) at moderately high count rates (m<17 at 1Å resolution, m<14 at 0.1Å resolution), whereas for longer exposures at lower count rates the IPCS should be considered. This, of course, only applies in the blue (wavelengths less than about 5000Å); the low quantum efficiency of the IPCS in the red makes it an unsuitable choice at longer wavelengths.
In observations where wavelength coverage is important, the size of the detector in the dispersion direction will be an important factor. For the IPCS, Thomson, RCA and GEC the sizes are as follows:
|Number of Pixels in the |
At any given dispersion, the IPCS will therefore give 1.2 times the wavelength coverage of the largest CCD.
The multiplex advantage afforded by the increased wavelength coverage with the IPCS may be iimportant for observations with the UCL echelle where individual orders can be longer than the size of a CCD frame. Twice as many CCD as IPCS frames may be needed to cover the same wavelength range on the UCLES. In such a case the IPCS may therefore be the preferred detector even if the integration times required to reach the desired SNR in each resolution element are 2-3 times longer.
If CCDs with 2000x2000 pixels, good blue response and low readout noise become available, the IPCS is likely to be superseded for all but a few specialized applications.
An advantage of the IPCS is the ability to assess the data in real time. Exposures can be terminated as soon as the desired SNR is reached, with more efficient use of the observing time available.
A major drawback of the IPCS is its behaviour at high count rates. At count rates much above 1 Hz per pixel, coincidence losses start becoming severe (> 10%). This makes the IPCS unsuitable for high signal-to-noise observations of bright objects (see above) unless neutral density filters are used to reduce the count rate, in which case exposure times become very long.
Cosmic rays are an important source of additional noise in CCDs, producing narrow noise spikes in the data when they strike the chip. Typical cosmic ray event rates are 1300, 630, 400 and 100 events/hour/frame for the TEK, RCA, Thomson and GEC chips respectively (though a true comparison of these rates should also comapre the areas of the CCDs). This can produce spurious emission or absorption lines in the final spectrum. Cosmic rays are best removed from the data by breaking long exposures in to a series of shorter ones and subsequently inter-comparing the separate frames. However, as discussed above, this has the serious disadvantage of increasing the effective readout noise.
Although CCDs suffer from a number of cosmetic defects (e.g. trapping sites, `hot' pixels, bad columns) there are now sufficiently large defect-free areas on each CCD for this factor to be unimportant in the overall choice of detector. The IPCS has no similar defects but does suffer from large geometrical distortion. This can, however, be corrected for although some residual distortion will remain at the edges of the data frame when large formats are used.
For spectroscopic observations, count rates may be estimated by the following:
Approximate values for the non-telescope dependent parameters in the above expression are given in Table 6.8. For slit spectroscopy of an unresolved image, f is essentially the fraction of the light from the object entering the spectrograph through the slit. In Table 6.9, f is given for a number of different seeing/slit width combinations.
Table 6.8: Parameters for calculating instrumental throughput
Table 6.9: Slit transmission as a function
Values for the telescope and spectrograph efficiencies (including grating efficiencies) may be obtained from the relevant users manuals and should be used when making accurate calculations of exposure times. However, as an approximate guide, for the RGO spectrograph + 25cm camera, 0.25 for the RGO spectrograph + 82cm camera and 0.1 for the UCL echelle spectrograph.