In the following section various terms which are used when describing CCD datasets are explained, a little of the rationale for the existence of the various CCD data types is also given. A pixel in following context is one of the CCDs light sensitive elements and should not be mistaken for a data pixel, although there is a one to one correspondence between them.
The bias level of a CCD frame is an artificially induced electronic offset which ensures that the Analogue-to-Digital Converter (ADC) always receives a positive signal. All CCD data has such an offset which must be removed if the data values are to be truly representative of the counts recorded per pixel.
The readout-noise is the noise which is seen in the bias level. This is produced by the on-chip amplifier and other sources of noise in the data transmission before the signal is converted into a digital representation by the ADC. Typically this can be represented by one value which is an estimate of the standard deviation of the bias level values.
In order that the bias level of the CCD system can be constantly monitored (it may at times move due to thermal changes and very occasionally, discontinuous steps) values (columns or rows) are read from the CCD without moving any charge into the output registers. These extra readouts are usually found at the sides of the real data and are often referred to as bias strips or over-scan regions (see Figure 1).
The analogue to digital converter, samples the charge which is returned from the CCD and returns a digital value (usually a 15 or 16 bit value). This value does not equate to the actual number of electrons detected in the pixel in question, but is proportional to it. Typically the proportionality constant is determined by noise considerations — the variance of the actual detected electrons is poissonian, hence the variance in the output from the ADC should equate to this (plus a few other terms such as the readout-noise), so the constant ADC factor can be derived. The output from an ADC is measured in analogue to digital units (ADUs). The ADC factor is multiplicative and converts ADUs into detected electrons.
The capability of pixels to hold charge (charge is entered into a pixel every time a photon is detected) is not infinite and after a certain limit is exceeded the pixel then stops accumulating charge. When the charge in such a pixel is clocked along the CCD (on route to the output registers, from where it is amplified and transferred to the ADC) the excess from it ‘bleeds’ along the readout columns and sometimes even across them. Before saturation slight non-linearities in intensity occur. Data values which exceed this non-linearity limit should be removed from the final datasets and generally cause no further problems. However, because of charge bleeding, contamination may occur around the vicinity and care should be taken when using such data.
All CCDs, at some level, exhibit the phenomenon of dark current. This is basically charge which accumulates in the CCD pixels due to thermal noise. The effect of dark current is to produce an additive quantity to the electron count in each pixel. The reduction of dark current is the main reason why all astronomical CCDs are cooled to liquid nitrogen temperatures. Most modern CCDs only produce a few ADU (or less) counts per pixel per hour and so this effect can generally be ignored. This, however, is not the case for Infra-Red arrays.
The transfer of charge between pixels (and hence along columns) suffers from inefficiencies. Usually this amounts to a charge loss which is never read out from the CCD well - this level is often referred to as the ‘fat’ or ‘skinny’ zero to confuse matters; I refer to it as the deferred charge value. When observing objects with low sky backgrounds (and/or low counts themselves) this loss of charge may be significant (at least in some older CCDs). To overcome this CCDs can be pre-flashed. This amounts simply to illuminating the CCD with a uniform light flux just prior to the actual object exposure. The object counts are then simply added to this pre-flash level of charge in the CCD wells. Note, however, that this method is of no use for very low counts as the signal to noise level which is required after pre-flashing is higher than before (the noise from the pre-flash photons adding to the noise of the object photons). Correction of data for pre-flashing is achieved by subtracting the pre-flash ADU count from the final data (before flatfielding).
The sensitivity of a CCD to incident photon flux is not uniform across the whole of its surface and before data can be said to be properly relatively flux calibrated this needs to be corrected for. The variations in CCD response can be on the large scale (one end of the CCD to the other) and pixel-to-pixel. The relative flux levels on different parts of the CCD are also vignetted by of the optics of the instrument and telescope, this variation also needs correcting for and is performed together with the CCD sensitivity corrections 4.
Flatfield calibration frames are usually taken of a photometrically flat source using the same optical setup as that used to take the object frames. In the past images of the interior of the telescope dome have been used for this purpose, however, it now generally thought that images of the twilight/dawn sky are more representative of a true flatfield, having the same global illumination as the data and having a good signal level (remember that calibration frames will be applied to the object data at some stage and hence will introduce a noise contribution to the final data values, it is therefore essential to get a good set of calibration frames with lots of signal if this process is to introduce the absolute minimum of noise, CCDPACK provides calibration frame combination routines to produce ‘best bet’ calibration frames with very low noise levels), but these frames have a colour response which may be not representative of the colour of the night time sky. If this factor is important then specially taken night sky flatfields must be produced. These can be taken of star free parts of the sky or produced from many object frames whose (contaminating) objects are removed, before median stacking to remove more spurious data values. Note in this final case that the noise levels required to correct for small scale variations are very time consuming to meet.
Some CCD data show an effect known as ‘fringing’. This usually has the appearance of a series of ‘ripples’ in the sky regions. Fringing is caused by the multiple reflection and interference of the night-sky emission lines in the CCD substrate. The effect is considerably enhanced in CCDs whose substrates have been machined thinned to increase the blue sensitivity, the thickness of the substrate being comparable to that of the incident radiation, hence any deviations from a planar geometry cause these ‘Newton Ring’ like effects.
The fringe pattern is an additive effect and must be subtracted. To de-fringe data it is necessary to get special exposures of an object clear part of the night sky, or, alternatively, remove all the contaminations (objects) from data frames with large areas of night sky. These frames should then be combined to give complete spatial coverage and to reduce the noise contribution. This ‘fringe-frame’ should then be scaled to the fringes present on the data frame (after normalisation — MAKEMOS) and subtracted.
4An additional effect of interest, which cannot be fully corrected, is the colour sensitivity of the CCD pixels. Most pixels on a typical CCD frame are exposed to the night sky which has a specific colour, this, however, may not be the same colour as the object itself, so the best case response is that the object and night sky colours mix to produce a response not typical to the night sky dominated parts of the frame, if the object is much brighter than the sky then its colour will dominate and ideally the flatfield should be produced with a source mimicking this colour response.