7.2 Noise and Uncertainty in CCD and CMOS
79
Chap. 11. This section deals largely with the noise sources directly linked to the
CCD or CMOS, while this section discusses how to remove noise.
Not all pixels are created equal; there will always be a variation in the QE between
individual pixels, which may change between exposures. This variation can usually
be adequately addressed by the application of what is known as a flat field, which
is discussed in detail in Practical 2. Knowing the flatness of your camera and the
effectiveness of your flat field is essential in undertaking very sensitive photometry.
The electrons in each pixel are held in place by the application of a small charge.
Intolerances in the makeup of the CCD or CMOS and impurities in the silicon from
which it is constructed causes electrons to move between pixels, a process known
as charge diffusion. Given the nature of the causes of charge diffusion, it will tend
to be found in some parts of the CCD or CMOS and not others. It is also to some
extent wavelength dependent. As with QE, flat fielding will deal in part with charge
diffusion, and for most applications that you will encounter, it will not be a problem,
as other sources of noise may well be dominant.
To fully understand the operation of the CCD, you will need to understand the
processes that lead from an electron being liberated to an image being produced. For
physical reasons, we cannot read all the pixels at the same time. Instead, the camera
firmware and hardware undertake the following read process, as illustrated in Fig. 7.3.
You can consider the CCD to be a matrix of pixels, with each pixel being identified
by a row and column number. In general, the only pixel that is read is the one at 0, 0.
Prior to being read, and as an additional precaution, a small number of additional
electrons are added to the read pixel. This addition, known as the pedestal, is done
to avoid a pixel going effectively negative (and potentially wrapping around, so that
zero becomes a 65,535 count) during later operations. The ADU then converts the
number of electrons into a count by applying the gain. This figure is then passed to an
array. Once the 0, 0 pixel is read and the electrons removed from the pixel so that it
now reads zero, the electrons in the next pixel up in the zero columns are transferred
into the pixel at position 0, 0, and all the pixels in the column are moved down one.
This process continues until the whole column is read. The camera then transfers the
next column over to the now-empty first column, and all columns are moved over
one. This cycle continues until all pixels have been read. The consequence of this
is that read noise increases as we move towards the read column. Some very large
CCDs are in actuality multiple CCDs combined, and therefore have multiple read
edges. In this case, you will see the level of noise increase towards the edges of the
array, as opposed to towards one edge.
The read process for CMOS is slightly different due to each pixel having its own
ADU. In the case of a CMOS, there is no column transfer; each pixel reads out its
value from its own ADU when requested, with the normal order being that the pixel
at position 0, 0 reads out first. Although this process is faster than the readout of a
CCD, it can suffer from a problem known as shutter blur when running at a high
frame rate. Given that very high frame rates are very uncommon in astronomical
imaging, shutter blur is unlikely to be an issue.
Most modern CCD cameras will enable you to perform a function known as
on-chip binning. Binning is a process whereby neighbouring pixels are joined to
79
Chap. 11. This section deals largely with the noise sources directly linked to the
CCD or CMOS, while this section discusses how to remove noise.
Not all pixels are created equal; there will always be a variation in the QE between
individual pixels, which may change between exposures. This variation can usually
be adequately addressed by the application of what is known as a flat field, which
is discussed in detail in Practical 2. Knowing the flatness of your camera and the
effectiveness of your flat field is essential in undertaking very sensitive photometry.
The electrons in each pixel are held in place by the application of a small charge.
Intolerances in the makeup of the CCD or CMOS and impurities in the silicon from
which it is constructed causes electrons to move between pixels, a process known
as charge diffusion. Given the nature of the causes of charge diffusion, it will tend
to be found in some parts of the CCD or CMOS and not others. It is also to some
extent wavelength dependent. As with QE, flat fielding will deal in part with charge
diffusion, and for most applications that you will encounter, it will not be a problem,
as other sources of noise may well be dominant.
To fully understand the operation of the CCD, you will need to understand the
processes that lead from an electron being liberated to an image being produced. For
physical reasons, we cannot read all the pixels at the same time. Instead, the camera
firmware and hardware undertake the following read process, as illustrated in Fig. 7.3.
You can consider the CCD to be a matrix of pixels, with each pixel being identified
by a row and column number. In general, the only pixel that is read is the one at 0, 0.
Prior to being read, and as an additional precaution, a small number of additional
electrons are added to the read pixel. This addition, known as the pedestal, is done
to avoid a pixel going effectively negative (and potentially wrapping around, so that
zero becomes a 65,535 count) during later operations. The ADU then converts the
number of electrons into a count by applying the gain. This figure is then passed to an
array. Once the 0, 0 pixel is read and the electrons removed from the pixel so that it
now reads zero, the electrons in the next pixel up in the zero columns are transferred
into the pixel at position 0, 0, and all the pixels in the column are moved down one.
This process continues until the whole column is read. The camera then transfers the
next column over to the now-empty first column, and all columns are moved over
one. This cycle continues until all pixels have been read. The consequence of this
is that read noise increases as we move towards the read column. Some very large
CCDs are in actuality multiple CCDs combined, and therefore have multiple read
edges. In this case, you will see the level of noise increase towards the edges of the
array, as opposed to towards one edge.
The read process for CMOS is slightly different due to each pixel having its own
ADU. In the case of a CMOS, there is no column transfer; each pixel reads out its
value from its own ADU when requested, with the normal order being that the pixel
at position 0, 0 reads out first. Although this process is faster than the readout of a
CCD, it can suffer from a problem known as shutter blur when running at a high
frame rate. Given that very high frame rates are very uncommon in astronomical
imaging, shutter blur is unlikely to be an issue.
Most modern CCD cameras will enable you to perform a function known as
on-chip binning. Binning is a process whereby neighbouring pixels are joined to