QSI 532 analysis

A thorough analysis of the main characteristics of a QSI 532ws CCD camera employing the Kodak KAF-3200ME chip is given here. Please follow the links at left for more details. The topics include:

  • Bias frames: structure and variation with temperature, long-term stability, analysis of read noise.
  • Dark signal: variation with temperature and time, analysis of 'particle' events (cosmic rays and radioactive decay).
  • Transfer curves: photon transfer with estimate of gain (electrons/ADU) and pattern noise, dark signal transfer curve.
  • Flat fields: estimate of large scale photo-response non-uniformity, vertical banding.
  • Residual bulk image: analysis of charging and decay of traps, simple model for RBI, consequences of RBI effect for photometry.
  • Linearity: measurement of CCD linearity, application of non-linearity correction to data.

During the course of these investigations, several thousand frames were taken using GoQat, often running unattended overnight for long periods. The camera behaves very reliably under Linux and there have been no operational faults caused by hardware, firmware or software during that period. Shutter and filter wheel performance is excellent and the cooling is fast, responsive and stable.


Bias frames contain some structure in the horizontal direction, being essentially a shallow concave curve of about 220 ADU in the centre, increasing by about 4 ADU at the edges. There is a very small, low frequency ripple superimposed on this. Bias frames show a shallow slope in the vertical direction. There is a small variation in the bias level with temperature. Bias frame stability is very good. The effective read noise is 10.5 electrons at -20C. The read noise histogram shows a very clean gaussian profile.

The dark signal generally increases linearly with time and shows the typical doubling about every 6K or so at low temperatures. A small proportion of hot pixels show counts that increase very much faster with temperature than the majority. The signal in these pixels tends to show a declining rate of increase with time. The mean dark current is about 5 ADU per pixel per hour at -20C. Cosmic rays and radioactive decay add about 900 events per hour.

The photon transfer curve is smooth and clean. A typical value for the gain of 1.33 e/ADU is obtained, and the fixed pattern noise (over a square area 128 pixels on a side) is found to be 0.19%. Full well is reached at 53000 ADU, which implies about 70000 electrons. The dark signal transfer curve gives a dark signal non-uniformity of about 3% (excluding the hottest pixels).

Flat fields show a vertical banding of about 0.15% peak-to-peak, with a periodicity of seven to eight columns. The large scale photo-response non-uniformity is no more than 0.25% for white light illumination.

Residual bulk image effects show at least a double exponential decay, requiring 1200s or so for the traps to discharge completely at -15C. There are about 240 electron traps per CCD well. A simple mathematical model is presented that explains the main features of the RBI behaviour. If careful photometric practice is followed, the consequences of RBI are likely to be at the one or two millimagnitude level for relative photometry.

The camera shows a smooth linearity curve, with a deviation of about 2% from zero counts up to full well. By applying a non-linearity correction to data frames, a systematic error of only 0.06% has been readily achieved under laboratory conditions.


Whilst making these investigations I have often referred to the review by Richard Berry, published in Astronomy Technology Today, Volume 2, Issue 2, February 2008. A copy of this publication was received with the camera from QSI. It can be obtained as a PDF document from the QSI website. You can also read more on Richard Berry's website. I have quoted some of his results in places for the convenience of the reader.

The measurements presented on the Linearity page use a very similar range of exposure times to those of Christian Buil, and the presentation of the relative flux curve intentionally follows the style of Christian Buil's relative gain curve. I have quoted his non-linearity parameters to enable the reader to assess the consistency between cameras.

All data reduction was performed with tasks from the Starlink Project. The software continues to be developed at the Joint Astronomy Centre, Hawaii, and is open source. (I am not at the Joint Astronomy Centre and have no connection with it).