Jan 16 2010

SBIG ST-8XME Read Noise Measurement
Posted by admin in astrophotography
Add comment

As discussed in a previous post (here), read or readout noise is the noise or uncertainty added to every pixel value read out of a CCD array. This noise is largely associated with the imperfect conversion of an analog signal into a digital value.

The method to determine a CCD’s readout noise contribution to the final signal is as follows1:

    1. Obtain two bias frames (0 length exposures).
    2. Subtract one bias frame from the other bias frame.
    3. Find the standard deviation for a region of the resulting frame.
    4. Calculate Read Noise (e- RMS) = StdDev * Gain / SQRT(2)

1From “Handbook of CCD Astronomy Second Edition” by Steve B. Howell

The results of performing these operations on ten pairs of bias frames is summarized in the table below:

Pair # Std. Dev Read Noise2,4 Read Noise3,4
Bias Pair 1 11.38 ADU 18.5 e- 20.9 e-
Bias Pair 2 11.26 ADU 18.3 e- 20.7 e-
Bias Pair 3 11.25 ADU 18.3 e- 20.7 e-
Bias Pair 4 11.08 ADU 18.0 e- 20.4 e-
Bias Pair 5 11.44 ADU 18.6 e- 21.0 e-
Bias Pair 6 11.35 ADU 18.5 e- 20.9 e-
Bias Pair 7 11.23 ADU 18.3 e- 20.6 e-
Bias Pair 8 11.24 ADU 18.3 e- 20.7 e-
Bias Pair 9 11.30 ADU 18.4 e- 20.8 e-
Bias Pair 10 11.31 ADU 18.4 e- 20.8 e-
Average 11.28 ADU 18.3 e- 20.7 e-

2Based on published gain = 2.3 e-/ADU

3Based on measured gain = 2.6 e-/ADU

4Published read noise for this device = 15 e-

UPDATED 18 Jan 2010: All values in the table above were updated on this date to reflect the correct Standard Deviation calculation. In the first posting of this data, I was using Maxim DL to subtract one bias frame from another bias frame. I subsequently found out that when Maxim DL performs this subtraction, all pixel values that calculate to a value less than zero are assigned a value of zero since negative numbers are not allowed in 16-bit unsigned integer space. After finding this problem with Maxim DL, I switched to CCDStack for performing the subtraction since CCDStack converts all FIT files from 16-bit unsigned integers to 32-bit floating point values before subtraction. This change meant that instead of the read noise coming out to be slightly less than the published readout noise value (~ 12 electrons per read), the read noise turns out to be slightly more than the published value (~ 20 electrons per read). NOTE: This problem with Maxim DL can be easily avoided by adding an offset value (say 200 ADU) to all pixels in one of the frames then subtract the other frame from this offset frame. This will avoid the less than zero equals zero problem and not affect the calculation of Standard Deviation.

 

Comments are closed.