The LMIRCam detector is a Teledyne H2RG device with sensitivity from 1-5 µm wavelength. From May 2011-November 2011 the detector used is/was S/N 264, which is on loan from the ETH Zurich group. The detector is anticipated to be replaced in late 2011.
Low bandwidth high bandwidth
Fast Single 10.1 15.5
Fast CDS 11.1 18.6
Slow Single 5.2 6.6
Slow CDS 3.3 6.4
My method for performing the analysis is as follows:
- for a given set, I choose two frames and take their difference, which removes physical variations across the detector.
- I then plot the histogram of the pixel distribtution and fit a Gaussian, which fits the distribution very well
- The sigma of the Gaussian fit then represents the standard deviation of the pixel values in differenced frame
- I divide by the sqrt(2) to obtain the standard deviation for a single frame
- I repeat for all combinations of frames and find consistent values
If there's no noise from the controller or other external sources, then the final value should represent readnoise.
Fast Mode Single: sigma_single = 9.37 DN
Fast Mode CDS: sigma_CDS = 11.59 DN
--> sigma_single = 8.19 DN
Slow Mode Single: sigma_single = 9.62 DN
Slow Mode CDS: sigma_CDS = 14.02 DN
--> sigma_single = 9.92 DN
Slow Mode CDS (motors off): sigma_CDS = 6.64 DN
--> sigma_single = 4.69 DN
In addition, I notice significant pattern noise in the slow mode frames (especially when the motors are running), so I ran my pattern removal routine through the entire data set (essentially, take the median of the channels to extract the pattern noise and remove it). I suspect that with the addition of the preamp, the pattern noise has been reduced sufficiently that readnoise dominates while in fast mode. Here are the results after extracting and subtracting the pattern noise:
Fast Mode Single: sigma_single = 8.34 DN
Fast Mode CDS: sigma_CDS = 9.54 DN
--> sigma_single = 6.74 DN
Slow Mode Single: sigma_single = 5.26 DN
Slow Mode CDS: sigma_CDS = 3.48 DN
--> sigma_single = 2.46 DN
Slow Mode CDS (motors off): sigma_CDS = 3.22 DN
--> sigma_single = 2.27 DN
There is probably other noise from the unintentional case grounding that doesn't manifest itself as an easily extractable pattern. I suspect that once we remove this grounding, we will see significant noise improvement.