NEWFIRM nonlinearity tests

Using linearity test data taken by Probst on 13 February 2008.

Linearity observing sequence (first with J filter and lights on, then repeated with dark slide in):

	6 x 10.0s reference
	6 x  1.213s 
	6 x  2.5s
	6 x  5.0s
	6 x  7.5s
	6 x 10.0s reference
	6 x 12.5s
	6 x 15.0s
	6 x 17.5s
	6 x 10.0s reference
	4 x 20.0s
	4 x 22.5s
	4 x 25.0s
	6 x 10.0s reference
	4 x 27.5s
	4 x 30.0s
	6 x 10.0s reference
	4 x 32.5s
	4 x 35.0s
	6 x 10.0s reference
	4 x 37.5s
	4 x 40.0s
	6 x 10.0s reference
	4 x 42.5s
	4 x 45.0s
	6 x 10.0s reference
	4 x 47.5s
	4 x 50.0s
	6 x 10.0s reference
	4 x 52.5s
	4 x 55.0s
	6 x 10.0s reference
	4 x 57.5s
	4 x 60.0s
	6 x 10.0s reference

The following analysis considers mean behavior over each of the four NEWFIRM arrays, and is based on sigma-clipped averages computed for each of the four detectors. Pixel-to-pixel analysis will follow.

Note: array order here [1,2,3,4] is as the images appear in the FITS extensions. This order is not the same as the order used for the data from Nov/Dec 2007!

The first plots showing the detailed behavior (frame by frame) of the count rates for the dark and illuminated sequences. Blue points are all exposure times other than 10s, while the red points are the interleaved 10s reference exposures. Note that for the lights-on test, the count rate for the 10s reference exposures first decreases, then increases slightly as the sequence progresses. This might be genuine drift the illumination level, or it could reflect some sort of signal latency from the "bright" frames (where the exposure times get >> 10s and therefore the total recorded count levels are significantly larger). If there is any trend in the 10s reference frames from the dark sequence, it would be a very small decline (increasingly negative averages) through the sequence.

Dark sequence, 4 detectors shown	J-band lights on sequence, 4 detectors shown
(J band - Dark) averages, 4 detectors shown

The next group of plots show the general trends vs. exposure time and count level for the four arrays. The last figure shows the linearity behavior schematically, where I have fit a line to the behavior of the count rate vs. counts data from exposures with average count levels < 10000 ADU, and then normalized the relative count rates from all data to the intercept of that line at n=0 counts (i.e., defining the data to be exactly linear at zero count level). The 10s reference exposures are plotted in red, and were not included in the fitting. The dispersion in their values is evident.

Counts vs. exposure time	Count rate vs. exposure time
Count rate vs. counts	Relative nonlinearity. Linear fit (slope = s, intercept = 1) to data from with counts < 10000 ADU