As was documented previously, there is a significant problem in the linearization of channel 3. I show below the means by which the new linearization is computed.

The new values were determined through a brute force solution that stepped through a seach space of calibration factors. It's a set of interlocking IDL and csh scripts that does the following:

1. Generate a new fowlinearize input file, with a fixed linearity curve for all pixels.
2. Feed this to the pipeline fowlinearize module, linearizing all the pipe0 data with it.
3. Multiply the resulting images by the gain correction map and dividing by the flatfield.
4. Locate the standard star, recentroid, and perform new aperture photometry in a 3-pixel radius aperture.
5. Fit to the resulting flux (DN/exptime) vs peak DN in the pipe0 data.
6. Adjust the linearity function coefficient, return to step (1).

It's not elegant but it finds a coefficient that achieves the desired result: namely, that the measured flux of a star is always the same, regardless of how far up the well the brightest pixel has been driven. The results are shown below. Note that in both cases the non-linearity of the Si:As arrays is actually lower than the InSb! The InSb arrays typically have coefficients of around -2e-6.

Ch.3

The corrected data are linear up to about 90% full-well. The coefficient is -6.5e-7. That is a non-linearity of about 3% at 90% full-well. The slope of the line indicates that the data are linear to below 0.1% from 20-90% full-well. This data cannot measure the linearity below 20% full-well.

Ch.3 Flux vs peak pixel unlinearized DN. Red - pipe0 data, flat-fielded and gain-corrected but not linearized. Blue - pipe0 data, flat-fielded, gain-corrected, and linearized using the current online coefficient. White - pipe0 data, flat-fielded, gain-corrected, and linearized using the new coefficient. Green - fit through the white data points.

Unfortunately, there is no way to create a pixelwise solution. We know from the ground that on average all the pixels are about the same, but there were a few "rogue" pixels that had very peculiar linearity curves. The noise observed in the measurements on the curve are almost certainly measurement noise or noise in other cal files, and do not reflect pixelwise linearity differences.

Ch.4

I also rereran channel 4 through the same process. The results indicate a small change to the coefficient from the one currently in use, namely -6.2e-7 instead of -7.1e-7. Otherwise the two channels are very similar, with the channel 4 solution perhaps breaking down slightly earlier, at 80% full-well. At the current time I do not know why the scatter is so much larger near saturation in channel 4 than in channel 3.

Ch.4 Flux vs peak pixel unlinearized DN. Red - pipe0 data, flat-fielded and gain-corrected but not linearized. Blue - pipe0 data, flat-fielded, gain-corrected, and linearized using the current online coefficient. White - pipe0 data, flat-fielded, gain-corrected, and linearized using the new coefficient. Green - fit through the white data points.