Black vs Room at 586 cm-1

Curtis took some ND pairs of Black and Room while we were on the TAAS during CD26.  I wanted to see if the KRS5 lens in front of the blackbody could be causing any of our fringing.

The data are

;;;     cold.lab.00178.fits  mirror
;;;     cold.lab.00179.fits  room temp blackbody
;;;     cold.lab.00180.fits  sky

"sky" is actually looking at the capped FCLS.  Blackbody and FCLS power had both been off a few hours before taking data, so the sources should be at room temperature.

There were 2 ND pairs in each file.  The pattern was N0 D0 T0, so there was a throwaway after each D. 

For sky and BB, the 2nd ND pair was 0.2% lower counts than the first one.  Almost like reduced response once we got going.  It would be interesting to play with this and see if more Ts helps and/or if the response drops any more with further ND pairs in the file.  

The division of room/BB is not perfect.  Room is 1.5% larger than BB. At 586 cm-1, this would be the difference between 294.5 K and 293 K.  However, there also seems to be some structure to the extra light; it isn't evenly at all pixels.  

BB divided by Room at 586 cm-1.  Image was median filtered by 5 pixels.  Atmospheric line is around (800,600).  This region is closer to 1.000 given the stretch.

Not sure what is causing extra light.  Scattering in room that isn't in BB?  If so, the scattering isn't even.  

As a side note, I looked at the S/N as a function of signal in the three sets.  To get the noise, I differenced the ND pairs in each file (and corrected for the root(2) increase in noise).  I calculated the Standard deviation of difference values for all pixels at a given number of counts.  Room and BB are basically on top of each other.  Mirror seems to have a linearly increase S/N indicating the noise is fixed (not photon?).  Would like to investigate this more since I'd have thought we'd be out of read noise by this count level (60 counts).  Also, the S/N behavior at higher count levels in Room and BB seems to plateau at too low a level.  We shouldn't be going non-linear at 1800 counts or so.  To explore further, we would need a much longer time series so the S/N could be determined at each pixel, rather than with this standard deviation of all differences at a given signal level.

S/N vs Counts for mirror (black), BB (green), and Room (red).