The function EXES_readraw.pro is supposed to reduce the raw data to linearized data scaled to flux in 1s. I found a bug where the wrong interval was being used for Fowler modes before. There are still some issues however.
I took some spectral data of the room with the various OT patterns from ND to NNNNNNND, and two more patterns with some spins in the middle.
 |
| Subimage of room spectrum used for pattern flux comparison |
 |
Pipeline flux per sec with readraw. Red is before correcting the interval length in the Fowler mode. Blue is after the correction. We still have problems with NND, NNND, and NNNNNNND, which would have gone through the SUTR part of the readraw code. Green line is a simple difference of [First -Last]/ interval of the raw data (doesnt pass through pipeline). It is scaled by 1/PA_GAIN, as is done in readraw, in order to match the levels. Clearly shows that the flux is linear with interval, aside from some non-linearity.
|
Work in progress. I need to get all patterns to get to the same counts/second in order to do the flux calibration. The problem seems to be with the SUTR part of the code. May just disable that readout scheme and check to see if S/N is affected for those few patterns that use it.
2 comments:
I think we want to have the pipeline default to ND pairs and get rid of the Fowler and SUTR options. I don't think we are read noise dominated at all.
I have a ND pair version working now. The S/N improves for many spectra. examples S/N 35->37 for AFGL 2136 1748cm-1 & mucep himed 384cm-1. But it worsened the really high S/N stuff like IRC 10216 in hilo. Still checking this out, but what I see is the baseline continuum is more bowed in the ND reduction versus the Fowler. I suspect I'll find that the flat field 1st read, 2nd INT is effected by the latency we see when we don't have a T at the end of the pattern.
Post a Comment