*) Primary ref for instrumental photometry & stray-light correction is
 Rowe et al. (2006)
 - background handled using
   i) 2-D polynomial fit to remove gradients
   ii) subtraction of flattened, cosmic rejected background
 - instrumental photometry a combination of aperture & psf photometery
 - but background signal still appears even when subtracted away
 - suggests some kind of non-linearity in stray light effects

*) See also Hareter et al. (2008, CoAst, 156), presenting an
 alternative reduction pipeline to Rowe et al. Relies on Reegen's
 decorrelation method.

*) See Huber & Reegen (2008, CoAst) and Reegen et al (2006, MNRAS,
 367) for discussions of decorrelation

*) Observation: best correlation between sky and source is plotting
   sky vs *total* aperture flux; cf. 

*) Sky ADU per pixel in sig Ori E observations can apparently be up near
 the saturation limit (c. 16k counts); however, have images been
 stacked (so saturation limit applies only to sub-exposure)?
  - stacking mentioned in Rowe et al. (2008, ApJ 689)
  - also in Hareter et al. (2008)
  - Huber & Reegen (2008) discuss impact of stacking on decorrelation

*) Secondary ref: in emails from Jason 

   Dated 3/2/2009, 3:43pm CST

    I've had a look at the light curves.  Star 03 appears to be saturated,
    hence the errors have no correlation with the scatter seen.  For sigma
    Ori E (star 02), the scatter appears to be mostly noise.  The errors
    in the data files are shot noise only.  Most of the scatter left over
    is most likely due to intrapixel gain variations.  This will be
    proxied by the X,Y centroid of the star.  The pointing performance of
    the satellite repeats with the orbital period, which adds an
    additional correlation with the stray light.  This makes untangling
    artifacts in the lightcurves difficult.  Hope this helps.

   Dated 1/23/2012, 4:56am CST

   -no background gradient was fit (this was not necessary)

   -pure aperture photometry was used to estimate photometric flux.  An
    aperture with a radius of 4 pixels was used.

   -The 'sky' was estimate from all future [further?]  than 8 pixels from the
    photometric center.  Centroids where measured by fitting the PSF with
    a 2-dimensional Gaussian.

    -Photometry variability of sigma Ori E was observed to be much larger
    (~0.1 mag) than instrumental variability related to the amount of
    stray-light (~0.02 mag).

    -An 8th order sinusoidal series was fit to the data centered on the
    primary period of 1.1908 days and removed from the time-series.

    -A polynomial was fit and subtracted from the residuals [in] the
    relationship between the sky-background and instrumental magnitude.
    If the above step is not taken than the stray light correction can
    produce aliases related to 14.1 c/d

    -outliers where clipped at the 5-sigma level.

    -The sinusoidal fit was added back to the data.

    Dated 1/23/2012, 1:47pm CST:

    The fit did leave behind some residuals, but it doesn't have to be
    perfect, in fact, overfitting can also cause problems.  The satellite
    has an orbital period of ~101 min (~14.1 c/d).  During each orbit the
    background is strongly modulated.  Background levels go from ~3000 ADU
    to 16000 ADU per exposure.  For whatever reason (gain variations?) the
    instrumental magnitude is strongly correlated to the background level.
    To correct this one usually just fits a polynomial to magnitude vs
    background level.  The assumption is that any stellar-variations will
    be averaged out and not affect the fit to the correlation.  For sig
    Ori E, this assumption is not true because the astrophysical
    variations are much larger than the stray-light amplitude.  Since the
    sig Ori E variations are coherent I can pre-whiten the signal, then
    fit for the stray-light.

    The free parameters here are: what order to fit the stray light with
    and how may harmonics to remove from the light-curve.  This was
    answered by looking at improvements (or lack of) from a chi-squared
    perspective.  I used 6,8,10 and 15 harmonics to fit for the
    astrophysics and 2,3 and 4th order polynomials to fit for the stray
    light and looked at the chi-squared statistic of the stray-light fit.
    There was minimal improvement (less than 1 sigma) between 8th, 10th
    and 15th order fits and likewise the 3rd and 4th order polynomials
    showed little difference.  I adopted the least complex functions to
    correct the data.

*) Cameron et al. (2006, CoAst 148) -- use 15-minute smoothing filter
   to eliminate stray light

*) 
