[daip] DFTPL question-2

[Eric Greisen]

Hanno Spreeuw wrote:
> Hello Eric, another question about DFTPL.
> Is it advised to run DFTPL on single channel residual datasets
> (if the source is at the tangent point, so when BW smearing can be 
> ignored) ?
> My lightcurve was quite noisy with 21 channels, but that changed after 
> running UVFIX on the multi channel dataset which moved the transient to 
> the tangent point and then running SPLIT (with APAR(1)=1, NCHAV=total # 
> of channels, CHINC=0).
> I get much smoother lightcurves now.
> Is that due to to the boxcar spectral smoothing from SPLIT?
> The reported error bars in DFTPL (with BPAR(9)=2) are significantly 
> larger than on the multi channel dataset, but also much more realistic 
> (especially after the sqrt(2) correction).
>

DFTPL is not one of the more commonly used programs - I happened to have 
worked on it recently when I discovered that SHIFT was not implemented 
in a uniform manner.  If the time variable source is in the continuum, 
your operation with UVFIX/SPLIT is a correct one - move the phase 
stopping point to the source so that all baselines should be coherent on 
it and then average the channels for best S/N.  DFTPL should have been 
able to move the phase center for you and average the channels so 
UVFIX/SPLIT should not have been essential.  But doing a single channel 
would be noisy.  If you do the multiple channels, the rms is figured 
entering each channel individually in the sums and so should be bigger 
than with SPLIT channel averaging so your low noise previously is a bit 
hard to explain.

Eric Greisen