Here's a litle info on what are in the data file, plus a few comments
and observations.  Scott Hannon, 16 May 2000

What is in the files: (example = m4d_149_avg_srf.mat)
---------------------
  Name             Size 
   favg           256x1  = average freq points
  indbad           1x18  = indices of predetermined bad chans
  indgoodall       1x306 = indices of chans used in average (see testall)
  indsum           1x43  = indices of SRF points (1-256) used in fringe calc
  module           1x2   = module name (char string, no M)
  noiselev         1x1   = max allowed SRF noise level
  savg           256x1   = average SRF
  stdavg           1x256 = standard deviation (std) of data used in average
  stdcentral       1x1   = max allowed std for defringed indsum SRF points
  stdraw           1x1   = max allowed std for raw SRF
  stdwing          1x1   = max allowed std for defringed non-indsum SRF points
  stretch          1x1   = flag: 0=no, 1=yes : rubber-band stretch SRFs?
  temp             1x3   = temperature name (char string=149,151,161, or all)
  testall          1x306 = number of test used in average (see indgoodall)


Things to look at:
-----------------
To see the average SRF,
plot(favg,savg)

To see the standard deviation of the data that went into the average,
plot(favg,stdavg)

The average SRF is often screwy on the first and first points (1 and 256).
I recommend you ignore these two points.

To see which channels (specified by index) of which tests were used for
the average,
plot(indgoodall,testall,'o')


How this data was created:
-------------------------
The raw SRF data is first checked for noise (noiselev) and standard
deviation (stdraw).  Those channels that pas and that are listed as
bad channels (indbad) are passed toteh defringing routine.  After
defringing, the data is again checked for stdard dev using stdwing.
The data that passes the check is then centroid aligned and averaged.
The data may also be rubber-band stretched (if "stretch"=1) to
align the widths.  After alignment but prior to averaging, the data
is checked for standard devition one last time. We use two limits;
one for the central part of the SRF (stdcentral), and another for
the wing (stdwing). All the channels that pass are then averaged.


Comments and observations:
------------------------
This averge SRF data is new and I've not yet had time to check it out.
I think it is OK, but no guarantees of that right now.

The Bruker sinc ringing starts to kick in noticeably around 1300 cm-1.
For modules M5 thru M12, the Bruker dominates the SRF wing.  If the
SRFs are rubber-band stretched to adjust all the channels to the same
width prior to averaging, it will screw up the averaging of the
ringing in the far wing.  The SRFs can be stretched without causing
any problems for M4b to M1a, and things are not too bad for M3 to M4d.
Compare the average for M4d and M4d "stretch"; the bumpiness in the
non-stretched average is the Bruker ringing.

There is significant wing info for M4d thru M1a.  For modules M2b
to M1a, the SRFs were not zero filled.  For M3 and M4a-d, we calc'ed
the average SRF for both zero filled and non-zero filled SRFs.  
If you look at the average SRFs for M4d thru M1a, you will see there
is some structure in the wings.  It is not random noise; it is
consistently there in the measurements, at least for M2b - M1a where
the noise is very low and multiple test averaging isn't needed to see
this stuff.  I went back and checked the old 386/387 data and the
structure is there too.

The wings for M12 thru M5 are a mess.  The Bruker dominates the
measurements, making it difficult to estimate the wings just by
eyeballing the plots.  The wing seems to be pretty much the same
as in the other modules.

The SRFs are a bit asymmetric.  This seems to be real.  I looked
at the raw SRFs for some of the channels that fall between the
fringes, and they are a little asymmetric.  From the 155K CO
trace gas data, we know the Bruker was very well aligned...the
CO lines look great (symmetric), so this can not be blamed on
the Bruker, it seems to be real.  The M12 and M11 SRFs are
a bit asymmetric too. (We do a phase corrrection so it's
probably not that)


Modelling the SRFs:
------------------
We have tried a few different models, but George's G-L model works
OK.  I have not yet done fits on this new average SRF data, but
I have done fits of old averages.  Things vary a bit module to
module, but our "generic" AIRS SRF model is the G-L module
(see matlab routine zx_srf.m) with values:

   gfrac=0.975
   gexp=3
   gslope=0
   lexp=1.8

Scott.