R. S. Bogart & R. K. Ulrich
v. 0.8 94.07.12
-
1. Observables
-
At each pixel in the full-disk mode (2" square): two observables
are calculated from the five averaged line filtergrams F1 - F4,
Fc, and the constant ADC bias according to eqns. 1-4 and 14-17
of the IPFR. One of these quantities, sensitive to spots and
pores, is the Continuum Intensity proxy Ic (eqn. 17); the other,
sensitive to faculae, is the Line Depth proxy Idepth (eqn 14).
The Continuum Intensity proxy is the same observable as used
by the Flux Budget part of the Structure Program (v 0.9). This
specification only refers to the Line Depth proxy under the
explicit assumption that the Flux Budget program will use an
identical spatial and temporal filter for the Continuum proxy.
The TAI time of observation (center of time average) and the
camera location with respect to the Sun (SOHO ephemeris) are
also required.
-
2. Cadence & Duration
-
Observations to be made once per minute as part of the low-rate
Structure Program, without interruption, from nominal commencement
through nominal end of Structure Program (2 years on station).
-
3. Spatial Filtering
-
The full image 1024*1024 per pixel data are boxcar averaged into a
128*128 square array (8*8 pixels per averaged pixel). Bad pixels
(dark current > 5 dn) are not ignored in the spatial averaging.
Two separate buffers for temporal filtering are maintained. In
the current specification these buffers are spatially coincident.
In future revisions it is possible that one buffer will be offset
from the other by 4 pixels in each direction, so that it would only
be a 127*127 square array, the 4 rows and columns at each edge of
the image being stripped.
-
4. Temporal Filtering
-
The data are averaged weighted by a Gaussian truncated to 0 at
tc +/- 12 min (thus, only 23 one-minute samples are included, and
2 accumulation buffers are required). The weighted averages of
the Line Depth proxy are sampled concurrently with the weighted
averages of the Continuum Intensity proxy once every 12 minutes
for telemetry. The Gaussian weights, normalized to unity at tc,
correspond to a standard deviation of 204 seconds (3.4 minutes)
as follows:
tc 1.000000 tc +/- 6 0.210748
tc +/- 1 0.957669 tc +/- 7 0.120106
tc +/- 2 0.841129 tc +/- 8 0.0627770
tc +/- 3 0.677549 tc +/- 9 0.0300931
tc +/- 4 0.500553 tc +/- 10 0.0132301
tc +/- 5 0.339149 tc +/- 11 0.00533448
The actual integer weights will be adjusted to avoid overflow in
the temporal sums (TBD).
The specified temporal window provides a low-pass filter with a
smooth roll-off to a value of 0.1 (-20 db) at 1.7 mHz, 0.01 at
2.4 mHz,first zero at 3 mHz, and transmission < -60 db (0.001)
for all frequencies > 2.8 mHz. With this filter no more than
about 40% of the power below 1 mHz should be due to aliasing of
higher-frequency p-mode power.
-
5. Calibration
-
No calibration (in the sense of correction of values calculated
on board to take account of for example known drifts in the filter
response function) is presently planned: the level 1 (calibrated)
data as delivered are identical with the level 0 (extracted and
sorted) data.
-
6. Data Organization
-
Averaged pixels lying wholly outside the maximum solar diameter are
discarded prior to telemetry. The remaining averaged pixel values
are arranged in a fixed stream, order TBD. The order of this stream
is preserved as the data ordering through level 0 and any level 1
processing.
-
7. Data Representation and Precision
-
It is assumed that individual filtergrams are digitized as 16-bit
integers normalized to values of 4095 for full-well, that the typical
exposure value is one half of full-well (~2048) per pixel at disc
center, and that the counting statistics are such that the shot
noise to signal ratio per pixel at disc center for each of the
accumulator quantities S_a, S_b, etc. is approximately 0.001, so
that 10 bits are required to represent the signal portion of the
quantity. Then the relative noise-to-signal per pixel for the
Line Depth proxy should also be approximately 0.001.
The spatial averaging will reduce the noise-to-signal ratio in
the resulting bin by the square root of 64, so that 13 bits will
be required to represent the normalized average without noise.
The temporal averaging involves averaging over an effective 8.5
samples (the sum of the Gaussian weights), implying a further
reduction in noise-to-signal by the equivalent of 1.5 bits.
The quantities to be transmitted should thus be represented as
normalized 15-bit quantities prior to lossless compression.
The above numbers are conservative estimates, as they are based on
disc center values. Away from disc center, the signal-to-noise
ratio will be lower and the corresponding precision less. It may
be possible to normalize the signal to 14 bits prior to compression.
If the average number of bits per bin of compressed averaged data
for the Line Depth proxy is 10.5, then the required telemetry is
175 bits per second.
-
8. Expected Future Revisions
- 0.9
- Refinement and specification of clipping mask
- Specification of requirements for flat-fielding, calibration;
desiderata for tuning
- Reflection of updated IPFR
- Possible use of Gaussian spatial filter
- 1.0+
- Possible change in locations of spatial masks for dithering
- Possible revision of observables
- Possible extension of Structure Program
-
9. Notes, Exceptions
-
The Flux Budget Continuum Intensity observations of the Structure
Program as specified in V 0.8 are also required for this program.
If the GOLF observations commence before SOHO is on station and if
it is possible to do so, both the Continuum Intensity and Line Depth
proxy observations should commence at the same time.
One or more campaigns may be planned to be run and analyzed prior
to the final definition of the GOLF observable and calibration, in
which case the initial observing sequence will be provisional.
It may be desirable or necessary at the level 1 calibration to
reconstruct the data stream into images and map into another
coordinate system, e.g. heliocentric.
Both the spatial and temporal filters discussed here differ from
those in the IPFR. In particular the temporal filter is more
complicated than the asymmetric filter of the IPFR.
-
10. Revision History
- 0.1 - 94.03.24:
- Revision history added (10)
- Changed baseline temporal filter from triangular to
trapezoidal taper (4)
- Calibration description reworded (5)
- 0.2 - 94.06.06:
- Changed baseline temporal filter from trapezoidal
to truncated gaussian (4)
- Changed second observable from average of F2 and F3
to calculated line core intensity, less sensitive
to Doppler shift and tuning (1)
- Added suggested but unused criteria for throwing out
bad or dark pixels from spatial averages (3)
- Added suggestion to spatially average numerator and
denominator of magnetic proxy separately (9)
- 0.7 - 94.07.06:
- Advanced revision number to nearer 1.0 (10)
- Changed observables from core filtergram and magnetic
proxy to continuum intensity and line depth (1)
- Added explicit tie to Flux Budget program (1, 4, 9)
- Changed spatial binning from 10*10 to 8*8 and added
requirement of separate buffers to allow for dithering
in future revisions (3)
- Removed irrelevant comment on averaging of pixels for
magnetic proxy (3)
- Changed sampling frequency from 20 minutes to 10
minutes (4)
- Added section on future revisions (8) and renumbered
- Removed superseded notes and exceptions regarding
choice of observables (9)
- 0.8 - 94.07.11:
- Changed sampling frequency from 10 min to 12 min,
samples from 19 to 23, and Gaussian width from 3.2 min
to 3.4 min (4)
- Added section on data representation and precision (7)
and renumbered
- Updated references to Flux Budget specification version
(1, 9)
