Pointing Offsets and Tracking

RA and DEC Offsets

Since April 2013 we record pointing offsets as reported by the tim command of the kentools package in the telemetry database. The data can be found as ra_offset and dec_offset in the exposure table. Note that ra_offset is given in linear arcsec on the sky; you need to divide by cos(dec) to get the total ra offset.

Analysis of the April, May and June Regression Data (Also see the page on the analysis of the June and July regression data)

(work in progress - more text and discussion will be added. The studies are done by Lucas Beaufore (OSU))

April exposures are from exposure id ranges : 200826 - 201063 and 201254 - 201454
May exposures are from exposure id ranges : 206150 - 206210
June exposures are from exposure id ranges : 215724 - 215923

Note: The first set of April exposures, as well as any exposures with incomplete data, were not included some of the analyses below.

Average Pointing offsets for April - June:

These are the values in arcsec averaged over all exposures taken that day (calendar day, not mjd)

When comparing hexapod setting between the three engineering periods one of the biggest differences is the temperature. Since we had a problem with one of the upper truss sensors in June the plot below is based on the domelow temperature. All this shows is that it gets colder in June. The second plot shows focus as function of temperature. The 110 micron per degree relation is seen as expected. Note that the temperature look up table is disabled and that focus is controlled by the AOS. The AOS does not use a hardcoded temperature relation. The data in the plot is for all filters. The offsets are not removed which explains some of the jitter.

Focus vs Temperature (Domelow since UPTRTEMP was broken)

More entries are present in the April sample (red) because the regression script was run twice that month.
The temperature dependence of the focus is obvious and the following plots of hexapod z vs. temperature shows the expected 110 micron/degree relation. The second plot shows that the AOS system via the trim correction drives the focus adjustments. Note that the temperature LUT is not in use when the AOS is operational (which is our default since last year)

Hexapod Position vs HA and DEC

A standard regression script is run every month during the engineering time. We have performed an initial study using these regression images for April, May and June. In this section we show the how the hexapod settings (x, y, z, tip and tilt) vary from month to month. A strong temperature dependence is expected for focus (hexapod z) where the other 4 coordinates should be more or less the same every month. The following plots show that this is not the case.
In all plots, red markers indicate data from April, blue is May and green marks data taken in June. The first plot shows a spherical view of the coordinate grid.

Coordinate Grid for Regression Tests

Hexapod coordinates as function of HA for the three engineering runs:

HA vs Hexapod x HA vs Hexapod y HA vs Hexapod z
HA vs Hexapod tip HA vs Hexapod tilt

Hexapod coordinates as function of declination for the three engineering runs:

DEC vs Hexapod x DEC vs Hexapod y DEC vs Hexapod z
DEC vs Hexapod tip DEC vs Hexapod tilt

Not surprisingly, the hexapod parameters show some dependency on the telescope position. There is reasonably good agreement among the three engineering runs. The next set of plots shows how the pointing offsets depend on the telescope position. No difference is expected between the three time periods - except for focus with its strong temperature dependence. [Aaron, all 5 hexapod coordinates must have some temperature dependence. How strong is that for x, y, tip and tilt?]

RA offsets as function of Hexapod coordinates for the three engineering runs:

RA offsets vs Hexapod x RA offsets vs Hexapod y RA offsets vs Hexapod z
RA offsets vs Hexapod tip RA offsets vs Hexapod tilt

DEC offsets as function of Hexapod coordinates for the three engineering runs:

DEC offsets vs Hexapod x DEC offsets vs Hexapod y DEC offsets vs Hexapod z
DEC offsets vs Hexapod tip DEC offsets vs Hexapod tilt

We observe a significant difference for the data from the June run. There appears to be some trend in the measured offsets as function of the hexapod position.

Here are two plots showing the pointing offsets as function of wind speed:

RA offsets vs Wind Speed DEC offsets vs Wind Speed

We see the same differences between the June and the April/May data. The wind speed doesn't appear to have a significant effect on the pointing offsets.

Hexapod Position Repeatability

We studied the repeatability of the hexapod position by comparing exposures taken each month at the same telescope position.
The following histograms show the difference in each of the hexapod coordinates (except z) for these exposures. The regression script takes multiple exposures at any given position. During this sequence the AOS system is allowed to adjust the hexapod position. In an attempt to exclude this effect we take the position difference only between exposures from the same position in the sequence - for example, first exposure in May minus first exposure in June, third exposure in April minus third exposure in June etc. The first set of plots shows data for the hexapod position followed by two plots showing the difference in telescope_lut x and y positions (these reflect the static component of the hexapod position adjustment). The last plot shows the difference in x and y values of the trim correction which is accumulated from the tweaks sent by the AOS.

Difference in hexapod coordinates

Regression test: Variation in hexapod x (same position) Regression test: Variation in hexapod y (same position)
Regression test: Variation in hexapod tip (same position) Regression test: Variation in hexapod tilt (same position)

Difference in hexapod coordinates

Regression test: Variation in telescope LUT x (same position) Regression test: Variation in telescope LUT y (same position)

Difference in hexapod trim values

Regression test: Variation in trim x (same position) Regression test: Variation in trim y (same position)

Bcam Value Repeatability

The following histograms show the difference in bcam values by comparing exposures taken each month at the same telescope position.
This is calculated the same way the hexapod position differences are (see Hexapod Position Repeatability above). The first two plots show the diffence in
bcam dx and dy values (correspoding to hexapod x and y values), the second two show the diffence in bcam ax and ay values (corresponding to hexapod tip and tilt values).

Hexapod Position vs Bcam Values

The following plots are composed of the June engineering regression exposures. They compare the hexapod positions against the corresponding Bcam values.

Hexapod components vs Offsets

The following sets of graphs show the same subset of exposures from the June engineering regressions (215724-215751).

These plots show the relations between the hexapod and the offsets. They show that the exposures are in groups of 4, and within these groups
the offsets correlate with hexapod position. Specifically, hexapod_x and trim_x have a negative correlation with ra_offset, and hexapod_y and
trim_y have a negative correlation with dec_offset.
What we are seeing raises questions about how quickly the alignment system converges. There is now some evidence that this doesn't happen at all during the regression script (with the large slews every 4 exposures this is of course not the same as DES observing.
Color coding: Blue - Position from the Telescope look up table; Green - Final hexapod position used for this exposure; Red - Trim value for this exposure. The trim and tweak values have been corrected for the database/telemetry timing problem.

Hexapod_x, Telescope_LUT_x and trim_dx vs exposure Id (effectively time) on the left and ra offsets as measured by kentools vs hexapod_x on the right. The 4 exposure block structure used by the regression script is clearly visible.

Hexapod_y, Telescope_LUT_y and trim_dy vs exposure Id (effectively time) on the left and dec offsets as measured by kentools vs hexapod_y on the right.
This is from the beginning of the script. We see that for the first 12 exposures the AOS is chasing a good y value for the hexapod. After each image the AOS tweak (== the step in the red triangles) is around 150 micron which I think Kevin told me is the maximum they allow. Then the trims level off and there is much less of change for the next 15 exposures or so.
Note that the telescope continues to change the position every 4 exposures. The plot doesn't show the the coordinates but you can see the steps in
the LUT values. They are of similar size for the first half ("going up") and the second half ("going down")

The 12 exposures on the right plot on the diagonal from the upper left corner to the lower right corner correspond to the first half of the exposures shown in the plot on the left. The cluster in the lower right corner corresponds to the second half of the exposures shown in the plot on the left. As the AOS is chasing the best y value the dec offsets change proportionally. Once the AOS tweaks become small and the trim is stable the dec offsets become stable.

RA/DEC offset plots vs hexapod x/y position for the entire regression script data sample (June)

The same study but on the plots on the right (and the plots in the 3rd row) show the offsets as function of trim instead of hexapod position.
Again, the first dozen or so exposures correspond to the entries on the downward diagonal in the dec_offset vs trim_dy plot.

Hexapod tweak variation for same position

To test how well the alignment converges we took advantage of the fact that the regression script takes 4 exposures in the same position. If the alignment shows any sign of convergence from the first to the last exposure in each block. Since the AOS corrections are one exposure behind, we plot the tweaks recorded for the second, third, forth and the first exposure of the next block.

Hexapod tweak_dx:

Hexapod tweak_dy:

The first plot in the tweak_dy series indicates large corrections (the two peaks at +/- 150 micron) which is not too surprising after a slew but the width of the distribution in plots 2, 3 and 4 doesn't seem to decrease a whole lot.
Aaron, could you remind us about the expected resolution for the donut tweaks in x and y? Could we be at the noise limit? (we could check if the tweaks for exposures 2,3,4 in each blocks are random or monotonically increasing/decreasing)

Hexapod tweak_xt (tip):

Hexapod tweak_yt (tilt):

Hexapod tweak variation for all 2013B-001 Exposures

Hexapod tweak_dx:

Analysis of offsets from pointing model

Analysis can be found here

Further studies
  • Change in position as measured by the offsets for consecutive exposures taken at the same telescope position
  • Compare the AOS corrections/hexapod trims for the 3 engineering runs