NRAO Home  >  Green Bank  |  Wiki Topic:    GB > PTCS > FocusPerformanceNotesMay07

Changes | Index | Search | Statistics | Go

Focus Performance Notes -- May 2007

Introduction

The subreflector focus performance seems to have become degraded in recent months, as described by Kim in a message of April 19th in which he notes that the quality of the fit of the focus model to the data is somewhat degraded in the data after August 17, 2006. The earlier data had 68th percentile residuals to the fit of about 1.2 mm, while the later data had 6.6mm. He also noted a large residual of -20 to -25 mm to the best model fit for the data of TPTCSPNT_070319.

The more recent data has been processed with the new python-based "prepoint", whereas the earlier data were processed with glish code "dcr.g". Is the new software to blame for the apparent worsening of focus performance, or is there a real problem with the focus model or dynamic corrections?

These notes summarize investigations of this situation.

Basic subreflector control

The Gregorian subreflector is commanded in an (X,Y,Z) coordinate system defined by the servo system, in which "Y" is toward and away from the main reflector along its axis of symmetry. We will be concerned here only with "Y". The servo (X,Y,Z) are slightly off from the ideal optical axes (see GBT Commissioning Memo 11), but not enough to concern us here. There is a focus tracking model, determined by astronomical measurements, of the form:

Y_model = YDA + YDB cos(EL) + YDC sin(EL)

There is a temperature-dependent dynamical model, also determined empirically, Y_dyn, which is a fuction of combinations of structural temperatures.

The model coefficients are stored in the Antenna FITS file main header for each scan.

There is also a "Local Focus Correction", Y_lfc, which is determined as a result of doing a focus scan on a calibrator.

The Antenna control software commands the subreflector to a Y coordinate given by:

Y_command = Y_model + Y_dyn + Y_lfc

---

Differences : glish dcr.g vs python prepoint

When processing focus scans and writing into the "ast" database files, Dana's glish code, "dcr.g", differs in a few respects from the python "prepoint" program.

	dcr.g	prepoint
baseline fitting	3rd order polynomial fit to area outside 50% of scan length centered on the peak	3rd order polynomial fit to region outside 1.33FWHM from location of peak*
peak fitting	gaussian fit to 50% of scan length centered on peak	gaussian fit to region within 1 FWHM of location of peak

*note: if there is not enough baseline, the average over the scan is subtracted (i.e, the average value is the baseline.)

In the GFM routine used by prepoint, the fitting is done in the coordinate with respect to the Y_model; hence the result is relative to the model. prepoint has a "-d" option, which, if specified, will add the Y-model before writing to the "ast" database file. This is the normal state of affairs, so the "Focus Offset" in the "ast" file is the absolute "Y" coordinate. If one doesn't use the "-d" option, then the result is relative to the Y_model.

In dcr.g, the fitting is done in the absolute Y coordinate system, and the Focus Offset results are also in this system; no Y-model is added or subtracted.

Ron Grider's analysis of the glish dcr.g code shows at least one error, and perhaps one questionable implementation decision that may serve to explain some of the dissimilarities between these results and PrePoint processed data. The issue centers around the possible misapplication of focus tracking model parameters from an earlier PFM; namely, Model3d. See Glish ReduceFocus Issues.

Some Comparisons

Here we list differences between the glish and prepoint processing. Note that the glish "dcr.g" used in these tests was the original form used by Dana, not the version that was modified to match the GFM processing. Differences are millimeters in the sense "prepoint minus glish".

Project	N	type of data	mean diff	stddev diff	max diff	min diff	histograms
060818	84	LCP, focus value	-0.40	0.53	0.72	-2.12	histogram (both LCP,RCP)
		RCP, focus value	-0.33	0.62	1.01	-2.20
	84	LCP, focus error	0.013	0.020	0.073	-0.054
		RCP, focus error	0.010	0.020	0.111	-0.040
061010	81	LCP, focus value	-0.26	0.56	2.29	-1.70	histogram (both LCP,RCP)
		RCP, focus value	-0.16	0.63	3.05	-1.69
	81	LCP, focus error	0.0001	0.0033	0.011	-0.016
		RCP, focus error	0.0005	0.0029	0.009	-0.013
070319	91	LCP, focus value	-0.99	0.87	0.89	-5.88	histogram (both LCP,RCP)
		RCP, focus value	-0.93	0.76	1.25	-2.90
	91	LCP, focus error	0.030	0.030	0.116	-0.128
		RCP, focus error	0.035	0.025	0.121	-0.031

---

Focus Measurements compared with model

The focus tracking model, without dynamic corrections or LFCs, is plotted along with the measured focus offsets for several recent pointing runs. These plots are based on the "prepoint" processing. In all cases, the upper plot shows the focus offsets versus elevation, with the model plotted as the solid curve. The lower plot shows the residuals to the model; the mean residual and its standard deviation are on the plot. The different plot symbols are for the two polarizations ( LCP: + ; RCP: * ).

Comparisons with Focus Tracking Model

Project	type of run	mean residual	stddev resid	plots
060818	single-sources	18.45 mm	15.18 mm	plot
061010	all-sky	3.46 mm	1.87 mm	plot
070319	all-sky	-25.25 mm	1.60 mm	plot
070330	single sources	-10.64 mm	0.73 mm	plot
070410	all-sky	-4.97 mm	1.41 mm	plot

Comments

First it should be noted that although there are small systematic differences between the prepoint and glish processing, the plots of focus offsets versus elevation look virtually identical. The results for the residuals in the last table are virtually the same regardless of the type of processing.

The weirdest looking plot is the one for project 060818. This session was single-source tracks on two pairs of sources, the first (0119+3210 and 0126+3313) are the points that follow the tracking model fairly well, although with larger scatter at lower elevations; the 2nd pair (1130-1449 and 1139-1350) are responsible for the deviant points near a focus offset of +40mm in the elevation range 25-40 degrees. This warrants further study.

The other plots show the data following the model fairly well, but with a constant shift, which is quite large in the cases of 070319 and 070330. The scatter about the mean is fairly small, about 1-1.5mm, which suggests that there is no short-term wonkiness in the subreflector control. The next step is to see what the dynamic corrections were for these runs.

Comparisons with Dynamic Focus Correction plus Tracking Model

In this set of plots, the solid line is the focus tracking model, and the green diamonds show the tracking model plus dynamic corrections. The lower of each pair of plots shows the residuals of the focus measurements with respect to the model plus dynamic correction.

Project	type of run	mean residual	stddev resid	plots
060818	single-sources	18.88 mm	11.16 mm	plot
061010	all-sky	12.85 mm	1.38 mm	plot
070319	all-sky	-16.81 mm	1.84 mm	plot
070330	single sources	3.37 mm	1.26 mm	plot
070410	all-sky	3.47 mm	1.21 mm	plot

More Comments

The use of dynamic corrections appears to reduce the large constant shift in some cases (070410, 070330, 070319), but actually made things worse for 061010. The scatter about the model has not changed much. The dynamic corrections apparently do not always make an improvement.

-- FrankGhigo - 02 May 2007

Attachment:	Action:	Size:	Date:	Who:	Comment:
kimc.19apr2007	action	1509	11 May 2007 - 21:13	ToddHunter	email from Kim 19Apr2007 re: New focus data
image001.jpg	action	33096	11 May 2007 - 21:14	ToddHunter	email attachment: image001
image002.jpg	action	33621	11 May 2007 - 21:14	ToddHunter	email attachment: image002
image003.jpg	action	67412	11 May 2007 - 21:15	ToddHunter	email attachment: image003
image004.jpg	action	65773	11 May 2007 - 21:15	ToddHunter	email attachment: image004