Targeted Test Observations of 2005 August 13

The test project (to002) was observed on 2005 August 13. It was designed to demonstrate the effects of EOP errors and to verify that we are able to correct them. One pair and one triplet of calibrators with well known positions were observed, with about 40 minutes on the pair alternating with about 62 minutes on the triplet. Three times during the observation, about 45 minutes were spent looking at calibrators all over the sky. Unfortunately one of the triplet was too week to be useful for purposes of the test. The pair was 1215+303 and 1213+350, which are separated by 4.7 degrees. The surviving pair from the triplet was 1111+149 and 1116+128, which are separated by 2.4 degrees. For the processing shown here, 1213+350 and 1116+128 were treated as phase calibrators. Less effort was put into high quality phase referencing than would be likely for a science observation. Specifically, there was no attempt to image the calibrators or to do careful editing. The weather was not great for these observations which degraded the phase referencing for the later parts of the observation. The examples shown here are from early in the observation in order to emphasize EOP effects rather than weather effects. Also, we have not yet tried an atmospheric calibration based on the segments with many calibrators. The observations were made in dual S/X band mode (13 and 4 cm bands) with a geodetic style broad frequency spread.

The test observation was correlated twice. One pass was done with the rapid service EOP that we would normally use when the EOP bug was not active. The other pass used hand altered EOP parameters that differed from the good values by about 3 ms in UT1-UTC, 2.8 mas in X and 6.4 mas in Y for the times nearest the observations. The 5 day sequence of altered EOP were contrived to have a jump in the offset from the best values in an attempt to trigger an interpolation problem if there was one. The values used are tabulated in Table 1.

Figure 3: Plots of phases from a 4 cm IF from the test observation for baselines involving Mauna Kea, Kitt Peak, and St. Croix -- a set of long baselines. These are from the processing pass in which poor EOP values were used. The left plot shows phases calibrated using a CL table copied from the data set processed using good EOP. That CL table includes results from fringe fitting so, when used on the the good EOP data set, it causes the phases to be near zero. When applied to the poor EOP data set, the residual phases will be only those resulting from the EOP errors. The right plot is the result of applying CLCOR to the CL table used in the left plot. It closely reproduces the phases in the good EOP data set calibrated with uncorrected CL table.

An impression of the magnitude of the phase offsets introduced by the poor EOP can be seen in Figure 3. The two test data sets only differed in the EOP used in processing, so calibration tables used for one can also be used for the other. The data set processed with good EOP was fully calibrated, including fringe fitted, to the point where the residual phases of the calibrators are all near zero. The CL table with that calibration was copied to the data set correlated with poor EOP and applied in making the plot on the left. The plot shows the phases on 3 long baselines. There is rapid phase winding, due entirely to the EOP offsets. The plot on the right is made with a CL table produced by using CLCOR to add corrections for the EOP offset to the CL table used in the left plot. The phases are now flat and near zero so the correction has been very effective. This also demonstrates that essentially all of the phase offsets in the left figure are actually due to EOP offsets.

Figure 4: Phase referenced images of 1215+303 at 13 cm from the test observation. The left image was made from data processed with good EOP. The right image was made with EOP offset by amounts typical of what was seen while the EOP bug was in effect. There is a clear position shift, but at this relatively low frequency, not much change in image quality.

Figure 5: Phase referenced images of 1215+303 at 4 cm from the test observation. The left image was made from data processed with good EOP. The right image was made with EOP offset by amounts typical of what was seen while the EOP bug was in effect. The position offset seen as 2.3 GHz is seen. At this higher frequency, the image quality is also somewhat degraded

Figure 4 shows the 13 cm phase referenced images of 1215+303 from the good EOP correlation pass (left) and from the poor EOP pass (right). For this low frequency, where the rotations caused by the poor EOP are not much larger than the beam, the image quality is not much affected. But notice the shift of about 2 mas in position. Figure 5 shows a similar pair of images of the same source, but at 4 cm. Now there is a noticable degradation of the image quality in addition to the position shift.

Figure 6: Phase referenced images of 1116+128 at 13 cm from the test observation. The left image was made from data processed with good EOP. The right image was made with EOP offset by amounts typical of what was seen while the EOP bug was in effect. The offset between calibrator and target is smaller than for the source in Figure 4 so the image shift is smaller. Again there is little change of image quality at this frequency.

Figure 7: Phase referenced images of 1116+128 at X band (4 cm) from the test observation. The left image was made from data processed with good EOP. The right image was made with EOP offset by amounts typical of what was seen while the EOP bug was in effect. A position shift is seen and degradation of the image quality is pronounced.

Figure 6 shows the 13 cm phase referenced images of 1116+128. Figure 7 shows the 4 cm images of the same source. Again the left images are made from data with good EOP while the right image is made using poor EOP. As with the other source, the effect at 13 cm was primarily a position shift. But at 4cm, despite the smaller calibrator/target separation than the other pair, the phase referencing for this pair did not seem to work as well. The position offset scales about with the separation (about 1 mas in this case with about half the separation.), but the image quality was more affected by the poor EOP.

Note that these tests show a significant impact on image quality for observations at 4 cm. Phase referencing observations at higher frequencies is common and will be even more affected, although they typically use calibrators that are closer to the target.

Figure 8: Difference between the CL table geodelay values from the processing pass done with good EOP and the geodelay plus the CLCOR correction for the pass done with poor EOP. The left plot is for Kitt Peak and the right plot is for Mauna Kea. If the CLCOR corrections were perfect, the difference would be zero. Note that 1 ps at 8.4 GHz is 0.0084 turn of phase or about 3 degrees of phase, so the differences are small.

CLCOR was run on the data set correlated with poor EOP to test the CLCOR corrections. The usno_finals.erp file used for the corrections was edited so that, over the days of interest, the values matched those used for the ``good'' EOP correlation pass. The most direct comparison was made by differencing the total delays from the good EOP pass and the corrected poor EOP pass. The total delay here is the geodelay column in the CL table plus the delay column for a CL table that has only had CLCOR affect the delay (no FRING etc). Figure 8 shows the results of that comparison for the station delay for 2 representative stations, Kitt Peak and Mauna Kea. What will actually matter is the difference between these station delays. The differences, after some debugging, are now down to the level of about 1.4 ps max in the test data. Note that 1 ps is equivalent to about 3 degrees of phase at 8.4 GHz. One slight concern is that the difference is increasing with time through the observation as is the scatter between sources during the 3 time periods (beginning, 0.9-0.93 day, and end) when sources scattered around the sky were observed. That effect is not yet understood. It is not clear how a long observation would be affected. A project is in progress to correlate a 24 hour run twice with different EOP to explore this effect, but the results are not ready for this memo.

Another test was done by comparing the final calibrated phases for reference and target sources for data correlated with good EOP and data correlated with poor EOP and corrected with CLCOR. For this test, the data were calibrated and fringe fitted separately on each dataset (A CL table was not copied from one to the other as in the test described earlier). The differences are a few degrees and of variable sign so they seem to be within the fluctuations due to the different correlation passes and different fringe fits. Without the CLCOR corrections, the calibrated target source phases are different by significant fractions of a turn. Note that, with this type of processing, the phase calibrator phases should be the same, as seen, because of the post-correction fringe fit.

An image equivalent to that of Figure 5, made with poor EOP data corrected by CLCOR looks identical to that in the left panel of Figure 5, the one made with good EOP. A simple fitted position for the peak of the source gives a position different from the one made with good EOP by 27 as, which is probably limited by noise. The offset of the peak in the image made with poor EOP is 2.4 mas, or nearly 100 times worse. The image off-source noise, from an histogram fit, for the corrected data is 6% higher than for the good EOP data set, but, since no effort has been made at editing or other cleanup, that may just be differences in the processing. It is unlikely to be due to the EOP correction.

The conclusion of the tests so far is that the corrections available with CLCOR are good to the levels required by most users. Users with extreme requirements (order 10 as is possible with care), should consider redoing the model if possible if they have data processed with poor EOP.