Pointing model 18, feb-mar12

Links to plots:

The input data used to compute model 18 (.ps)  (.pdf)
The model 18 fit with residuals (.ps)  (.pdf)
Checking the model by removing a source at a time and recomputing the model (.ps)  (.pdf)
The azimuth encoder table results (.ps)  (.pdf)
Measuring the constant offset terms for the other receivers (.ps)  (.pdf)
Variogram of the raw errors and pointing residuals (.ps)  (.pdf)
verify model18 using sbn (.ps) (.pdf)

Links to sections:

Data used to compute the model.
Fitting the model.
Checking the validity of the model.
Azimuth encoder table.
Measuring the constant offsets for the "other" receivers.
Variogram of the pointing residuals.

Current status 05jun12 model 18A installed

Background.   (top)

Data used to compute the model.  (top)

    The data used to compute model 18 (.ps) (.pdf) was taken using model17A  (the previous model). Figures 1-5 show these errors. Figures 6 and 7 remove the model 17 correction and show the raw telescope pointing error. All errors are great circle arc seconds.
  1. Fig. 1 is the azimuth/zenith angle coverage for the input data.
  2. Fig. 2 is the pointing error (za error top, az error bottom) plotted versus azimuth. This is relative to model 17. The left half of each plot is the northern portion of the dish (southern sources with declination < 18.2 degrees). The right half of each plot is the southern portion of the dish (northern sources).
  3. Figure 3 is the pointing error (za error top, az error bottom) versus zenith angle for the input data relative to the previous model17. There is a linear ramp in za error vs za. (-1.6 + .28*za). It is about 4 asecs at za=19. Back in dec07 (when model17 was built) we had two more compressors than we do at the model18 time. This is about 500 lbs less wait on the dome for model18
  4. Figure 4 is the za and azimuth errors plotted by source order. The sources are color coded.
  5. Fig. 5 is the magnitude and direction  of these errors plotted versus azimuth and za. 1 tick mark is 5 arc seconds. At the bottom is a table of the average magnitude and rms for the entire dish and computed for every 5 degrees in za.
  6. Fig. 6 has the raw az, za errors plotted versus azimuth. The model 17 correction has been removed. Model 18 will be fit to this data set. Fits to 1az, 2az, and 3az have been over plotted with the amplitude and phase angle of the maximum. Part of the 1az term is probably a latitude error in the AO position.
  7. Fig. 7 shows  the same raw errors plotted versus za. T

Fitting the model.   (top)

    The model is fit to the raw errors. An encoder table spaced every .5 degrees in za is computed for azimuth and zenith angle errors and then removed. The final residuals are great circle errors.  The telescope must move in that direction from the computed position to point at the source.
     The model 18 fit with residuals (.ps) (.pdf) are:

za residuals az residuals total residuals [asecs]
mod18 noEncTable 5.80
mod 18 with Enc Table 2.13
  1. Fig. 1 plots the residuals versus za for the azimuth and za errors. The encoder table has not yet been removed.  The computed encoder table is over plotted in red.
  2. Fig. 2 plots the azimuth and za (raw Errors - ( model + encoderTable) ) residuals versus za.
  3. Fig. 3 plots the azimuth and za (raw Errors - (model + encoder table) residuals versus azimuth. There is more scatter in the azimuth residuals that the za. The tilt sensor measurements show a 6az term over part of the  dish. The encoder rack gear for the azimuth also has some runout. It will cause a azimuth scatter (with a za dependence since these are great circle errors).
  4. Fig. 4 plots the za and azimuth residual errors by source.
  5. Fig. 5 shows the za, az model residuals plotted versus source declination.
  6. Fig. 6 has the residual error plotted versus azimuth and zenith angle. 1 tick mark is 5 arc seconds. A table of the average error and the errors every 5 degrees za is at the bottom of the plot. Also included is the model parameters and values.

Checking the validity of the model.   (top)

    The validity of the model is tested by removing a source at a time from the data set and recomputing the model (.ps)  (.pdf). This was done for all 33  sources in the model.
  • Fig 1 has the model residuals removing one source at a time. 0 is J530+135, 1=J0745+101.. to 32=J1840+240. The black line is the total rms residuals while the red is the azimuth and the green is the zenith angle. The top plot does not include the encoder table while the bottom plot includes it. 
  • Removing the 15th source J0532+075 makes the largest improvement in the model. Looking at  figure 6  from the previous section it looks like the last 2 points before setting for this source had large errors. Maybe the tiedowns had lost tension?
  • Figure 2 plots the mean pointing error and its rms for each source track that was not included in the model. The model was evaluated without source i, then the mean and rms of the pointing model along the az,za track for source i was computed.
  • the 16.6 dec source J0521+166 had only one point (so there is not rms)
  • the 27.8 dec source (J0519+277) did have large az residual errors.

  • Azimuth encoder table.   (top)

        An azimuth encoder table for azimuth residuals was built by smoothing the great circle azimuth residuals in azimuth and then removing this from the (model-zaEncTbl) azimuth residuals. I first tried smoothing the little circle errors  (azErr/sinza) thinking that the azimuth encoder wrack gear was the largest culprit and it should give a little circle error. The residuals didn't get much better. The low za errors were messing up the averages. This must mean that the azimuth residual errors are great circle and not little circle.
        The table step has 1 degree steps in azimuth. Different az smoothing was tried. The az encoder table results (.ps) (.pdf) are shown in the figure: (the azimuth encoder table has not been installed).

    Measuring the constant offsets for the "other" receivers.   (top)

    Data sets used:

    sbn verify sources all nighte

    sbw,sb,lbw,cbh,lbw,sbn .. verify run using 07may sources

    ran with model17a but tietest..

    327,430,800,alfa,sbh verify using sbn sources.

    alfa- B0758+143 .. redo model verify from 15th

    The model includes constant terms (great circle) in azimuth and zenith angle for each receiver. These terms can differ receiver to receiver because of positioning error of the horn on the rotary floor. The model is made with sband narrow. After the model we need to compute what the constant offsets for the other receivers are (ideally it would remain constant).

        On 7,8,12,5,21 may12 5 sources  were tracked with sbn and the new model 18 installed.  The sources were:

    Some of these sources were then retracked with the other receivers to measure the receiver constant az,za values from the sbn model. Before each receiver was used, an offset was included to get the receiver close to where it was supposed to be.

        The plots show the tracking error for sbn and the other receivers. (.ps)  (.pdf).
    The first 11 plots show the sbn error and the  other receiver error (one per page). Black is the sbn measurement. Red, green, blue, purple are the up to 4 frequency bands of the "other" receiver. The left column has azimuth errors while the right column is za errors.  The numbers printed are the mean(sbnErr) - mean(rcvrErr) in arcseconds. The figures are:

    1. 327 MHz.  3 frequency bands,
    2. 430 MHz . 3 frequency bands ,
    3. 800 MHz . 1 frequency band, 2 sources.. The large za offset is because i put the 75 Arcsec za offset in backwards when updating the model file for the verify. I removed this error before actually computing the real offset.
    4. lbw. 4 frequencies, 2sources.
    5. sbw. 4 freq, 1 source
    6. sbh.  4 freq, 2 sources
    7. cband. 4 frequencies, 2 sources.
    8. cbandHi. , 4 freq, 2 sources
    9. xband. 4 frequencies,2 sources
    10. alfa 2 sources, bm 0. the Za errors for B0758+143 varied between the first and 2nd day of this source.
    11. Mean(sbn)-mean(rcv) for each source and frequency band. black is the azimuth error and red is the za error.
        Ideally the (sbn-rcv) value should be the same for all sources and frequencies of a receiver.
        The offsets for the individual receivers as calculated from the above data is shown in the table below.
    model 18 receiver offsets.
    rcvr azOffset asecs za offsets asecs
    sbn -34.07 -101.64
    327 -95.53
    430 -4.23
    lbw -35.93
    sbw -27.91
    sbh -36.07
    cb -34.90
    cbh -34.20
    xb -35.48

    processing: x101/model/feb12/verify.pro

    Variogram of the pointing residuals.   (top)

        A variogram of the raw errors and pointing residuals (.ps)  (.pdf) shows the correlation of the measurements versus separation of the points. The residual error and raw pointing error difference is computed for all points on a pair wise basis. A metric is then defined for the point separation and is used to bin the data. The variance of the pair differences for each bin is then computed and plotted versus the distance. For each figure the top plot is the pair wise difference of the pointing residuals (including the zaencoder table) while the bottom plot has the pair wise difference of the raw errors input to make the model.
        This data can be used to interpolate the residuals onto an az,za grid (it gives the nugget (y intercept), range (where the variance increases), and  the sill (value where the variance levels off) for the krigging routine)
    1. Fig. 1 is the variogram using the great circle angular separation of the points as the metric.
      1. The separation was binned to .3 degrees steps.
      2. The za correlation increases until za=2. degrees  and then levels off
      3. The az residuals variance increases till about 5 degrees. The 25 foot spacing of the north south main cables is about 1.6 degrees (1.5 degrees is close to the 25 foot spacing of the main cables ). The large correlation in the bottom plot is the 1az term of the raw pointing errors.

      Fig. 2 projects the points into the xy plane and then measures the distance (since the kriging would be done in this plane). It looks the same as that of figure 1.
    processing: x101/model/feb12/doall.pro