[daip] FRING

[Z.-Q. Shen]

Hi Jim,

Thanks for your diagnostic message. It is very helpful.

After some testing runs, I found the setting of APARM(4)=1 helps.
In this case, I used APARM(5)=1 to solve for multi-band delay over
8 IFs. This would increase the SNR for each spectral data point, right?
But I wonder if we can solve for single-band delay (i.e., APARM(5)=0 or 2)
with APARM(4)=1. I did try APARM(5)=2 and found more solutions
compared to APARM(5)=1 solutions, but these solutions seem to be not
good with quite large scattering. I did not try APARM(5)=0 because of very
limited SNR.

This is VLBA only observations at 3mm. Antenna 7 is OV. I suppose
that high SNR is related to the noise spike. The observing log said that
OV GPS Delta was drifting at - 0.1 us per hour. You also mentioned in
the VLBA Scientific Memo No. 25 that a common cause is when OV is
looking at the White Mountains. I wonder where is White Mountains
to the OV station.

I have another question regarding running FRING. This is about DPARM(4).
It is said that we should use DAPRM(4) to specify the correlator integration
time, which is 1 sec in my data. But when I accidentally set it to 2 sec,
I did not see any difference compared to the results from FRING with the
correct DPARM(4)=1. My question is how important is the DPARM(4) in
FRING.

Thank you and a happy new year to you.

Eric

At 21:35 2003-12-23, Jim Ulvestad wrote:
>There are several reasons for this message.  Often, there is a
>single bad antenna that gives a poor starting point for the
>solution, so that chi-squared is unable to get to a minimum
>from its starting point.  I have found that this often happens
>with Saint Croix, because the integration time is a bit
>too long and the antenna loses coherence.  The snippet of
>results you sent indicated that antenna 7 has much higher
>SNR than the other antennas--is 7 the VLA or GBT?  If
>so, you might try using 7 as the reference antenna.
>
>This result can happen when you have low signal/noise,
>so that an antenna starts off at a noise spike rather than
>a true signal in the FFT.  Try increasing the SNR
>by setting the parameters to average between IFs
>and polarizations.  This is done by setting APARM(3)
>and APARM(4) = 1.
>
>Take a look at your SNPLT from the original FRING
>run in order to find places where the solution did not
>converge.  Then, run FRING on only that small piece
>of data, and turn up the print level to APARM(6)=1.
>This may show you a particular antenna that has very
>low SNR (either resolution or poor coherence) or
>a bad starting point because a noise spike is being found
>out of the FFT.  You may then try running FRING without
>that antenna and see if convergence is better.  If so,
>you might want to flag the antenna.
>
>Finally, you can set the window for the search window
>to be fairly narrow by using DPARM(2) and DPARM(3).
>If you already have set the clocks using the phase cals
>or the FRING on a strong source.  If so, you can set a
>small delay window; I would recommend DPARM(2)
>of 60 to 100 (60 nsec to 100 nsec total width, meaning
>+/- 30 to 50 nsec).  DPARM(3) also can be set to
>restrict the window in fringe rate.  The value you would
>use here is frequency-dependent, since the fringe rate
>in milliHz is proportional to the observing frequency.
>The fringe-rate error for VLBA antennas tends to
>be atmosphere-dominated.  For 5 GHz, I'd guess
>that DPARM(3)=20 or 30 is usually adequate.
>But for 43 GHz, you may want to use 100 or more.
>
>Finally, if you use low values of DPARM(2) and DPARM(3)
>as I mention, you may want to reduce the SNR
>threshold using APARM(7)---try setting it to 4.
>It also may be that SOLINT is too long for the
>atmospheric coherence, but you may be able to
>shorten it some if you increase the SNR by using
>APARM(3) and APARM(4) to average data.
>
>Bottom line is that it appears to me that you have
>fairly low SNR on at least some baselines, and
>that this gives you some bad starting points in your
>solutions.  Therefore, your goal should be to
>identify which antennas have the poorest SNR,
>to increase the SNR as possible or flag the low
>SNR data, and narrow the initial search range down
>to focus on where you really expect the fringes
>to be.
>
>A final thought--it may be that the source you are
>running FRING on is fairly resolved.  Making an
>initial map and self-calibration, then feeding this
>image back into FRING as a source model (using
>IN2NAME and NCOMP) might help things
>converge better.
>
>Best,
>
>Jim
>
>Shen Zhiqiang wrote:
>
>>Hi,
>>
>>In running FRING, I saw many pieces of message as below
>>
>>FRING      Time=   0/ 20 58  4, Polarization =   2
>>FRING     B= 01 - 04 IF=  1 R=     -22.5 D=      -5.9 SNR=  15.5
>>FRING     B= 03 - 04 IF=  1 R=     -21.5 D=     -19.5 SNR=  10.4
>>FRING     B= 04 - 07 IF=  1 R=     -22.5 D=     -16.6 SNR=171.1
>>FRING     B= 04 - 08 IF=  1 R=     -22.5 D=     -16.6 SNR=  10.6
>>FRING     FIT DID NOT CONVERGE FOR IF    1
>>FRING      This probably means that the starting value for the
>>FRING      delay or rate for one or more antennae is bad.  You
>>FRING      may want to set search windows and try again.
>>
>>Obviously, this would lose some good detections. Is there any
>>way to avoid this? What does it mean to set search windows?
>>Are the windows used too small?
>>Again, this happened many times in a single run of FRING.
>>Best regards,
>>Zhiqiang Shen
>>
>>_______________________________________________
>>Daip mailing list
>>Daip at listmgr.cv.nrao.edu
>>http://listmgr.cv.nrao.edu/mailman/listinfo/daip
>>