Single pol/brd spectral line observing with wapps only provides 1 pol
Jan 05
Problem:
The spectral line gui allows you to setup the wapps
for 1 polarization per board. You can select polA or polB for the first
board and the opposite pol for the 2nd board. This allows for more channels
per spectra (but uses up more boards). An example would be:
-
Brd 1: 3 level auto correlation, polA. Assume 100 Mhz
bandwidth. This gives 8192 channels/spectra (there are 16K lags per board,
but you end up with only 8192 because 100 Mhz needs to do interleaved sampling).
-
Brd 2: 3 level auto correlation, polB. Assume 100 Mhz bandwidth.
This gives 8192 channels/spectra.
The problem is that the gui configures the wapps to provide polA to both
boards (actually it could be polB, i'm not really sure... the important
thing is that you only get 1 polarization). This problem has been there
since at least nov04 and probably all the way back to may04. This problem
does not occure when the gui is used in pulsar mode.
People who have been observing with 1 polarization
per board have been losing sqrt(2) in sensitivity since they are not getting
both polarization's recorded. The plots
show an example of what is happening.:
-
Fig1 top: These spectra are 1 polarization per board.
Black is polA (brd 1) and red is polB (brd 2)
-
Fig1 bottom: These spectra have 2 polarization's per
board. Black is polA (brd 1) and red is polB (brd 2).
-
Fig2 Top: 1 polarization per board. The spectra (1
second integration) has been normalized by the scan average (60 seconds
long) to remove the bandpass in polA and polB separately. You can see that
the noise in polA and polB are correlated (the lbw receiver was setup for
linear polarization.. the hybrid was out). This is incorrect since the
noise in polA and polB should not be correlated.
-
Fig2 middle: 2 polarization's per board. The spectra have
been normalized by the scan average. The two traces no longer look correlated.
This is what is expected.
-
Fig2 bottom: These two plots measure the correlation of polAxpolB
for the two methods of taking data (1 pol per board black, 2 pol per board
red). 60 1 second records were used for the correlation. Let spc(Ci,Tj)
be the data for a single spectra where Ci are the 2000 channels and Tj
are the 60 1 second records. The processing was (let Ci stand for channel
i and Tj for time sample j):
-
Remove the mean by channel and normalize to the rms of each channel.
-
spcAvg(Ci)= sumJ(spcCi,Tj) .. compute robust average for each channel
-
spcRms(Ci)=rmsOverTj(spc(Ci,Tj)) .. Compute the rms for each channel
-
spcN[Ci,Tj]= (spc(Ci,Tj) - spcAvg(Ci))/(spcRms(Ci) .. subtract mean,
normalize to rms
-
Compute the correlation of polA vs polB using the normalized spcN:
-
cor(Ci) = sumJ(spcPolAN(Ci,Tj)*spcPolBN(Ci,Tj))/ sum(j)
The correlation polAxpolB for the single pol/brd is
close to one the correlation for the 2 pol/board is centered on 0.
This problem was reported by chris salter in nov04.
I reported a similar problem on 20may04 and again on 31jul04 stating that
the single pol per board header values were incorrect (i mistakenly thought
that the data was ok, just the headers were bad).
The gui continued to allow users to select polA for
board 1 and polB for board 2. People have been observing in this mode and
have been losing sqrt(2) in sensitivity as well as quoting the wrong sensitivity
levels. When we find things like that that may need some time to correct,
we should disable the feature in the gui until it can be fixed. This would
prevent users from inadvertently using broken options.
This problem is scheduled to be fixed on 3feb05.
processing:x101/050131/singlechan_jan05.pro
home_~phil