test 16 bit sampling full stokes (Galfacts setup)
Spectral averages (1, 12 seconds and rms by channel)
Saturation in beam1 when 1290 radar points at us.
On 17jun08 full stokes data was taken
with the Galfacts setup to test the system. The setup was:
- Alfa receiver with RF center frequency=1375, two 172 MHz
at 1300 and 1450 MHz.
- 172 MHz sampling, 4096 channels, dump time= 1 millisecond
- pshift =0xffa
- ashift=polA,B: 14, stokesU,V: 15
- packing=16 bits
- telescope sitting at az 270, za=10.
- Before taking data the A/D rms was adjusted to be 30 counts.
- 12 seconds of data was taken to give 1 complete radar rotation.
- Radars present in the data:
- aerostat 1242, 1256
- remy radar: 1290
- faa radar: 1350
- Punta salinas radar modeA: 1232/1247 and 1241/1256.
Spectral averages (1 and 12 seconds) and rms
by channel for 7
Three pages of plots were made for each beam:
- Page 1 and 2 : 1300 MHz band spectral bandpasses and 1450 MHz
bandpasses. For each of these bands:
- Average the 1 millisecond spectra over the 12 seconds. This
gives the mean number of counts for the 1 millisecond spectra versus
the bandpass shape.
- Plot 1: polA (black) and polB is (red)
- Plot 3: stokes U (black), stokes V (red)
- Average the 1 millisecond to 1 second. plot the 12 1 second
spectra with offsets to see how things were changing with time.
- Plot 2: polA (black) and polB is (red)
- Plot 4: stokes U (black), stokes V (red)
- Page 3: The rms by frequency channel for the first 1.2 seconds of
- PolA (black), PolB (red), StokesU (green), StokesV (blue).
- Plots 1 and 2. Rms vs frequency in counts. plot 1 is 1300 Band,
plot 2 if 1450 MHz band.
- When packing the spectra, polA,B are upshifted by 14 while
stokes U,V are upshifted by 15. This is to make the rms of polA,B
and stokesU/V more equal. Jeff is computing 2*polA,2*polB, and
stokesU,V. The extra upshift of stokesU,V brings them closer together.
- Plots 3 and 4. Rms/Mean vs frequency.
- For polA and polB i divided the measured rms by the 12
second averaged bandpass.
- For stokes U,V i divided the measured rms by
sqrt(avgpolA*avgPolB). After doing this the rms/mean was sqrt(2) too
small so i multiplied the result by sqrt(2) (i'll go looking for
the sqrt(2) factor someday...).
Saturation in Beam 1 when 1290 radar
points at Ao.
The beam1 plots above showed that polB had lots of
spectral artifacts during the 3rd second of data. I used about .6
seconds of 1 millisecond spectra about this occurence to take a closer
- Dynamic spectra of .6 seconds with
1 millisecond of data. The image was scaled so the full range is 3
sigma of the noise. The images have frequency
- polA dynamic spectra (.gif)
- polB dynamic spectra (.gif)
- .07 seconds to .19 seconds: 1290 remy radar sweeps in front
- 1242,1256 is the aerostat radar.
- 1232,1247 is mode A of the punta salinas radar.
- 1280 is coming from the 1290 radar when it is strong.
- 1290 is the remy radar
- 1310 is the frequency image of the 1290 radar.
- 1350 is the FAA radar
- 1358 is the frequency image of the 1242 MHz Aerostat.
- 1343 is the frequency image of the 1256 MHz Aerostat.
- 33 1 millisecond spectra
overplotted for beam1 (.ps) (.pdf):
- The 33 spectra where divided by the median spectra spectra over
639 millisec and then converted to a db scale.
- Page 1: polA 1300 MHz band
- Page 2: polB 1300 MHz band.
- The ipp for the 1290 radar in 2762 useconds so you see it every
2 or 3 1 millisecond spectra.
- When the system saturates there are copies of the signal at:
- 1290, 1290 +/- 10
- This pattern is repeated every 40MHz about the 1290 signal
- The pattern matches a clipped sine wave that includes a dc
offset (see Clipping
a sine wave). So the 1290 radar (which is a sine wave for the
digitizer) is driving the a/d into saturation. This is also reflected
in the recorded a/d overflow status from the pdev spectrometer.
- Count for PolA/B spectral
- 600-800 Counts in 1450 band, 300-600 counts in 1300 MHz
- The A/D input rms in counts was set to 30 for both bands.
- The radars in the 1300 MHz band made it hard to reliabley set
the a/d sigmas. When the radar ipp hit, the rms would increase by
factors of 2.
- The rms is computed sampling 1 million points 4 times: 5.8
millisecs *4 =23 milliseconds (probably a little longer since the sum
and sum of squares is computed after each million points).
- The radars take about 50-80 milliseconds for the beam to
- We'll need to figure out how to handle this. Doing it a few
times and then taking the minimum might be a way out but it would
increase the power adjust time. We might also want to compute a
spectrum and use the spectral density value to set the attenuation.
- The rms/channel for the 1450 band is larger than the 1300
MHz band (for the same reason as the spectral density).
- The expected rms/mean value
for a 172MHz, 4096 channels, and 1 millisecond sampling is:
- 1./sqrt(172e6/4096*.001) = .154
- The measured rms/mean for polA,B was .16 which is close
- The measured rms/mean for stokesU,V was sqrt(2) less than .16 .
This must be a problem with my scaling. I used:
- rmsU/sqrt(meanPolA*meanPolB) .. this gave sqrt(2) too small.
The plot has the extra sqrt(2) applied.
- The rms/mean clearly shows the images of the radars appearing
in the other half of the frequency band.
- How much of the band do we lose
at the edges:
- The rms/mean starts to increase 5 MHz away from the edges of
- Centering the bands at 1300 MHz and 1450 MHz gives an 11 MHz
overlap at the center of the band.
- Throwing away 5 MHz on the outer edges of the band would give a
freq coverage of:
- (172/2. - 5)* 2 + 150 MHz=312 MHz total bandwidth.
- This covers the 300 MHz of alfa. If you wanted to get back
the outer edges of the band you could increase the separation of the
band centers by a few MHz (although the polarization response is
going south out there).
- The radars and rfi:
- The faa, remy, aerostat, and
punta salinas are all visible
- There is a frequency image of the aerostat 1342
- This is probably to be expected. Complex sampling needs
the I,Q phase to be correct for the image to cancel. When the radar
pulse hits the baseband mixer, things are non linear and the I,Q phase
difference is not going to be 90 deg.. so we should see images.
- On a few 1 second average spectra there are birdies spaced
across the band:
- beam 1 polB when 1290 is strong
- beam 5 polA when 1290 is strong
- beam 6 polA when 1350 is strong.
- This is coming from the a/d being driven into saturation and
clipping the radar "sine wave" input.
- The beams that had saturation are probably a function of az/za.
If we moved then the other beams would probably also have this problem.
- The polyphase filter and 12 bit sampling is doing a good job of
not letting the rfi spread. Even during strong pulses the rest of the
band looks ok (except for the ones that get wiped out in the mixer).
- To do
- Test how effective the a/d