23jun20 lbw rfi using puppi
average data: spectra and total power for 83 seconds
spectra using .17 average spectra (.png):
total power vs time and the power from the 4 radars
Hires total power vs time for each radar
spectra with 240Usec resolution at deepest compression (.png)
spectra with 40usec resolution during compression
40 usec spectra from within the 78.9725 sec compression (.ps)
gns satellite passes over AO during the
83 second movie of the
dynamic spectra (.167 sec resolution) (.avi)
Project p2789 was running puppi in pulsar search
mode using the lbw receiver on 23jun20 from 00:30 am to about 3am.
They reported that more than 50% of the band was corrupted with rfi.
Their setup was:
- Standard lbw puppi setup with rf center freq set to 1680MHz.
- this maps 1380MHz rf to 1200MHz IF (which is the center of
the 800 to 1600 puppi band).
- Puppi used search mode with 2048 channels, 800MHz bw, 40 Usec
To debug the problem i looked at file
which started at 23jun20 00:52:12 AST.
The first few minutes of the file were used.
The telescope was located at az=346 and za=13.2 (looking south)
Plotting the average spectra and total
Each row in the fits file had 4096 spectra
(40Usec each). I computed the average spectra for 500 rows (83.8
seconds of data).
The first plots shows the
average data (.ps) (.pdf)
- Top: Average spectra (for 84 seconds).
- The colored dashed lines show the frequencies of various
- blue: faa radar freq: 1257.59,1252.41,1349.59,1344.41 MHz
- Green: punta borinquen radar (Pb) freq:
- Red : punta salinas radar (Ps) freq:
- purple: Aerostat radar (aero) freq: 1261.25,1246.2 MHz
- I've also labeled
- gpsL1 : 1575
- gpsL2 : 1227.6 (this is close to a punta salinas freq)
- I failed to label a few more gns satellites (gpsL4,beidou,
- Bottom: total power vs time
- the total power (1120 to 1750 MHz) was computed for each .17
second row of data and then plotted vs time.
- the data was normalized to the median value.
The Average spectra has lots of small
spikes in it.
spectra image was made using the .17 average spectra (.png):
- 500 .167 second spectra were used.
- The median spectra was used to flatten the image
- Colored lines flag the radar frequencies.
- there are stonger bursts of rfi every 12 seconds when the
rotating radars point at AO.
- eg (green lines at 1328, and 1332 punta borinquen radar at
4 and 16 seconds).
- during the rest of the 12 second rotation ao still sees
- There are some wide (in freq) interference with a drifting (in
- These are gns satellites after normalizing the image to the
- 25mhz bw area around 1575.
- This is a gps satellite's L1 frequency
- 1175 MHz. this could be gps L5
- 1228: probably gps L2
- 1207:this could be galileo or beidou
- 1267: beidou or galileo
- There are horizontal stripes of black.
- These are caused when the if/lo and or backend goes into
- the median correction causes them to be black since most
of the time there is no compression.
- These can occur when a radar points at AO (every 12
- If you look at the black trace at 48 seconds you'll
see that it lines up with the the yellow aero stat radar
- A large number of the short interference glitches occur in
these black lines (though not all)
- when the system goes into compression, it can create
intermods of signals within the band.
- Most of the narrow vertical lines have a spacing of about
The first summary plot shows the total power vs time and the
power from the 4 radars radars (.png)
The 2nd set of plots is a blowup showing
the total power and a single radar power:
- Each plot has been normalized to the median value.
- The radar plots use a single channel
from the 800 MHz spectra (390KHz bw).
- the radars transmit at more that 1 frequency. I've only
plotted one freq from each radar.
- Top: total power vs time (1120-1750 MHz) using the
40usec samples (although there aren't enough pixels to show the
- The values that go below 1. are when the system compresses
(eg 78 seconds)
- Some of the blank areas are when a particular radar blanks
in our direction.
- 2nd: Aerostat radar (1262.25 MHz)
- The wide blanking region is when the aerostat points at AO
(eg 20 to 22 secs)
- The narrower blanking region is when the aerostat points at
the San Juan airport.
- The total power compression corresponds to peaks in the
- 3rd: Punta Borinqeun radar (remy) at 1332.59 MHz
- This has a much narrower blanking region (eg at 50
secs). It blanks for about +/- 2 degrees of rotation.
- There is a large sidelobe 6db above the blanked main beam
12.6 degrees from ao boresight. This is a known problem with
- 4th: FAA radar (1349.59 MHz)
- The Punta borinquen and FAA are similar carsr radars (using
different sets of freq).
- The faa blanking is also +/- 2deg
- the off ao boresight level is more constant the the PB
- Bottom: Punta salinas radar (1294.5 MHz)
- this has a larger blanking area when it points at AO.
- The deepest compression occurs around 78 seconds
- The longest compression occurs 47 to 49 seconds
- The compression will turn out to be from the aerostat radar.
- It does not occur around the blanking region when the
aerostat points at AO
- It must occur at different sidelobes of the radar
- It occurs more that once in a 12 second rotation.
Dynamic spectra blowup at deepest
One row of data from the fits file (.167 secs,
4096 spectra) at the deepest compression (row 470.. cnt from 0.. @
78.93 seconds) was used to look at the affects of compression.
The dynamic spectra shows the
.167 seconds of data (.png)
- The 4096 spectra were smoothed in time by 6 ( 240usecs
resolution) to fit on the screen.
- The horizontal compression lines line up with the aerostat
- Looking at the aerostat total power vs time, this period was
about 3 seconds away from when the aerostat pointed at
- The colored lines show the radar frequencies (ignore the
turquoise noise line :).
- The dark horizontal lines (where the compression occurs) line
up in time with the individual aerostat ipps (occurring about
- The bursts every 90 milliseconds at 1620MHz is an iridium
satellite (or a ground user).
A second dynamic spectra
blowup using 9 milliseconds of time about 78.973 secs was made
- This has 215 40usec spectra. I've blown the vertical
scale up by a factor of 3 (pixel replication).
- The colored lines mark the different radars
- The aero rf pulses are:
- 160 Usecs at 1262 usecs folllowed by 160 Usecs at 1246 MHz
(4 40usec sec spectra followed by 4 40usec spectra).
- Each pulse is a chirped signal of 1MHz bw.
- there are 7 time intervals for the pulses that then repeat:
(1371,3504,3076,2809,2903,3289, and 3676) usecs
- The compressed regions clearly occur during the aerostat
- The compression starts right after the 1262 Pulse start and
ends toward the end of the 1246 mhz pulse.
- During the band compression there are a series of
pulses across the entire band. These pulses are not there
outside of the band compression
7 40 usec spectra from
within the 78.9725 sec compression were plotted (.ps)
- Each spectrum is plotted using a different color.
- the spikes go full scale (8 bits)
- The spikes are also present in the 1000 to 1150MHz part of the
- The 1-2 GHz IF1 filter cuts off power in this part of the
- So the spikes are not coming from the AO IF/LO.
They must be coming from the puppi backend.
- The most likely source of the spikes are:
- Clipping at the puppi A/D converter
- clipping in the xilinx butterfly stages of the polyphase
filter bank (18 bit fixed point arithmetic).
- The final fft is done in the gpus (which doesn't have this
problem since they use floating point)
- either of these can generate lots of spikes (my guess is the
The Global Navigation satellites
The row averaged (.167) dynamic spectra showed
wide rfi at some gns bands:
- gps:l5: 1175, l2:1228,l1: 1575
- beidou: 1207, 1268
- galileo: 1260-1300
- (more info)
plots were made of the various satellite constellations positions
relative to AO on 23jun20 1:00
- Each frame of a page holds 8 satellite (each plotted in a
- The y axis is the ao Za for the satellite vs time.
- The x axis is the satellite azimuth relative to ao (this is
the source azimuth, not the feed azimuth).
- the right col lists the satellite name, yymmdd hh:mm:ss time
for the minimum za for the pass
- the * plot the orbit. it is stepped in 2deg of orbit phase
- the square symbol is the first point of the pass.
During the 89 secs of the dynamic spectra the telescope was
at az=346, za=13 (looking south).
Satellitesthat could have caused what we saw:
What can be done:
- The if/lo system was being compressed by the aerostat radar
- It is not normally up in the air at 1:00 am
- spikes across the entire band were being generated during the
- They were being created after the if/lo filters.. so they
must have been created in puppi.
- Either the A/D converter or the butterfly stages of the
polyphase filter bank.
- The wideband (in freq) continuous rfi was coming from gns
- The lband wide system has had some problems, but the 2degK
increase in Tsys during this period did not cause any of these
- I didn't look into any dropped packets by puppi.
- The aerostat is normally 20db higher than any other radar
since it is floating 5000 Ft up in the sky, looking down at the
ground (no terrain shielding)
- For ground based radars, a plane flying by can give large
signals (it gets around the ground shielding). In the case of
the aerostat i don't think this is the case.. (so we can't just
say this was a plane flying by).
- experience has shown that the signal level you get from a
particular radar can vary with az and za. (by many db).
- If the system is being overdriven, then the only thing you can
do is add more attenenuation (before the piece that is
- This could be the rf attenuators (before the first mixer) if
it is the first mixer saturating.
- the if attenuator (upstairs after the first mixer and before
the fiber optics) if the fiber optics is saturating.
- 2nd if2 gain or the puppi attenuators .. if the problem is
- doing this can cause the system temperature to go up a
little (eg.. too much attenuation before the hot fiber optic
tx..). But an increase in tsys may be better than no data :).
- since puppi was causing the spikes, increasing the if2
or puppi attenuation would have helped in this case.