spectral jumps at 2^n channels
24may10
Spectral jumps have been seen at power of 2 channels
previously. The jumps correlate with the input power levels. Higher
input levels make the jumps go down. Not doing the first down shift in
the butterfly stages also makes the jumps smaller. They may be an
underflow in the butterfly processing. Computing the rms values in
counts when the jumps occur seems to leave values that should not be
underflowing.
20may10: high res decimation test
- Shows jumps vs decimation 1 to 20.
- The size of the jump decreases with increasing input level. The
input level changes with the decimation used (but not by more than a
factor of 2 in voltage).
23apr10: spectral jumps
for adpwr=30,45,120 and pshift=0x1ff5, 0x0ff5.
- Not doing the first butterfly down shift make a big difference.
08jul08: spectral
jumps vs A/D and pshift levels.
- Shows jumps for adjpwr:15,20,25,30,80 with various pshifts. Has
fftlens 8192,4096,2048,1024,512
23apr10: spectral
jumps with noise source.
Data was taken using the if2 noise source. The setup
was:
- 8192 channels, 172 Mhz bandwidth.
- integrate for 60 and 300 seconds.
- Take data with adjpwr of 30,45 and 120.
- Use the standard pshift=0x1ff5. Also try pshift=0x0ff5 (where
first butterfly down shift is not done).
- test numbers:
- pshift=0x1ff5, adjpwr=30, 60 secs
- pshift=0x1ff5, adjpwr=45, 60 secs
- pshift=0x1ff5, adjpwr=120 , 60 secs
- pshift=0x1ff5, adjpwr=120 , 300 secs
- pshift=0x1ff5, adjpwr=45 , 300 secs
- pshift=0x0ff5, adjpwr=45 , 300 secs
- pshift=0x0ff5, adjpwr=30 , 300 secs
- pshift=0x0ff5, adjpwr=120 , 300 secs
The plots show the spectral jumps for this
test (.ps) (.pdf):
- Page 1: The jumps at channel 2048 for the 9 tests.
- A linear fit was done to channels 2007-2047 (fit1 red) and
(2048-2088 fit2) green.
- The jump was: (fit2[2048] - fit1[2048])/mean(data[2007:2088]) -
1.
- The vertical scale is Tsys close to Channel 2048.
- Page 2: jumps, power vs test number:
- Top: The size of the jump vs test number. Black is polA, red is
polB
- Bottom: the A/D rms in counts fore each test.
- Page 3: Size of jump vs a/d rms input level
- polA is +, polB is *
- Black has the standard pshift=0x1ff5, red has pshift=0x0ff5
- The size of the jump decreases as the rms input level increases.
- Not doing the first down shift in the butterfly stages makes a
large decrease in the jump size.
processing: x101/100423/mockdlp.pro
<-
page
up
home_~phil