If2 noise source test
26jan07
The downstairs iflo has a noise source than can
be used to inject power into the downstairs iflo system. On 26jan06
this noise source was used to feed the pdev spectrometer (pdevS0). The
setup was:
-
170 Mhz sampling tied to station clock
-
8192 fftlen
-
dana's mixers used to mix from 250 Mhz down to base band.
-
Band0 (the lower band) centered it's band at 250 - 60 Mhz before down converting.
Band 1 was centered at 250 MHz + 60 Mhz.
The data set taken were:
-
3600 seconds of data with 1 second integrations. 32 bits were saved.
-
60 seconds of 1 millisecond integrations. 16 bits were saved.
The sections contain:
Spectral average of 3600 seconds showing Birdies
in the bandpass.
Total power stability.
Spectral average of 3600 seconds.
3600
sec average of 1 second spectra shows any birdies (.ps) (.pdf)
-
Page1 3600 second average spectrum. Black in polA, red is
polB. Top plot is band0, center plot is band1, bottom plot are both spectra
with the 60 Mhz offsets of the mixers. You can see a birdie at -65 Mhz
in polB of band0.
-
Page2 blow up of average spectra. The top two plots are blowups
(+/- 10 MHz and +/- 3 MHz) of band 0. The bottom two plots are blowups
of band1. You can see 220 Khz birdies in band0. This is probably coming
from the power supply. The structure a +/- 1 Mhz is probably coming from
the IF processing/amps before the pdev device.
-
Page3 remove the bandpass: The bandpass for the 3600 second integrations
was removed by transforming each spectra, removing the lower most 10% of
the frequencies (with cos^4 window) and then transforming back. The
vertical scale is in units of Tsys. T top 2 plots are band 0, the bottom
two plots are band1. You can now see some birdies sticking up above the
noise. Only birdies that appear on both sides of DC come from the spectrometer.
Birdies on only one side are injected before the I,Q mixing.
-
Page4 bandpass removed spectra (vertical blowup).: This has the
page 3 plots with a fixed vertical scale (+/- .002 Tsys). The birdies at
base band are flagged in color. The single sided birdies are labeled in
black (-66,-65, -50 Mhz).
-
All bands see the 80 Mhz birdies (this is the bus frequency for the power
pc on the spectrometer). PolB has the 83.34 Birdie. The 70 Mhz birdie is
seen in band1 and polB band0.
-
Page 5 bandpass corrected spectra at IF: The spectra are plotted
at their IF frequencies.
Total power stability:
The total power was computed for each time sample for
the 1 millisecond and 1 second data sets. The average spectra over each
data set was computed and then used to normalize each spectra before computing
the total power. 200 Khz about Dc was excluded from the total power.
The plots show the
total power results (.ps) (.pdf):
-
Page 1&2 have the 1 second data:
-
Top plot: total power vs time. Black is polA and red is polB. The
vertical scale is in units of Tsys. The gain of the signal is varying by
.002*Tsys over the 3600 seconds.
-
Middle plot. Spectrum of total power vs frequency: The maximum frequency
is .5 hz (1 second sampling). PolA shows a number of birdies in the total
power spectrum They (and their harmonics) are flagged in color. The
birdies appear in both the low and the high band of data.
-
Middle plot. Spectrum of total power vs period. This is the same
data as the middle plot now plotted against period.
-
Page 3 1 ms total power vs time: The data has been smoothed to 5
milliseconds before plotting. The top two plots are polA (band0 and band1)
while the bottom two plots are polB (band0 and band1). PolB has a
lot of junk in it (about 1 second period).
-
Page 4 Spectrum of 1 millisecond total power data. The top two plots
are polA (band0 and band1) while the bottom two plots are polB. The dotted
red lines have been plotted every 60 hz.
-
All of the bands show 60 hz.
-
Band 1 pola and b also show 180 hz.
-
242.46 hz is also present in all bands. It is much stronger in band1 polA
and B.
-
Page 5 Blowup of 1 millisecond total power spectra: This shows 0
to 10 hz of the 500 hz spectra. The junk in polB total power can be see
to be birdies at 1.05 and 1.87 hz (and their harmonics). PolA does not
have this (so it is probably in the if/lo and not the spectrometers).
-
Page 6 rms vs integration time: The rms was computed for power of
2 integrations of 1 millisecond to 128 seconds. The data below 1 second
come from the 60 seconds of 1 millisecond sampled data. The data above
1 second come from the 3600 seconds of 1 second data (the 1 second point
comes from both data sets). The * are pol A, the + are polB. The black
solid line with diamonds is the expected rms for the bandwidth and integration
time. The top plot is the standard rms. The bottom plot is the square root
of the allen variance:
AllenDev[i]=sqrt(.5*mean((y[i]-y[i+1])^2))
-
Both the rms and allendeviation do not follow the expected rms (although
the allen deviation is closer).
-
The standard rms for the 1 second data is essentially flat with integration
time. This is being caused by the gain variations that are seen in the
input data set.
-
Since polA and polB follow each other, the problem is probably not in the
spectrometer (since the spectrometer handles polA and polB the same).
-
The noise source is a single noise diode that is split. The differences
in polA and polB must be coming from the amplifiers after the diode and
before the spectrometer.
-
We need a better source of noise to check if the noise decreases as 1/sqrt(bw*tau).
processing: x101/070126/cmptp.pro plottpRms,plotTpSpc1ms.pro,plotTpSpc1sec.pro
<- page
up
home_~phil