Alfa Noise test 02mar04..
The alfa if/lo processor was tested on 02mar04
while it was installed in the receiver room. One noise source was connected
to pixels 1ab,2ab,3ab,4b,5ab,6ab,7ab . Pixel 4a had a separate noise source
connected to it. The noise source provided a signal at lband. This was
split, routed to the rf amplifiers, mixers, through the fiber optic xmter/rcvr,
and then into the wapps. A 100 Mhz bandwidth (3 level sampling) was taken
with 4096 usecond dumps for 60 seconds.
The IF/LO processor was placed in rack 1 below the
radios. To check for rf pickup, synthesizer 1 in the downstairs racks was
set to: 1308 Mhz, 14 dbm, and front panel output (with a bnc to sma connector
present). This birdie was picked up by the alfa front end IF/LO.
The plots are:
second average spectrum.
correlation (of power) by freq channel using pixels 4a and 4b.
Fig 1 top. The 7 pixels (coded by color) for polA. The 1308 Mhz pickup
is seen as well as a resonance in pixel 1a at 1289 Mhz. The bottom plot
Fig 2 This is the strength (in units of Tsys) for the 1308 Mhz birdie versus
pixel number. Black in polA, red in polB. We do not know if the variations
of coupling into the boards is being caused by the boards, or by the illumination
of the box by the synthesizer.
1000 samples (4 seconds of data) was used to compute the correlation
coefficients (by channel) between pixel 4a and 4b. These pixels had different
noise sources for input. Dynamic spectra spc[nFrqChn,nSmples] were used
to compute the correlation coef.:
corcoef[i,j]= sum(spc4a[i,*]*spc4b[j,*])/1000. where the * is summed over time
The normalization removed the robust mean and then divided by the
robust rms before doing the multiplication and summation (robust
means exclude outliers). Note that power (not voltage) was being correlated.
Pixel 4a is on the horizontal axis and pixel4B is on the vertical axis.
The average bandpass is plotted on the left and at the bottom (they are
not part of the correlation coefs). The large black lines are the 1308
Mhz birdie. The only correlation is the vertical line for pixel 4A above
1308 Mhz. It must have been that the 1308 Mhz coupling changed over the
4 seconds, and pixel 4B must have been slightly in compression because
of the large birdie. This would cause all channels of pixel4B to move with
the changing sine wave. Since pixel4A 1308Mhz would also see this
variation, it has some correlation with all theother channels (thanks desh
for beating this into me..). It is a bit strange that there was any compression
when the size of the sine wave (in a single channel) was 40 * Tsys. With
2048 channels this is 2% (or maybe 4% since we hanning smoothed). It may
also be that the 3 level correction is not being done perfectly...
coef (power) for pixels 3a,3b,4a,4b
The same channel auto and cross correlation coefs. were computed for
these 4 pixels.
There are 10 different correlations: (3ax3a,3ax3b,3ax4a...).
The auto correlations should be 1 (since it was the same signal).
The correlations between different noise sources (all cross correlations
with 4a) should be 0 since the different noise generators were not correlated.
The cross correlations using the same noise source (but different digitizers)
should be 1 if all of the noise power was coming from the noise source.
We see a value near .65 . On the edges, most of the power is coming
from the amplifiers after the filter (which are not correlated). The .65
could be explained if the power levels into the if/lo were set too low.
If the fiber optic noise was a not trivial part of the noise , then you
would get a correlation less than 1 (we need to recheck this).