From John Effland via Charles Cunningham:

 
The stability data presented at Groningen along with the setup used to obtain the data.  Please note the following:

The actual measured data from the mixer-preamp is shown by the red trace at higher frequencies and by red, pink, and dark blue traces at lower frequencies.  The data are reconstructed from a number of sweeps on the FFT analyzer using different maximum frequencies.  This is required to obtain adequate resolution at the lower frequencies.  A check of the calibration is given by the purple X, which is the theoretical noise assuming an 8 GHz bandwidth.  It's not surprising that the measured delta V/V is higher
that the theoretical value because the shape of the noise passband from the mixer-preamp has a drop in the middle of the IF band. That most likely yields an effective noise bandwidth that's smaller than 8 GHz.

The light blue trace shows the setup variance when the setup is configured to measure just the noise contributions from the IF system.  In the setup diagram, the Radiall Switch is connected to the hot load.  Note that this trace falls close to the theoretical 8 GHz bandwidth mark.

The green trace is a calibration that shows the noise when the measurement bandwidth is reduced from 8 GHz to 500 MHz using the K&L 4B380-4750/500-0 filter (shown by the dashed lines in the block diagram of the setup).  Note that the theoretical level using this filter, shown by the green square, is just slightly higher than the measured
green trace.  This is explained by recalling that the filter's noise bandwidth is certainly greater than 500 MHz, which is just the 3 dB bandwidth.

 Larry notes:

.  The various plots on top of each other are hard to read.  There are
two red ones and two blue ones, as well as other colors.  I've used
what seems to be the best of them, the lower red plot.

.  Presumably what's plotted is the *single sided* voltage spectral density, since the stated theoretical value for effective bandwidth B is \sqrt(2/B); for two-sided, it would be \sqrt(1/B).

.  To the best of my ability to measure from the plot, and ignoring       the peaks at 2.3, 38, and 60 Hz, I find:
      f_c = 200 Hz    (corner frequency)
      \alpha = -1.14  (2*slope)
      B  = 4.3 GHz    (VSD = 2.14e-5/sqrtHz)

   This yields a flicker coefficient of the gain fluctuation PSD of    G_1 = 9.8e-8 /Hz   (see Memo 466 for explanation).

   The apparent effective bandwidth is surprisingly low, even    considering John's explanation of a gain hole in the band.

   Comparison against Table 1 of Memo 466 shows that G_1 is worse    than almost all HFET amplifiers reported.  Based on scaling those
   reported numbers to the characteristics of the ALMA IF amplifiers   built in Charlottesville with "Cryo 3" wafer InP transistors, I had
   predicted G_1 = 1.2e-9 /Hz (bottom of p9 in Memo 466).  Of course,   that was just for the cold preamp, and this measurement is for a
   complete receiver.

.  The data for the IF system alone (light blue curve) gives
   f_c = 90 Hz
   \alpha = -1.2
   B = 7.8 GHz

   and this yields G_1 = 2.8e-8.  This is quite bad as well, so with   this system it would not be possible to measure a really good   receiver.

.  The peak at 2.3 Hz is no doubt gain modulation from the cryocooler, either from vibration or from temperature modulation (it would be nice to know which).  And the 60 Hz peak is no doubt from the power line. Do we know the cause of the 38 Hz peak?  It looks like the 60 Hz modulation is in the IF system, since it's just as strong for the IF alone.  The other peaks occur only with the full receiver.

Test Setup