Setup/processing  01aug06:  mmic amp,  noise src, cbh using square law detectors.
Setup/processing of 08aug06  data: cbh using the wapps.
Plotting the data: 1 second averages, allen deviation, acfs.
Plotting the data: spectra.
Conclusions

---

01/08aug06 4-6 ghz amp,cbh noise/gain stability.

---

Setup/processing  01aug06:  mmic amp,  noise src, cbh using square law detectors.  (top)

    The  4-6 Ghz mmic amplifier was set in a dewar in the cryo lab. A 500 Mhz band was mixed down to a cfr of 500 Mhz and then sent to the control room where it was square law detected with a 50 Usec time constant and then sampled at 25 usec. Data was taken for about 42 minutes using the RI. 3 other bands were sampled simultaneously:

• riDig channel 1 left: mmic amplifier from cryo lab (500 Mhz bw). The dewar is looking at a warm 50 ohm load.
• riDig channel  1 right: 500 Mhz noise src in receiver room.
• riDig channel 2 left: cband high polA with shutter closed (looking at absorber) (500 MHz bw)
• riDig channel 2 right: cband high 500 Mhz polB shutter closed (looking at absorber) (500 MHz bw)
    The first 3 signals used the square law detectors on the right rack. The 4th signal (cbh polB) used the square law detector in the left rack.
    60 Hz at about 2 millivolts was seen on the detected signals. AC coupling the input to the square law detector did not get rid of the 60 hz. AC coupling the output of the square law detector (before the  opamp) got rid of it. So the 60 hz is coming from the square law detector power supply.

The processing steps for the data were:

• Average the data to 1 second and plot vs time. This shows how the gain wandered around.
• Compute the allen variance (or deviation) for integration times of:
• 25,50,125,250,500 usecs
• 1.25,2.5,5,12.5,25,50,125,250,500 millisecs
• 1.25,2.5,5,12.5,25, 50 seconds
The allen deviation was computes by taking 1/2 the difference of adjacent integrations, and then computing the rms of that series.
• Using the difference time series where the samples were 25 useconds apart,  compute the acf for the first 1 million points (50 seconds of data). Use this to see if adjacent samples were independent.
• Compute average spectra of the power time series for the 42 minutes. Two resolutions were used:
• 64K long transforms giving 1 hz resolution, 20 Khz bandwidth
• 16K long transforms after averaging 50 samples. This gave 50 millihz resolution and 400 hz bandwidth

 
    The data was processed twice. The first time was with no filtering. The second time with filtering to try and remove some of the 60 hz. A mask of 1 hz width about the 60 hz harmonics (up to 960 hz) was created. The filtering was done by: grabbing 50 seconds of data (2^21 points) at a time, fft, multiply by the mask, and then inverse transform. This second data set was then processed similarly to the first set. The mask forced the frequencies about the 60 hz harmonics to go to zero. It probably would have been better to use the mean spectral density.

---

Setup/processing of 08aug06  data: cbh using the wapps.

    The cbh with the shutter closed was used. The wapps ran at 32 usec sampling, 3 levels, and 100 Mhz  bw. Data was taken for about 45 minutes. The 0 lag was 3 level corrected and then used as the total power. The processing was the same as the 01aug05 data (except that the maximum sample rate was 32 usecs rather than 25 usecs).

Plotting the 1 sec avgs, allen deviation, and acfs.  (top)

• Top plot: 1 second averaged time series:
• 01aug06:  cbhPolA (blue) had a jump around 100 seconds of .5% Tsys and a glitch around 1900 seconds. The noise src had a negative drop around 1900 seconds. I wonder if someone was moving the cables that went to the ri?? The filtered and unfiltered data sets are the same.
• 08aug06 : The data shows a 512 second cycle of about 1% of Tsys. This was not present on 01aug06. The turret room temperature was monotonically increasing during the time period so it is not coming from the gregorian dome. This data used the same amps in the downstairs IF/LO as the 01aug06 data (where the cycle was not seen). The analog unit of the wapps in the clock room was different. Either the ac cycle was different on 01aug06 or the gain variation with temperature is in the wapp analog chassis.
• Center plot:  the allen deviation:
• 01aug06: The integrations times were 25 usecs to 50 secs. The purple diagonal line is the expected rms for 2*integration times (since we subtracted two points).  The allen deviation shows:
• The measured differences start out below the expected deviation. The difference value for an average of 5 to10 agrees with the expected rms.
• The unfiltered data has a maximum around 8 milliseconds (1/2 of a 60 hz cycle).
• The unfiltered data again agrees with the expected rms when the integration time is 50 milliseconds. The difference time is 100 milliseconds. This is a multiple of 60 hz so the 60 hz component should cancel.
• The filtered data agrees with the  expected value up to about 50 milliseconds. This shows that the bump at 8 milliseconds was from the 60 hz.
• The test dewar has a peak around 250 milliseconds and then drops off. This is probably from the strong .8 second cycle from the cross head. The  noise source (with no refrigerator) remains flat. The cbh dewar rises but does not peak (it's thermal mass probably damps out the refrigerator cycle).
• cbhPolB rises slowly starting at .1 seconds. cbhPolA starts rising at 50 milliseconds and then remains flat. cbhpolB was taken with the square law detectors in the left rack. They seem to work better than the ones in the right (that cbhPolA used).
• 08aug06: The integration times are 32 useconds to 64 seconds. The allen deviation shows:
• The allen deviation is decreasing linearly (1/sqrt(time)) from 32 usecs to about .16 seconds where it flattens out. At about 10 seconds it starts to increase. There is little 60 hz nor 1.2 hz from the crosshead.
• The measured rms is 1.47 times (about sqrt(2)) times larger than the expected rms. I've included the factor of 1.23 for the 3 level sampling when computing the expected rms. I'm not sure why the measured value is not closer to the expected one. The counts in the 0 lag were close to the optimum values.
• Bottom plot: acf of time series difference:
• 01aug06: The acf is non zero until about lag 5 (see the filtered data). This explains why the first 3 allen variance measurements (1 deltaT, 2 deltaT, 5 deltaT) are lower than the longer integrations. The  time constant ( 50 usecs) rolled off slower than the sampling (25 usecs) leaving adjacent samples with some correlation. The actual allen variance is probably linear down to the 25 usecs differences.
• 08aug06: The time series from the wapps have no correlation all the way down to lag1.

---

plotting the spectra:  (top)

• Top plot: 1 hz resolution. full bandwidth: The spectra have been offset for plotting.
• 01aug06: 20 khz bandwidth, 64K length xforms. The filtered data have negative going spikes at the 60 hz harmonics since this power was removed. There is a spike around 8435 hz (118.5 usecs). It is not in cbhPolB (which used the leftmost square law chassis) so the birdie is probably coming from the square law chassis on the right. The power falls off with frequency. This is probably being caused by the square law time constant. The acf showed that the first few lags were non zero.
• 08aug06: 15.5 Khz bandwidth, 32K length xforms. The spectra is clean except for a spike at 60hz. The spectra is also flat with frequency. It does not fall with frequency like the 01aug05. The level is also higher. It should be a factor of 2 higher since the transform length was 32K rather then 64K.
• 2nd plot: 1 hz resolution blowup  (0-2500 hz):
• 01aug06:The 60 hz on the test dewar has more harmonics than the others. cbhPolB has the least 60 hz (it used the left square law chassis). The filtered plots show the negative going spikes at 60 hz harmonics. The test dewar has a few harmonics above the 960 hz filter cutoff. We looked at the 60 hz on the scope and it is a square wave rather than a sine wave.
• 08aug06: There is 1 spike at 60 hz and 1 at 180 hz. This 60 hz is coming from a sine wave rather than a square wave.

• 3rd plot: 1 hz resolution blowup ( 0-500 hz):
• 01aug06:  The harmonics of 60 hz are flagged. cbhpolb has the least 60 hz (square law detectors from left rack). The filtered version is plotted on the same scale. It has very little 60 hz.
• 08aug06:  There is a 60 hz and a 180 hz spike. Pol A is a little stronger than polB.
• bottom plot .05 hz resolution:
• 01aug06:  50 points were averaged and then a 16384 length xform was done. You can see the strong 1.2 hz birdie from the cross head for the test dewar. cbhPolA (blue) has a lot of junk from 0 to 1 hz with a peak at .5 hz. This is also evident when you look at the 1 second average time series. There is a lot stronger high frequency (.5 hz) noise in cbhPolA than the other data sets. This is probably why the allen deviation for chbpolA flattens out around .5 seconds.
• 08aug06: 75 points were averaged and then a 8192 length xform was done. You don't see the 1.2 hz birdie from the crosshead. There is more low frequency power caused by the gain cycling from the AC.

• Square law detector data:
• The  8 millisecond peak in the unfiltered allen deviation is coming from 60 hz in the square law detectors.
• The square law detector chassis on the right has more 60 hz hum than the one on the left. It also has an 8.5 khz birdie.  The 60 hz is a square wave giving many harmonics. It's power supply should be looked at.
• The second bump in the allen deviation around .25 seconds in the filtered data set is coming from the 1.2 hz crosshead  cycle.
• The time constant in the square law detector has correlation's up to 4 times the nyquist sample frequency. This causes the allen deviation to be smaller than expected for short integrations.
• The allen deviation decreases as 1/sqrt(integTime) up to about 25  milliseconds integration time (looking at the filtered data).
• The wapp data:
• there is very little 60 hz.
• The gain is changing by about 1% tsys on a 511 second cycle. This is probably coming from the analog units of the wapps. A lot of on/off position switching does a 5 minute on, 5 minute off, followed by a single cal. The cal assumes that the gain has remained constant for the on and the off. It woud be good to slow down the AC cycle some.
• The allen deviation decreases as 1/sqrt(integTime)  up to .1 second integrations. The measured rms is about sqrt(2) larger than the expected rms.

28jul06 test 4-6 ghz amp in test dewar. (top)

• Top:  total power vs time. The total power samples were avg. to 1 second and then plotted. They have been normalized to the average total power value. The red dashed lines are the expected rms for a 600 Mhz  1 second integration. Looks like the total power jumped a bit around 1200 and 1500 seconds.
• Bottom: The black plot is the allen deviation. To compute this we:
• Made n passes thru the data.
• On each pass avg. a different number of adjacent points (1,2,5,10,50,100,200,500,1000,...)
• After averaging, compute the difference of adjacent points and then compute the rms of these differences. This is what is plotted in black
• The green * are averages that are integral multiples of 60 hz.
• The red line is the expected rms for a data set with 600 Mhz bandwidth and avg*2 seconds (*2 since the subtraction of the adjacent points is the same as averaging by a factor of two more).

top: The detected power was sampled at  10 useconds. This shows the full 50 Khz averages over the 1600 seconds with about 1 second resolution.

2nd: Blowup of 0 to 3 Khz. Multiples of 60 hz have been flagged in red.

3rd: further blow of 0 to 500 hz. showing the first few multiples of 60 hz.

bottom: .064 hz resolution data 0 to 10 Hz. Multiples of 1.2 hz have been flagged (the refrigerator cycle).

processing: x101/060728/doallen.pro

<- page up
home_~phil