Intro

The setup:

• cband 4-6 Ghz
• mock spectrometer
  
• 172.032 Mhz bw, use all 7 boxes covering 4-5 ghz or 5-6 ghz.
  
• 8192 channels, .1 sec integrations, 32bits polA,B
• pshift, adjpwr setting are show in the table below:

• file	pshift	adjpwr a/d Counts	freq rangeGhz	time secs	Notes
  0000	0x1ff5	30	4-5	60	std pshift
  200	0x1555	30	4-5	60	noise pshift
  300	0x1555	90	4-5	30	noise pshift increase input signal
  400	0x0aaa	30	4-5	60	noise pshift, first butterfly no shift
  500	0x1ff5	30	5-6	60	std pshift
  600	0x1555	30	5-6	60	noise pshift
  700	0x0aaa	30	5-6	60	noise pshift, first butterfly no shift

  
  
• 0x1555  shift every other butterfly (this should work for noise)
• 0x0aaa   shift every other butterfly, but no shift on the first butterfly
  
• 0x1ff5    9 downshifts then alternate every other butterfly. This is what cima currently uses.

Processing:

• Each file was averaged
• Data was extracted and plotted about DC and channel -2048 (relative to DC)
• Set 1:
• about DC (+/- 100 channels) were normalized to their median and then plotted
• Set 2:
  
• 100 channels were extracted, . a 2nd order fit was made and used to normalize the data
• all 7 172 mhz freq blocks were then averaged
• This was done for channels
•  dc to dc+100
• -2048-50 to -2048+50 (channel 0 is DC).
  

• Page 1: +/- 100 channels about DC normalized to median value
• Black: polA, Red: polB
  
• each of the 7 172 Mhz frequency blocks (4 to 5 ghz) are over plotted with an offset for display
• Top PSHIFT=0x1ff5 . This  is the current pshift in use with the mocks.
• The dashed lines flag channels,4,8,16,32,64.
• You can see jumps in the spectra at 4,8,16,32,64, but only on the positive freq half of DC
• Bottom PSHIFT=0x1555. shift every other butterfly (for noise)
• the jumps in the spectra are no longer visible
• Page2,3 divide 2nd order polymial into 100 channels of spectra:
• the  2nd order polynomial was fit and then divided into the 100 channels
• The top frame is polA, the bottom frame is polB
• The colors show different files with different pshift and adjpwr
• the black trace has pshift=0x1ff5 (the current standard).
• the red trace uses 0x1555. shift every other butterfly (with a shift on the first butterfly)
• the green trace has pshift=0x1555 but the power level has been increased from 30 to 90 counts.
• the blue trace has pshift=0x0aaa. this is alternating every butterfly with no shift on the first butterfly
• Page 2: channels DC to 100
• The black trace has jumps of .005 to ..01 Tsys for channels,4,8,16....
• The other traces do not show any jumps
• The single channel jump at channel 15 for red,blue is probably rfi.
  
• Page 3: +/- 50 channels about chan -2048. remove 2nd order polynomila
  
• the black trace has a jump of .006 Tsys
• The other colors show no jumps

Summary:

• the spectral  jumps at 2^n channel spacing are caused by butterfly underflows
• The pshift value of  0x1ff5 currently used by cima causes jumps of .005 to .01 Tsys.
• When a  pshift of 0x1555 is used (shift every other butterfly) no jumps are seen
• Sensitivity:
• The integrations were 60 seconds,
• We added 7 frequency bands together so the sensitivity is close to a 400 second integration
• Increasing the power level, or decreasing the shift to 0x0aaa did not seem to improve things.
• Looks like 0x1555 is sufficient. the lower number of shifts, or higher input level would make the system more susceptible to pfb overflows from rfi.
• 4-5 Ghz has some rfi
• 5-6 Ghz has much stronger rfi
• Data taken with 0x1555 showed no pfb  overflows so it might be ok to use this setting for cband

todo:

• Need to do azswings with the 0x1555 pshift at dome around 15 degrees and see if the rfi gets strong enough to cause pfb overflows
• changing pshift from 0x1ff5 to x01555 will increase the voltage in the fpga by 2^4=16. we need to make sure that the rest of the path works for all of the different configurations that we have.
• For lband we probably need to keep the 0x1ff5. at least for the band that includes the radars
• It is going to be a challenge for the user if we allow a different pshift setting for each of the 7 bands of each box.

• The are from underflows in the pfb
• removing shifts on the left (1st butterflys removes the jumps faster than removing the shifts from the right.. this makes sense).
• single pixel observing without radars (when not using lband) can probably use regular noise like shifting (every other butterfly)
• Lband with the radars needs the aggressive shifting or we get pfb overflows on every radar ipp
  

<- page up
home_~phil