29sep12 spectral channel jumps (top)

• A pfb overflow means that the data has gotten too large for the 18 bit multipliers in the fft butterfly stages.
• When  going through the fft butterflys, noise voltage doubles every other butterfly. Sine waves (rfi) can double on every butterfly.
  
• We start with a 12 bit number in the uppermost 12 bits of the 18 bit register. This leaves 6 lower bits that are unused
  
• We downshift on every butterfly stage for the first 9 butterfly stages. So there is no way a pure tone can cause an overflow. So fftllengths of 512 or less should never have pfb overflow errors.
  

• use spectrometers: b0s1g0,b1s1g0,b2s1g0,b3s1g0.
• 86 MHz bandwidth, 172.032 MHz clock. The digital filter reduced the bandwidth by 2.

• b0-b6s1g0.
• The full 172.032 MHz bandwidth was used.
  
• The overflows disappeared after the full bandwidth was used.

• The data came from  B1919+21 file .00200 (4 frequencies).
• The first row was used. It had 10200 spectra samples at about 400 usecs.

• Each page shows a different frequency band.
  
• The top Frame is pola, the bottom frame is polB.
• Page 1: 1170Mz. This has two spikes on with different amplitudes. These may be rfi.
  
• Page 2: 1420 MHz
• polB has a number of single channel spikes
• The spikes have one of two possible amplitudes (this is not normal for rfi).

• There are 10200 spectra in each image
• the top frame is polA, the bottom frame is polB.
• 1170 MHz Dynamic spectra (.gif)
• The spikes on the left look wider than 1 channel
• there is some broadband rfi around 1.7 seconds
  
• 1420 MHz Dynamic spectra (.gif)
• PolA has a few spikes
• PolB has spike all over the place.
  
• 1520 MHz dyanmic spectra (.gif)
• This is clean
  
• 1620 MHz dynamic spectra (.gif)
• there is rfi around 1620. This is probably iridium.

• Each spectra was subtracted from the following spectra (400 usecs later) for each of the 10200 spectra.
  
• A histogram was then made of the deltaSpectra values for each frequency band.
  
• Each frame is a different freq band
• Dashed green lines were drawn at -(8192+4096) - 4096 4096 and (4096+8192) counts
• Black is polA, red is polB
  
• Frame 2 shows the 1420 MHz band (which had the jumps)
• red polB It has counts were the difference of the spectra were (2^12) and (2^12 + 2^13)
• black PolA has counts where the difference was 2^12
  
• They aren't exactly at 2^N since the noise can change between 1 spectra and the next 

Running the diagnostics:

• The test compares the first box used, with all of the other boxes in the group
• If the first box  has trouble, then all tests fail (since it compares all the other boxes with the wrong data.
• Boxes are excluded so that another box becomes the reference for all the others.
  

01oct12:

• test failed at first when b0s0g0,and b0s1g0 were included
• Group 1 ran ok
• reduced clock to 163.84 MHz. At first it failed after 2mb. After restarting it ran without problems.
  

02oct12

• ran diagnostics with:
• 172.032 MHz bw, 86.. ran with hardly any errors
• 178MHz., bw 89 . this is the timing report max frequency.
  
• ran 18Mb with a few errors
• failed immediately
• then ran for 46 Mb with no errors.
• 180MHz, bw:90 . this is beyond the simulation max allowable frequency.
• fails immediately
• 172.032 MHz, bw 86
• 12mb ok except for about 15 errors on b5s1g1
• b1s1g0 fails immediately
• 20 mb. b5s0g0 1 error
• 12 mb g0 no errors
• 12 mb g1 no errors
• So today it ran more reliably at 172.032 clock and bw=86 than it did on 01 oct12
  

home_~phil