When processing the clp data (from the usrp's) the  steps are:

• synchronize to the start of a clp ipp .
• zero the accumulation buffer.
  
• Input 1 ipp of data.
  
• for 10 ms ipp with 25MHz bw, this is 250000 complex samples
• remove dc
• remove rfi
• for each height:
• decode, compute power, accumulate.

    Ideally you would like the computation of DC to cover a period longer than any oscillations in the data (say 16ms for 60hz). This could be done by preprocessing the data:

• read in a few seconds of data, and use this to compute dc, and then use this for a long stretch of data.
• This assumes that DC will not change with time (I guess that is actually the definition...)

• 02aug16 world day data
• carriage house -> usrp
• 10 ms ipp, 440 usec rf pulse, 25 MHz bandwidth, complex sample every 40 nanoseconds
• The results are in usrp sample units:
• 16 bit sample mode
• but the height data typically has an rms of  200 to 300 samples.
  

    The test consisted of:

• input 10 seconds of data (250,000,000 complex samples)
• for each 10 ms ipp, compute DC  by:
•  

  region of ipp for DC computation	samples in region	time span useconds	Notes
  tx samples	11000	440.	Code has its own dc component
  blanked samples	4700	188	telescope signal blanked
  all height samples	238700	9548	from 80 km to end of ipp. includes the cal
  heights 700-800 km	16667	666

  
  

The plots show the DC computed in each region for the 1000 ipps (.ps) (.pdf)

• Black is the real sampler, red is the imaginary sampler
• for each region the mean and rms of the 1000 dc computations (1 per ipp) is shown
  
• Top frame: DC from xmiter pulses (11000 samples)
  
• the code has it's own DC (depending on where it is sampled)
• the large difference in real, imaginary is from the phasing of the code through the xmter.
• 2nd frame: DC from blanking region (4700 samples)
• the telescope signal is blanked here.
• the oscillation is probably aliased sampling of 60 hz (once per 10 milliseconds).
• 3rd frame: DC from 1st data height (80km) to end of ipp. 238700 samples
• 4th frame: DC from data between 700 and 800 km (16667 samples)

The table shows the results:

summary

• The most accurate measure of the data was probably the all hght data:
• 238700 samples.  mean: -.71, -.52
• The blanked region did a pretty good job of measuring DC: -.77, -.46  with a small rms
• The code is currently using the 238700 all heights region.
• I'm a little nervous about using the blanked region.
• Users can change the blanking region from the thumbwheel switches in the control room
• They could make this blanking region very short (if they wanted to).
  

processing: x101/atm/clpgpu/test/testDC.pro