mock spectrometer minimum integration times

Mock fft length and the minimum allowable integration times.

• terminology
• fftlen - number of channels in your spectrum
• smpTm= sample time  for bandwidth being sampled = 1/bandwidth. they are complex samples.
  
• bw = bandwidth 1/smpTm
  
• output dump: integrate N ffts and then output to disc. Make include 1 to 4 polarizations.
  
• NAccumSpc: the number of spectra to average together
  
• totalIntegrationTime= time to acquire and add N spectra
  
• Example computation fftDumpTime
  
• If your fflten is 1024 and your sample rate is 100 Mhz, then:
• 1 fft takes 1024*.01 usec = 10.24 usecs. 
• The total integration time must then be a multiple of 10.24 usec.
• If you accumulate 8 ffts (NAccumSpc then the total integration time for each output spectra is 10.24usecs*8=81.92 Usecs
  
• The i/o rate
  
• Let Npols be the number of polarizations being ouput (1,2,4)
• let BytesPerChn be the bytes in 1 spectral channel (2,4) (16 or 32 bits)
  
• iorate= bytes1output/secs1output
• total bytes 1 output dump= fftlen *BytesPerChn*Npols
• secsFor1Ouput=fftlen*smpTm*naccumSpc
• iorate=(fftlen*BytesPerChn)/(fftlen*smpTm*naccumSpc) or
• iorate= BytesPerChn/(smpTm*NAccumSpc) or
• iorate=BytesPerChn*BandWidth/NAccumSpc
  
• So the i/o rate is independent of the number of channels.

constraints imposed by the mock spectrometer:

• The mock spectrometer must always average at least 4 spectra
• The polyphase filter used has an overlap by 4.
• This limits the minimum allowable output times.
• The output time now becomes a function of the fftlen (since you must always acquire at least 4).
  
• The mock spectrometer can never average more than 65535 spectra
• the register to count the accumulations is only 16 bits
• The maximum i/o rate from mock to disc is output 70 Mbytes/second
• This is a gigibit ethernet between each mock spectrometer and  their file server.
• Each mock can output 1 or two complete bandwidths
• Single pixel observing uses a single band
• This limits a single mock to 172 Mhz bw
  
• ALFA observing uses both bands
• this can output 2 172Mhz bandwidths so a single mock can cover the 300 Mhz Bw of an alfa beam.
  

Summary:

• The mock spectrometer is an fft based machine. It must always acquire (avg) an integral number of spectra
• A mock spectrometer must always acquire (average) at least 4 spectra (because of polyphase filter band overlap).
• A mock spectrometer can never average more than 65535 spectra before outputing.
  
• The maximum i/o rate for a single mock is 70 Mbytes/secs
• The iorate= BytesPerSpectralChan*Bandwidth/NaccumSpc
  

Plotting minimum dump times:

• These plots are for 16 bit output spectra.
  
• Page 1 shows single pixel observations (1 mock band per mock board).
• Page 2 shows observations using alfa (2 mock bands per mock board).
• The vertical axis is minimum allowable dump time
• The horizontal axis is the bandwidth used
• Each fftlen is plotted in a separate color
• The * of each line are the allowable mock bandwidths with a 172 Mhz clock
• the lines show bandwidth divisors of 1..20
• The 3 frames on each page are:
• Top:  polarization adding
• Middle: polA, polB output separately
• Bottom: Full stokes (a,b,U,V)

Table of minimum dump times (for some configurations)

• Each table has the values for a single mock box. We have 7 x 2 sets of these boxes.
  
• Sum PolA and B. 16 bit fast dump data (pulsar observing)
• the rightmost columns show min time for single pixel and alfa
  
• Output PolA and B separately. 16 bit fast dump data.
• the rightmost columns show min time for single pixel and alfa
• Output Full Stokes. 16 bit fast dump data.
• the rightmost columns show min time for single pixel and alfa

<- page up to pdev hardware.
home_~phil