Spectral output vs Ashift parameter
05jun08
The pdev
spectrometer ashift parameter (see layout of spectrometer
scaling) upshifts the 40 bit accumulator before packing the
data for output (it is not used for time domain data). The output
packing can be 32,16, or 8 bits. With 32 bit data the ashift parameter
can be left at 0 (since you can't dump very fast anyway). For 16 and 8
bit data the value of ashift is more critical. You want to make sure
that the spectra ends up in the output bits and that you have a few
bits on the noise. The spectral bandpass will use some of the available
bits for the spectral shape of the filters.
On 05mar08 data was taken to see how the output
varies with ashift. The data taking setup was:
- 172 MHz sampling, 2048 channels, dump time= 1 millisecond
- pshift =0x7fd
- dshift=0
- Full stokes.
- packing=16 bits
- ashift: 13,12,11,10.
- telescope sitting at az,za. Only beam 0 was used.
The plots show the results for
various values of ashift (.ps) (.pdf):
The top plot shows the spectrum averaged for 1 second. The bottom plot
is the channel rms vs frequency for the 1 millisecond integration. Each
color is a different Ashift value.
- Pag1,2: pol A. Page 1 is 1450 Mhz, page 2 is 1300 Mhz.
- Top is the 1 millisecond spectra averaged to 1 second.
- Bottom is the rm along a channel for the 1 millisecond spectra.
- Pag3,4: Stokes U 1450 and 1300 Mhz
- Top is the 1 millisecond stokes U averaged to 1 second.
- Bottom is the rms along a channel for the stokesU.
- Pag5,6: Stokes V 1450 and 1300 Mhz
- Top is the 1 millisecond stokes V averaged to 1 second.
- Bottom is the rms along a channel for stokesV.
Conclusions:
- The default ashift for 1 millisecond, 2048 channels is the black
ashift spectra.
- Other configurations (sample time, lenfft) will be scaled by the
square root of the noise power in a channel relative to the
standard config (2048 channels, 1 ms).
- The rms for stokes U, V are about a factor of 2 less than
polA,polB.
- This is from the factor of 2 that jeff outputs. 2*polA,
2*polB, stokesU, stokesV.
- I've modified the default scaling so that the output values for
stokes U,V are scaled up by a factor of 2. There rms vs frequency
should now look the same as the polA,polB.
- This used beam 0 where polB has some real problems.. So the U,V
spectra are probably not typical. I need to look at all 7 beams.
processing: x101/080605/chklevels.pro
<-
page
up
home_~phil