Spectral jumps vs A/D and pshift levels.

08jul08

Introduction

Spectral jumps were seen in data taken back in 18mar08. That data had the rms fixed at 15 and changed pshift as well as fixing pshift and changing the rms. It was taken with a fixed fftlen of 8192. Since then most data has been taken with an rms=30 and pshift set to have 9 downshifts and then alternate shifting every other butterfly stage.
A new set of data was taken with the if2 noise source. The setup was:

down convert if2noise source to 250 MHz.
Set pdev lo's to 175 and 350 MHz to mix to baseband.
Sample a 172.032 MHz bandwidth
Take 120 seconds of data dumping at .5 and .1 seconds (for the shorter fftlens).
Take 6 sets of data. At each set, take data with fftlens = 8192,4096,2048,1024, and 512. fftlengths with even, odd number of butterfly stages will have rms voltage levels that differ by sqrt(2) (since the butterfly down shift changes voltage by factor of 2 and it takes 2 butterfly stages to increase the rms by a factor of 2).

The goal is to find a configuration that minimizes the rms while being free from the spectral jumps seen in the spectra when the a/d rms is low.
The tests done were:

Tests performed
	butterfly downshifts used	a/d rms	Relative power levels of spectra vs fftlen
	butterfly downshifts used	a/d rms	8192	4096	2048	1024	512
0	bandPass shape	80	3.6	7.1	3.6	7.1	3.6
1	9 downshifts then 0,1	30	.50	1	.5	1	.5
2	9 downshifts then 0,1	25	.35	.7	.35	.7	.35
3	9 downshifts then 0,1	20	.22	.44	.22	.44	.22
4	10 downshifts then 0,1	30	.5	.25	.5	.25	.5
5	9 downshifts then 0,1	15	.13	.25	.13	.25	.5

Plots where made to show the results. For the plots:

Colors differentiate between different tests
vertical dashed lines are plotted at powers to see if there are any spectral jumps.
1 set of plots is made for each fftlen.

	8192	4096	2048	1024	512
band pass shape: These plots show the bandpass shape for the various beams and fftlengths. The bandpass with rms=80 is plotted as well as the bandpass with rms=30 counts. You can see that the spectra with rms=30 has more bumps in it than that with 80 counts. I wouldn't expect this to be the case. 4 pages per plot showing: polA_loband, polB_loband, polA_hiband, polB_hiband Each page shows the 7 beams. the brown line is test 0 with rms=80 the black line is test 1 with rms=30. dashed vertical lines are at powers of two showing where jumps might be.	plot	plot	plot	plot	plot
Ratio spectra/spectraRms80: The spectra from each test has been divided by the spectra with rms=80 to get rid of the bandpass shape. 8 pages/plot. page 1,2: polA lowBand page 3,4: polB lowBa page 5,6: polA hiBand page 7,8: polB hiBand The ripples in the band pass get larger as the a/d rms level decreases (the attenuator value increases). rms=30 (black and blue line about the same) rms=25 red, rms=20 green, rms=15 purple have progressively larger ripple. I would have expected the ripple to go down as the gain went down (since attenuator value goes up). This must mean that the reflection is not passing through the attenuator. The ripple spacing is about 10.6 Mhz or about 10 meters with the velocity set to .68 c.	plot	plot	plot	plot	plot
Ratio spectra/spectraRms80 Blowup: The previous set of plots has been blown up about fftlen/4 and fftlen/2 to see the possible jumps. The left column shows fftlen/4, the right column shows fftlen/2 Most spectra have steps on the positive side of fftlen/2 (about dc).	plot	plot	plot	plot	plot

Conclusions:

still see the spectral jumps even with a/d rms set to 30.
ripples in the bandpass increase as the a/d rms decreases..

processing: x101/080708/nsetst.pro

<- page up
home_~phil