Date: Wed, 9 Nov 2005 16:46:10 -0700 (MST) From: Walter Brisken To: evlatests@parallax Subject: Sampler histograms I have modified one of Mike Revnell's test programs to request a series of data chunks from the evla deformatters (D351s). From these data I've created histograms showing the frequency of occurance of the various sampler levels. I took no real care to make sure attenuator levels were set to be sane, but in all cases I got reasonable looking RMS values. The resultant histograms are not nice smooth histograms but rather show that level frequency alternates with bin, indicating that the least significant bit (and to a much less severe level the second least significant bit) seems biased. In the case of IFs 16c and 16d the effect is fairly strong: adjacent bins with ratios of nearly 2:1. Similar phenomena were seen as early as June 6, 2005, but this is the first time I've systematicly looked at all the IFs. I have two plots, source code, and this email message stored in my new EVLA web repository : http://www.aoc.nrao.edu/~wbrisken/evla/ I'll keep things organized in directories named after the date of the corresponding evlatests report. -W Date: Wed, 9 Nov 2005 20:06:26 -0700 (MST) From: Walter Brisken To: evlatests@parallax Subject: Sampler statistics with fully attenuated T304s A few of us were wondering at one point what the sampler output would be for a zero input level. The closest I could come was to set both the RF and OUTPUT attenuators in the t304s to 31dB and place an upper limit on the self noise of the samplers. Here are the histograms in text form for the 8 samplers: level 14A 14B 14C 14D 16A 16B 16C 16D -6 0 0 0 0 0 0 0 -5 0 0 0 0 0 1 43248 -4 707 0 0 0 0 175431 934787 -3 985383 0 0 0 2 873047 70541 -2 62486 2640 0 12740 884886 97 0 -1 0 1045355 67102 1035809 163688 0 0 0 0 581 772727 27 0 0 0 1 0 0 208522 0 0 0 0 2 0 0 225 0 0 0 0 3 0 0 0 0 0 0 0 Ant 16 T304A seems dead. Note that for 14C all values of L304.OBSERVE_BAND_SELECT (except 5, which was used in the above) gave a wider distribution so it is likely that the width is due to some signal still getting through the attenuators. Aside from 14C (and not commenting about 16A) a limit of less than 1 sampler unit of noise is incurred in the sampling. -Walter