From harris Mon May 5 11:43:05 2007 Date: Mon, 7 May 2007 11:43:27 -0400 (EDT) From: Andrew Harris To: Brian Mason Subject: CCB Allan var Hi Brian -- Ok, here are the Allan variance plots for the CCB, now with the correct phasing. And in postscript. I've binned the time series in each plot by 20 to cut down the file size, but used the full set of data for the Allan variance calculation. In summary: no big changes from what I'd said at the telecon. Any differenced data have a transition from white noise to drift at 10-15 seconds. Combining detectors to make a correlation receiver neither improves nor degrades the stability. This makes sense if the instabilities come from the front-end imbalance. And channels 2 and 6 (my test channels) do show similar time series data shapes, but with inverted senses. Summing the channels pushes the transition time out to 50 sec or so -- but that phasing also removes the signal, so it's not too useful. But it does show that the instabilities are equivalent to an astronomical signal. Since the correlation receiver setup is not worse than the electronic beamswitching setup, I'd say we're all seeing the same thing. Ron's comment about the spectra having ideal noise is a little misleading for a comparison with the Zpectrometer results: it only applies to small sections of the baseline. We also see better stability at high lags than low -- my focus has been on understanding the origin of the worst of the structure, which causes largish-scale bumps. The main difference may well be the load differential stability, as you've pointed out in the past -- and when we weren't in the lab, we tested on a pretty unstable sky. To go into details, for the figures in the attached file: Page 1: Channel 2, phase state 1. No differencing. S(f)~f^{+0.7} for short times, drifts after ~1 s. Page 2: Channel 2, difference of phase states 1 and 2. White noise at short times, variance minimum at 10 s. Page 3: Channel 2, full phase differences. White noise at short times, minimum at 17 s. Page 4: Channel 6 (other hybrid output from Channel 2 data), full phase difference. Very similar to results for Channel 2, but structure in time series is inverted. avar[1] is the variance at the shortest timescale; it's 1.8 here compared with 2.6 for Channel 2. I'd attribute that to differences in diode responsivity and video amplifier gains. Page 5: Differenced hybrid outputs (Channels 2 and 6) as one would do for a correlation receiver. No surprise, it looks like the Channel 2 and Channel 6 data. Corner time 12 sec. Page 6: Differenced hybrid outputs, now weighting the channel data by the square roots of the avar[1] to account for the backend gain differences. No change. Page 7: Sum of the channel outputs, equal weighting on the channel outputs. The shapes in Channels 2 and 6 cancel well, leaving some residual structure with a minimum time ~50 s. In this mode the source would also cancel -- so at that point, it's more efficient to go to zero observing time! [ Part 2, "" Application/POSTSCRIPT 938KB. ] [ Unable to print this part. ]