Hello all -- On Thursday, September 21st 2006, Galen and I ran a series of tests to identify the source of the ~20 K input noise offset from the Ka-band receiver. We have isolated the problem to phase switch pickup on the bias of the first Ka-band amplifier in the cryostat. The most likely source of the pickup is coupling through a common power supply. Isolating the power supplies for the phase switch driver and changing the phase switching scheme for the Zpectrometer should suppress the imbalance to a reasonable level for most of the scientific observations. In more detail on the imbalance: We started by reproducing the imbalance signal we'd seen in the initial Zpectrometer tests by observing a room temperature load over both horns. The difference signal from this should be 0 K but is about 700 counts in the Zpectrometer's total power channel (adc229; 14.eps). The same correlator output is about 9700 counts when one horn is terminated at 80 K, so the equivalent temperature is about 20 K. We next turned off the first LNA stages, and the imbalance disappeared, with a CCF mean ~-4 and nearly featureless (18.eps). The pattern is essentially the same with the phase switch reference cable disconnected from the Zpectrometer (42.eps) and with the phase switch turned off at the control box on the receiver's side (50.eps). With the amplifiers back on, we reduced the fourth stage gain and saw the imbalance drop (34.eps). These two tests isolated the problem as coming from the first LNA, not further downstream in the system (noise from the common downconversion LO, for instance). It's not hard to get this size imbalance: with a gain of 70 dB in the amplifiers, a 20 K imbalance corresponds to a 2 microK change with phase switch state, or a gain change of 10^-7 amplifier gain change for a 20 K amplifier. In a separate set of tests that sampled the stability, we found that the imbalance is very stable on average, varying by half a percent over 80 minutes. The problem is that the noise properties are not good: white for only a few seconds, with 1/f and drift noise contributing increasingly to 30 or 40 seconds, and drift dominating after that (L229_060921.eps). We made a set of measurements to separate supply-line pickup from ground loops. The phase switch driver card contains two out-of-phase drivers with one driving the phase switch and the other a nearby ground point. This approximately balances the current flowing into the chassis. We tried connecting and disconnecting the second driver from the ground and tried different grounding points, on the card and on the frame. There is a very small effect, a few counts on 700 total, showing that ground loops are a small effect for this problem. Turning off the phase switch in the Zpectrometer's arm and switching the other arm gives a slightly larger imbalance signal with more structure at high lags. The conclusion from these measurements is that ground loops are present at a low level, but the bulk of the imbalance comes from a signal on the power supply itself. As I understand it, the amplifiers and LNAs share a common power supply. There are two fixes for this problem, and both should be implemented. 1) The Zpectrometer should be changed to drive two phase switches in quadrature rather than only one phase switch, as it did for these measurements. In corresponding about offsets for the CCB, Brian Mason explained his experiments with CCB switching. He finds about an order of magnitude reduction in the offset with the quadrature switching compared with a single switch. The rejection will depend strongly on the phase switching and the demodulation waveforms (especially the phase switch second harmonic content), so it is difficult to predict the offset reduction from this scheme. Rejection of 10-20 dB seems likely, but we need to check this experimentally next week. 2) The LNA and phase switch drive power supplies should be isolated from each other. It is probably best to isolate the phase switch drive from everything else, since we will otherwise see pickup at lower levels from components other than the first LNAs. I think I understand the observational consequences of the noise imbalance. Someone should check my logic, though. If we assume an order of magnitude imbalance reduction by going to quadrature switching, then we'd have a 2 K imbalance that integrates down for 3 seconds until white noise takes over (ignoring drifts for the moment). This sets a noise floor of a \Delta T_rms = 2 K/sqrt(3 sec * 16 MHz) = 0.3 mK rms in a single Zpectrometer channel. Drifts will increase this noise floor on the short term but should integrate down at some level. A noise floor of 0.3 mK is uncomfortably close to the line strengths we need to measure, 1 mK or lower. With a 50 K system temperature, in theory it means we will hit the noise floor after 31 minutes -- in practice, drifts will increase the noise but will kind of integrate down, allowing us to creep down toward this limit over hours. If the little analysis above is correct, it means that isolating the phase switch power supply is essential. It's also the right thing to do. This analysis also applies to Spectrometer data, which probably uses the quadrature scheme already. We do need someone to look at the imbalance level of the un-nodded Spectrometer data before we draw a conclusion here, though. *************** Notes on the "spike" in lag number 96 as a reminder to keep track of it -- it's higher order pickup somewhere... The spike's amplitude dropped faster than total power with decreasing LNA gain (14.eps vs. 34.eps). The spike changes polarity and increases in amplitude when the phase switches are turned off at the box on the Rx side (50.eps), although the CCF otherwise shows no sign of signal.