Concerns have been expressed on whether the 3-bit system is useful for C, X, and Ku bands, given the modest bandwidth increase, and (particularly for C-band) the high level of RFI. There are concerns that the latter could prevent the system from settling on the right power level for optimum feeding of the samplers. To get some quantitative measures of this, Emmanuel set up an experiment where 3C147, and an adjacent blank field, were alternatively observed with the 3 bit samplers at these three bands. To ensure meaningful comparison, the 8-bit samplers were arranged to observe at the same time. The data were taken on Feb 25, at about 5PM. The array was in D configuration. Full results of this test will be presented later (after Emmanuel finishes his technical reviews!). Here I give the basic results for C-band only. Data quality was just superb. Virtually no editing beyond the system flags was required. Delays and bandpasses were removed following standard procedures. The gains were set using the 3C147 flux scale. The SY tables were not used (and in this case, not needed -- no gain changes occured within the short window of time over which these data were taken). The 3-bit system gains were set in the usual way -- all data prior to the 'gain change' (easily seen in the visibilities) were flagged. There is no indication from the visibilities whatever that the levels selected by the system are incorrect. Sensitivities for each subband were determined using the histogram plotting program 'UVHGM'. The 'RL' correlator was utilized for this, to ensure minimal effect from any accidental background sources. Tests using other polarization pairs indicated no sensitivity to the particular pair chosen. RFI was *much* less a concern than I had expected. I used SPFLG to review each spectrum, in order to select which channels should be utilized in the histogram. The total number of channels seriously affected by RFI were: a) 207 in the lower half (4.0 -- 6.0 GHz) b) 100 in the upper half (6.0 -- 8.0 GHz) There are a total of 1024 channels in each half, with the resolution utilized in this experiment. Most of the 'lost' channels in the lower half are between 4.0 and 4.2 GHz -- satellite downlinks. The RFI here is not very strong, as we were pointed rather far north. My bet is that in A configuration, we'd likely not see this RFI at all. So, about 15% of the BW is lost. The rms noise values for the 3-bit and 8-bit systems is given below. In subband with RFI, I utilized the 'clean' part of the spectrum. For all others, the value comes from the central 26 channels. The channelwidth is 2 MHz (but was Hanning smoothed), the averaging time 3 seconds. Freq. A1C1 A2C2 B0D0 ---------------------------------- 4040 MHz 98 -- -- RFI dominated 4168 62 RFI free part only 4296 58 4424 54 4552 55 4680 54 4808 53 4936 52 5064 54 5192 53 5320 54 5448 51 5576 49 46 5704 50 45 5832 53 46 5960 52 44 6040 -- 48 43 6168 47 43 6296 46 42 6424 49 43 6552 49 6680 47 6808 45 6936 44 7064 48 7192 45 7320 48 7448 46 7576 49 7704 47 7832 48 7960 49* --------------------------------------------------- * lower half. 20% worse in upper half. Major conclusions are: 1) The 3-bit system appears to set up correctly in C-band. The RFI levels (at least for this experiment) are not high enough to cause obviously bad settings. 2) The ~ 10 -- 15% loss in sensitivity due to the 3-bit samplers is clearly present. 3) The lower half of C-band is about 15% less sensitive than the upper half -- this is also well known, and should be largely addressed by the planned thermal gap retrofit. 4) I estimate that the gain in sensitivity (in continuum) from use of the 3-bit system will be ~20 -- 25% over use of the 8-bit. (This includes the loss of BW due to RFI and the loss from the samplers. It does not include time lost due to setup).