This is a quick update on the Zpectrometer after our first stint of testing 
in the GB lab with the Ka-band frontend.  We set up and ran internal tests 
on Sunday the 17th.  On Monday we connected to the front end and got 
"first light" correlation functions that verify the system's basic 
operation.  I attach a probably indecipherable copy of the result; but 
it's a good result!  We spent Tuesday starting to chase down a problem 
I'll mention in a moment and in working with the software group to 
generate a complete set of FITS files to use in data flow testing.  The 
Zpectrometer went into the anechoic chamber for RFI testing Wednesday 
morning and, having passed, was back in the lab and back on the network by 
lunch.  

We have three hardware problems to solve, one of which was planned, one 
semi-expected, and one that was unexpected and serious.

The planned task is to flatten the gain across the four sub-bands now that 
we have measurements of the power levels as a function of frequency from 
the front end as seen through the channelizing downconverter.  Our 
downconverter is pretty flat, but it seemed sensible to get measurements 
at its outputs rather than try to guess at what we'd get.  We'll order 
gain equalizers early next week, and I expect that this will be sorted out 
by our commissioning time in mid-November.  If they don't arrive some 
parts of the band will be noisier than others.

The semi-expected problem is getting the power levels for our phase 
calibration tones within good ranges across the band.  We see a fair 
amount of structure across the Ka-band, some broad stuff and some narrow 
dropouts.  I think we'll be able to clear this up in our next lab visit in 
early October.  We need to clear it up to get spectra, but we can think of 
schemes that aren't fully automatic (changing pads around, etc.) for the 
first little while.

The unexpected problem is serious: we see the equivalent of a constant ~20 
K input noise imbalance, when we should see something much closer to zero 
(the whole point of the correlation receiver architecture).  The total 
power in the offset is 1/f or drifty for timescales longer than 3 seconds, 
making the receiver look like a moderately stable total power system with 
a 20 K receiver temperature.  We have pinned the noise source down to 
somewhere in the front end with pretty high certainty.  It is likely that 
it is internally generated noise that is modulated by a ground loop 
containing phase-switch driver currents.  Although we're no longer in the 
lab at GB, miracles of modern technology etc. means that the Zpectrometer 
is still accessible to us over the internet.  We will run tests today and 
tomorrow to try to isolate the problem further, with Galen in the lab and 
us on the computer here, connected by phone.

Left unfixed, this offset would cause baseline structure and a loss of 
sensitivity.  If the same effect is present for the Spectrometer it would 
account for at least much of the Spectrometer baseline problems.  (We're 
trying to find someone to look at Ka-band Spectrometer data to check 
this.)  There are work-arounds if we can't find and fix the offset, but 
they will involve some loss of resolution and sensitivity -- and then we'd 
have a total power receiver with no advantage from the correlation 
architecture!  I'm pretty confident we'll find the source of the problem, 
though, and we have potential fixes for some of the possibilities even 
now.

The software situation looks good.  Although we'd planned to run with our 
existing software for the first year or so, the GB software group felt 
that the interactions between the Zpectrometer and telescope and observer 
would be better handled by adding the Zpectrometer to the general system. 
We have provided one of our instrument computers so the manager has 
something to talk to, and the software group ran a full observation using 
that and other simulators for some of the GB systems.  Still some bugs, of 
course, but nothing fundamental so far and we're at a stage now where 
software work on data processing can proceed independently of the 
hardware.