Having the 8568B basically working again is great. Not so great were the correction coefficients, all seems to be a bit out of alignment.
Before alignment…. quite some deviations at 100 kHz, at the lower frequencies, and so on.
This is the order of the coefficients:
First, checked the log amp circuits, and all seems fine. No adjustment made.
The, some small tweeking of the A4A5 assy that controls the step gains.
Then, the major part, the xtal filters of board A4A7:
Each of its crystals can be switched to 10-30-100-300-1kHz bandwidth, at virtually constant gain.
To do the alignment properly, each xtal has to be tuned separately, and some bypass networks are needed.
The bypass network, a 2.8 Ohms resistor in series with a 47n ceramic cap – just use an old ATX power supply connector to get some suitable contacts. These come free of charge, whereas HP used to charge a dollar each, or more. Note that I did not have any 2.8 Ohms resistors at hand, so I used two 5R6 in parallel.
This is how it looks during the adjustment:
When all is done, the bandpass responses should be symmetrical, which they are, and the amplitude flat for all bandwidth. After running the calibration, these coefficients were found:
As you can see, it all looks great, except for the 10 Hz bandwidth. Carefully re-checked the aligment – 30-100-300-1000 is perfectly flat, but whatever I do, it seems A4A7 has extra loss when set to 10 Hz bandwidth. In priciple such situation can happen with misaligned crystals, or with some low-frequency issue of one of the amplifiers, which are rather unlikely. Maybe just some aged xtals? Will give it another try later, with flatness checks with some of the xtals bypassed, to see if the issue is caused by any particular of the stages. Found one of the 2.2 µF tantalum caps on the A4A7 to have high leakage current, replaced it, to no effect.
For now, it is good enough – specification at 10 Hz more relaxed anyway, and I never do 10 Hz bandwidth measurements without correction enabled, just considering that any modern analyzer basically relies on a good number of calibration and correction data stored somewhere in the instrument, and applied to all measurements.