One of the products that have been in the HP/Agilent/Keysight catalog for 3 or 4 decades, or more, the 11708A reference attenuator. Specified at +-0.05 dB, it is a remarkably simple device – it just provides 1:1000 attenuation, chiefly, 30 dB. It’s main application is the calibration of 8484A power sensors, from a 1 mW source – the 8484A needs a 1 µW reference level.
Unfortunately, it doesn’t come cheap, when ordered from Keysight today, at least for a hobbyist’s budget. So I got mine used, aged (30 years?), and at a minor fraction of the cost.
Before using it for a considerable number of power measurements, it is a good idea to confirm it’s performance. Measuring attenuation to +-0.05 or better is no easy tasks, but fortunately enough, a tractable one, with a 8642A signal generator, and a Micro-Tel 1295 precision attenuation measurement receiver. The Micro-Tel is specified to +-0.02 dB, plus +-0.02 dB for each 10 dB, say, +-0.08 dB. Actual performance, of a well-calibrated and well-heated-up unit is considerably better, but only in combination of other high quality components, like, a stable source (the 8642A has virtually no measurable drift), and, good test cables (using Suhner Sucoflex).
The Micro-Tel 1295 employs IF substitution to determine attenuation, and the IF attenuator works in 10 dB steps. Therefore, for best accuracy, the tests should be done at various power levels, to use various combinations of x0 dB segments, of the IF attenuator.
The results, quite remarkable!
One thing to consider for the test – the input and output matching losses. Neiter the source nor the cable/receiver are perfect 50 Ohm terminations – but the 6 dB pads will ensure only very minor losses. Obviously, you need to use high quality pads here, specified to small return loss, 18 GHz parts preferred.
First step – reference measurement is taken without the attenuator-under-test:
Second step – actual measurement is taken with the attenuator-under-test installed between the two 6 dB pads:
Before the start – best to check reproducibility and repeatability. With good cables and hardware, +-0.005 to +0.01 is achievable with the current setup.
Well, let’s say, chances are that the 11708A is +-0.02 off its nominal value, most likely, it didn’t drift at all over the last 30 years.