Thanks for the follow-up. This is my idea as well: feed a CW signal into a variable attenuator, split the signal into two paths, one to the receiver and another to a detector. Finally, control the variable attenuator with the error signal from the detector, close the control loop, and you get leveled sinewave. So I decided to do a quick investigation on its feasibility, and it was how I found your post.
The challenge is whether the design can be accurate enough to match an oscilloscope calibrator. I used the classic Tektronix SG504 as a reference, with the following specs (the original specs are all in voltages, here I converted voltages to power for easy comparison).
Frequency Reference: 50 KHz, 6 MHz
Frequency Range: 245 MHz to 1050 MHz
Leveled Amplitude: 0.5 Vpp (3.9 -2 dBm) to 4 Vpp (5 Vpp, 24 18 dBm @ VSWR 1.2:1, 0-35 C)
Amplitude Accuracy: 3% at reference (0.256 dB)
Amplitude Flatness: 4% to reference (0.340 dB)
...so Worst Case Absolute Accuracy is 7.12% (0.597 dB)
Harmonic Content: 25 dBc (2nd), 60 dBc (3rd and more).
As you can see, distortion shouldn't be a big issue if Tektronix could do it with an external diode detector from a signal source with only 25 dBc harmonic suppression. The main problem is the razor-thin tolerance for both absolute accuracy and flatness. Leveling at a power level as high as
24 dBm 18 dBm can also be problematic.
Being a novice, I have absolutely no idea about what kind of performance is achievable in a discrete design, and even if I do I currently have no equipment to characterize them. But I wanted to start plugging some numbers in and see what I get. So I was looking into RFICs with guaranteed specifications, log amps in particular. I found most log amps are optimized for use in RF receivers, they have the best accuracy and linearity around -20 dBm to 0 dBm. 0.25 dB or higher accuracy is achievable under favorable conditions. Many have RMS measurements as well, so no need to worry about distortion. But above 0 dBm they can't be trusted, there will be significant errors like 3 dB at 10 dBm. So it's necessary to put a 20 dB attenuator before the log amp to knock the power down. Then the attenuator's tolerance also becomes the problem, especially a large one like 20 dB. The best SMD RF attenuator from Mini-Circuits has 0.15 dB flatness under 1 GHz, but the reference accuracy is 0.4 dB max. Finally there's also the tolerance of the variable attenuator, these errors quickly add up. Of course, actual performance is likely better than guaranteed specs. The errors can also be removed with manual calibration. But proper characterization remains a problem.
What kind of performance is achievable in a discrete design? Any input is greatly appreciated.
BTW, the diode detectors used in SG504 and the low-frequency SG503 were all successfully reimplemented. These designs can be used as a starting point.
A replacement levelling head for the Tektronix SG504
http://www.perdrix.co.uk/SG504Head/Design and Implementation of a Substitute for the SG503 Peak-to-peak Detector (Part no. 155-0107-00)
https://www.eevblog.com/forum/projects/help-needed-calculating-trace-impedance/?action=dlattach;attach=1172968