Would you actually put two thermistors into some sort of bridge, such that changes in the ambient temp would not influence the reading?  At a place I worked in the mid-1960s, they made their own matched transistor pairs, thermally-coupled and potted to get around that sort of thing.

Hard to know without some experimentation, but maybe even some slow chopping, with a reed relay, could help, for instance measure for 3 seconds, and then disconnect the source for 3 seconds and so on. Wouldn't be a very fast way of getting a measurement though : ( I don't know how the existing sensors do it.

Many 'true r.m.s.' detectors in legacy audio signal-processing gear paired incandescent light bulbs with either CdS sensors or, later, silicon solar cells, either of which might (or might not!) exhibit flatter temperature response than a thermistor, which, after all, is a temperature-sensitive device by design.  But whatever is used for conversion, the desired r.m.s. waveform response may be a trusted issue, but the linearity of conversion over a 20dB or better range might require some additional nonlinear network or a lookup table to get the numbers to line up.  A daunting task for sure. 

So far the plan is a dual RMS converter setup: an AD637 from 10Hz to 50KHz, an AD8361+VCA824 from 20KHz to15MHz. In the crossover region a "track the highest" analog stage will combine the high and low frequency path. 

I've seen separate DC and AC path is several professional designs, never one that combined two frequency ranges. Have you seen anything like that?

Sounds like a solid plan, actually.  And, no, I've never encountered anything like that in a measurement application.  Be sure to post your progress!

I think you can do it like this but with some v. tricky maths needed if it's all to be analogue. If you don't mind using  a low speed adc and a micro to drive the meter, not so bad.

(the tricky maths is to take the square root of the sum of the RMS of the hp and lp signals.

MK

I am very tempted to throw in an ADS7028 to compare, once I start testing. The benefit of this part it has all the rms measurement core needed for the digital implementation. Analog in - SPI out the calculated rms value. 

As far as frequency range overlap, I'm planning a wider region to overlap - in other words, in the crossover regions of both converters expected to provide accurate, thus identical values. For example, the 20..50KHz band to be served by both converters with 1% accuracy. 

Did it, see blog.

community.element14.com/.../a-really-simple-hf-true-rms-detector

MK

The exit criteria is different in my case: I'm to verify the rebuilt voltmeter can met or exceeded the original 3400A specs.

 ggabe  Your HP3400A is an old-school analog meter. I personally have a FLUKE 113 which does "true RMS" I got mine in a pawn shop for about 50 bucks. but if you like new Lowes (Box Store) is about 150 bucks. Now it's time to ponder Fix or Replace. If fix Did you check out the parts available from HP ops? Agilent or now Keysite??