That was almost nostalgic for me particularly. I was part of the group that did the LMH67xx series of parts. I personally didn't design the LMH6702, but I was next cubicle over from the guy who did.
Excellent write up. I'm also a software guy that experiments with hardware, so I love seeing these type of projects. How much time was spent on R&D for this? I'm sure there are plenty of reasons, but it's insane to me that something like this would cost $3k.
3K in the RF world is cheap... I've watched as more than that in chips fried with a bad reflow profile, and that was with some very very cost-efficent people.
Instrumentation amps are meant primarily for amplifying low voltage rather than high-bandwidth. What you're looking for is a "fully differential" amplifier.
He apparently actually built a board using an LMH6702 (roughly $3 for 1.7GHz), but then never mentions anything about it.
And, what's with the downvotes on my question? If the answer is so obvious, what's the answer oh-so-brilliant-ones?
Yes, instrumention amps are often used for low-voltage amplification...primarily because the topology enables the elimination of impedance-matched input termination with the distinct benefit of minimizing parasitic load on the UUT. Apparently, readily-available integrated instrumentation amps on the market are designed for relatively low-bandwidth applications as a quick DigiKey parametric search suggests, but the canonical topology itself isn't fundamentally bandwidth-constrained...which is part of the reason why I pointed to max GBW results from the search and left it at that.
No, that's not what I'd be looking for, especially in response to your question with the specific constraint of "an integrated version of a differential amp"; given the sensitivity of the in-circuit measurement application being considered, this effectively suggests integrated instrumentation amp. With your suggestion of what I should be looking for, how would you go about input termination without loading the UUT to the detriment of SI?
The author does indeed briefly discuss his prototypes and even released his Eagle design on GitHub[1] with still more notes.
For the record, I didn't downvote on first response, but your "oh-so-brillant-ones" snark may have earned one.
I actually did originally go back to the GitHub repo, but I missed this in the README:
"So far I have only built the B variant of the probe, that is, the probe with separate OP amps for the inputs."
So, that's why he doesn't discuss the others. He never built them.
> With your suggestion of what I should be looking for, how would you go about input termination without loading the UUT to the detriment of SI?
I would almost certainly want to grab the LMH6552 which is a fully-differential. It's input bias current isn't dramatically worse (110uA (about 10K Ohm) vs 34uA (about 30K Ohm)) than the LMH6702 that he used and likely wins by being matched/same temperature because same die/no PCB routing mismatch/etc.
