Well, I copied the circuit of generic instrumentation amplifier and used an opamp that is available locally - OPA4354AID. My knowledge of op amps is very limited so don't laugh at my attempt. Here is my thought process:
The opamp is rail to rail ( 0.1V to rail). I used Vs/2 for ground to avoid going near the rails. Because we are measuring power supply that will often go above 2.5V, so That's the idea of C1 and C2 - to remove DC offset. What I forgot is that there will be a high votage spike if you connect it to already running power supply. So a zenner protection on the input would be needed.
This op amp has relatively high DC offset - 2 to 8 mV. C3 should remove the DC offset. The AC coupled probe would probably do the same job.
If you don't use an oscilloscope probe you could use 50 or 75 ohm coaxial cable with proper termination (R8 and another one at the scope. I forgot to put a note that when using
terminated coaxial cable
the gain should be 2 times higher because of the divider. For example output stage could have gain 4 instead of 2.
Matching resistors will be pain in the a$$. An option is to use 0.1% resistors or match them manually. Here I can get 100 smd resistors for about 20 cents. I bet that there will be few matched in every 100.
Converting C1 and C2 to 20 Mhz low pass LC filter could do a better job. It also means that you don't have to use your scope BW limit.
Ok, now ruin all my dreams and tell me what's wrong with the schematic (there should be something)
... well I just found something. The input stage could have DC offset 8 mV (worst case) (if we assume that it's always the same polarity). I'm not sure if it's possible to put capacitors between input and output stages to remove DC offset? Will the output stage low input resistance cause troubles with capacitors between the stages?
----
BTW I tried the math function on rigol DS2072A. I have 20-30 mV on both probes - absolutely the same even with the 50 ohm termination, regardless of power supply state - on or off. If I try to reduce range to 0.5 or 1 mV channel 1 and 2 waveforms are clipped and math is not correct. So I can't use that math to measure power supply ripple/noise. I tried to track it - I stopped lights, remote controlled stuff, soldering iron... same. The only thing I couldn't stop was the UPS which is probably the cause for this problem. A probe with a spring shows 2-3 mV ripple at 10mA and 10 mV ripple at higher currents (30-50 mA).