I need a very low noise power supply and ADC reference voltage for an 24 bit ADC measurement (I need noise level less than 5 uV peak to peak on both power supply and the ADC reference voltage. Supply current can be around 100mA)
I would like the ask the method mentioned in this link
Shortly it says adding 3 LDO output with an op-amp then dividing by 3, it reduces the effective noise on the final output. Will it work in your opinion?
What other methods do you advice?
That technique looks quite silly actually. The best one can hope or is an improvement of sqrt(3) at best for a lot of work. And he is using chopper stabilized op-amps which can leak tones into the ADC.
Good low noise design requires accurate analysis of all noise sources. I scanned the paper and saw no mention of thermal/Johnston noise as a bare minimum.
On a component/board level there are many op-amps that you can use and then filter. Instrumentation amps might be a starting place.
No offense, but I don't think you really want a 6 digit DVM. Suppose you're measuring a 1 V signal, and the meter displays
$$ \large{1.02435 V} $$
The "5" are tens of microvolts. Are you really going to write down all those digits in you logbook? For everyday use I would write "1.02 V", for high precision maybe "1.024 V", but never more, but then I need a high precision setup in the first place. And the last digit won't be stable, even the "3" may switch to a "2" from time to time. What is it then? Also, measuring to better than 0.1 % (which is 3 digits) requires high precision for every component; the chain is only as strong as the weakest link. An 8 cm copper trace 0.2 mm wide on your PCB has a 0.1 Ω resistance, which will give an error of 0.1 % on a 100 Ω resistance. A 10 kΩ resistor has about 2 µV Johnson noise over a 10 kHz bandwidth. On a hot summer day you may have different readings with the windows open or closed.
Frankly, I would specify 4 significant digits; it will be hard enough to get the last digit stable and correct. And unless you work at FermiLab or CERN you're not really interested in more. It's not worth the trouble.
edit
" in my application I need this level of precision unfortunately"
Possible, yet extremely unlikely, even in a lab setup. Convince us. (By the way, do you mean precision or accuracy?)
What's that success up to 5 digits? Did you really pay $17 for a single 0.01 % resistor with a 0.2 ppm/°C tempco? For 6 digits accuracy you'll have to build a temperature controlled oven for your complete circuit, or the 0.2 ppm/°C will render the last digit useless.
There's a huge difference between 6 digit resolution and 6 digit accuracy.
If I were you, I'd start by reading this application note
AN124 - 775 Nanovolt Noise Measurement for A Low Noise Voltage Reference http://cds.linear.com/docs/Application%20Note/an124f.pdf
which will help you with power supply noise measurements, as well as discussing the LTC6655, which claims 0.775uV noise voltage to 10Hz.
