(pre-)amplifier to measure power supply noise

[exe]

Hello.

I would like to measure noise of my linear power supply. I think I need a pre-amplifier as my oscilloscope's noise dominates the measurement. I've seen a few circuits before, but, sadly, didn't make a bookmark. So far I only found not too many circuits, one of them is this one: https://www.analog.com/en/analog-dialogue/articles/low-noise-inamp-nanovolt-sensitivity.html . Basically, it's a single IC solution using ad8248. The article suggest to put four of them in parallel to reduce the noise, but, give the price of the device, I think I'll start with one .

So, what do you think of ad8248? Is it good-enough to measure some tens of microvolts? Or may be there is a cheaper alternative? I see there are quite av few low-noise opamps on the market. I don't really need it to be very accurate in terms of gain or something. It's more for comparative measurement. Bandwidth is not that important. I think 1MHz is should be fine. Gain 100-1000x, not sure how much I really need. I think the trade-off here bandwidth and gain accuracy. More gain -- less bandwidth.

[iMo]

The lowest frequency of interest?

[exe]

10-20Hz, up to 50Hz if this makes a big difference (I want to see ripple, if any). But I'm open for suggestions. I think very low frequencies I can measure with my keithley 2000 (may be not, never tried).

Basically, for me it's a learning experience, I don't really have a need for this. I just want to play with my DIY power supplies and see what and how affects stability and response.

[David Hess]

For a cutoff frequency that low, a low input bias current operational amplifier should be used to make the size of the input coupling capacitor reasonable.  This only applies to the first stage; a majority of the gain can be in the following low impedance stage or stages.

If the noise is really that low, then a differential measurement may be necessary to remove noise from ground loops.

[radiolistener]

if you want to test power supply, you're needs at least 3-10 MHz bandwidth, in order to be able to see possible internal oscillations in the linear regulator circuit or noise in the switching regulator.

[exe]

May be I need two different amplifier? One is for low frequency measurement, the other one is with higher bandwidth (and possibly lower gain). Although ad8428 claimed to have a 3.5MHz bandwidth, which is probably enough for me.

I wonder how much gain do I need. With 3MHz bandwidth I can use x1 passive probes. My oscilloscope can display signals of amplitude 10mV (noise floor is around 1mV when input is shorted). Sounds like gain of 1000 should do the job. May be I'm a bit optimistic about environment noise. I'm pretty sure I'll need to shield the preamp.

BTW, the power supplies I'm going to measure are in plastic enclosure (=not shielded). So, may be, I don't even need sensitivity down to tens of microvolts due to EMI. I dunno .

[iMo]

Mind the "bandwidth" of an opamp is always a product of the closed loop Gain and the BW.
With an opamp with claimed 3.5MHz GBW and an amplification set to 100 the actual BW will be 35kHz.
For 3.5MHz bandwidth and amplification of 1000 you would need an opamp with GainBW product of 3.5GHz.
Therefore you need several opamps in series (AC coupled).. Like 3 opamps with 3.5MHz GBW each, set 10x gain each and as the result you'll get 1000x Gain with around 350kHz BW.

FYI - below is a 2 transistors Amplifier with 72.4dB (4190x) gain, with 135kHz bandwith (30Hz-135kHz). I bet with good noise params for your application.

[not1xor1]

Mind the "bandwidth" of an opamp is always a product of the closed loop Gain and the BW.
With an opamp with claimed 3.5MHz GBW and an amplification set to 100 the actual BW will be 35kHz.
For 3.5MHz bandwidth and amplification of 1000 you would need an opamp with GainBW product of 3.5GHz.
Therefore you need several opamps in series (AC coupled).. Like 3 opamps with 3.5MHz GBW each, set 10x gain each and as the result you'll get 1000x Gain with around 350kHz BW.

AD8428 is an instrumentation amplifier with a fixed gain of 2000 and 3.5MHz is the -3dB bandwith at that amplification rate.

IMHO that bandwidth should be enough for a linear PSU and if that is just an ordinary PSU (nothing related to metrology) a single AD8048 would work as well.
An alternative is AD8049, with programmable gain, but the full bandwidth (15MHz) can be achieved only with a gain of 1.

[not1xor1]

10-20Hz, up to 50Hz if this makes a big difference (I want to see ripple, if any). But I'm open for suggestions. I think very low frequencies I can measure with my keithley 2000 (may be not, never tried).

Basically, for me it's a learning experience, I don't really have a need for this. I just want to play with my DIY power supplies and see what and how affects stability and response.

Ripple is usually at 100Hz (unless you are using just a single diode instead of a rectifier bridge).
In any case you need a much wider bandwidth since ripple is usually full of harmonics.

[David Hess]

if you want to test power supply, you're needs at least 3-10 MHz bandwidth, in order to be able to see possible internal oscillations in the linear regulator circuit or noise in the switching regulator.

20 MHz is common for noise measurements but lower limits are fine if direct comparisons are not needed.

May be I need two different amplifier? One is for low frequency measurement, the other one is with higher bandwidth (and possibly lower gain). Although ad8428 claimed to have a 3.5MHz bandwidth, which is probably enough for me.

I commonly use my Tektronix 7A13 with a 5MHz bandwidth or 7A22 with a 1MHz bandwidth for power supply noise measurements.  Higher sensitivity oscilloscopes tend to be limited to 5 or 10 MHz because of input noise considerations anyway.

I wonder how much gain do I need. With 3MHz bandwidth I can use x1 passive probes. My oscilloscope can display signals of amplitude 10mV (noise floor is around 1mV when input is shorted). Sounds like gain of 1000 should do the job. May be I'm a bit optimistic about environment noise. I'm pretty sure I'll need to shield the preamp.

BTW, the power supplies I'm going to measure are in plastic enclosure (=not shielded). So, may be, I don't even need sensitivity down to tens of microvolts due to EMI. I dunno .

Just enough gain to exceed the input noise of the oscilloscope is needed.  1000x might be convenient but is pushing what is feasible without very careful construction.  See below.

Mind the "bandwidth" of an opamp is always a product of the closed loop Gain and the BW.
With an opamp with claimed 3.5MHz GBW and an amplification set to 100 the actual BW will be 35kHz.
For 3.5MHz bandwidth and amplification of 1000 you would need an opamp with GainBW product of 3.5GHz.
Therefore you need several opamps in series (AC coupled).. Like 3 opamps with 3.5MHz GBW each, set 10x gain each and as the result you'll get 1000x Gain with around 350kHz BW.

1000x in three integrated stages of 10x each is just barely feasible without too much trouble.  That is 60dB of gain however which will require very careful isolation and shielding between stages to prevent oscillation.

[exe]

Thank you very much, guys! I'll take suggestion into account and try to measure something.

Mind the "bandwidth" of an opamp is always a product of the closed loop Gain and the BW.

Good point, I had doubts about that as well. I started to look into the datasheet and it says "[two-stage architecture] This architecture results in a 3.5 MHz bandwidth at a gain of 2000 for an equivalent gain bandwidth product of 7 GHz.". I'm glad they are explicit about that

[exe]

FYI - below is a 2 transistors Amplifier with 72.4dB (4190x) gain, with 135kHz bandwith (30Hz-135kHz). I bet with good noise params for your application.

Nice old-school! May be I'll start with that, before buying expensive AD devices .

[not1xor1]

FYI - below is a 2 transistors Amplifier with 72.4dB (4190x) gain, with 135kHz bandwith (30Hz-135kHz). I bet with good noise params for your application.

Nice old-school! May be I'll start with that, before buying expensive AD devices .

That is a bad choice.
Those AD ICs are differential amplifiers. They would take care and get rid of all common mode noise while that simple 2 BJTs amplifier will show just EMI noise in your lab.
Look at "How to measure PSU noise and ripple" video and thread for more details.

[Gerhard_dk4xp]

Quite a strong allegation. Show me a baseline at -190 dBV with your differential amplifier.
The EMI noise in my lab must be something different.

<       https://www.flickr.com/photos/137684711@N07/24354944411/in/album-72157662535945536/        >

noise density of some LEDs,   0 dB = 1 nV/rtHz, no cross correlation.

regards, Gerhard