Op amp for amplifiying 5uV of noise to 200uV?

[c4757p]

I want to build a simple amplifier for measuring the output noise of my PSU kit. I'm looking to measure in units of at least 5uV. I want a gain of 40, so that my lowest noise "LSB" is at least 2 real LSB on my Fluke 45 on either "fast 300mV" or "medium 3V" range.

Now, 5uV isn't incredibly low. I just pulled up the TL072 datasheet (because I have a shitload of them), and while I've always thought of it as reasonably noisy, it looks like it should work for a 0.1Hz to 1kHz* range. The noise voltage vs. frequency plot only goes down to 10Hz, but it's rising at about 18nV/rtHz/decade, so if it's 45nV/rtHz at 10Hz, I imagine it would be 63nV/rtHz at 1Hz and 81nV/rtHz at 0.1Hz. Am I correct? That translates to 14nV at 10Hz, 63nV at 1Hz, and 256nV at 0.1Hz. Multiplied by 40, that "huge" 256nV is 10uV, which is big, but well under the 100uV LSB of the meter. So, can I use it for this? Or am I missing something?

*I'll switch to the faster, but lower-resolution, HP 3437A for noise at this end of the frequency range, or use true-RMS mode.

(Of course, I know about Johnson noise, input noise current, the need to power it from a battery to eliminate external PSU noise, and the need to shield it with a thick, metal box.)

[robrenz]

Isn't it going to need to be differential to block out all the ambient noise?

using a 7A22 diff amp here picture links are broken but the pics are in sequence at the bottom

[c4757p]

If it's floating on battery power (including the PSU itself), wouldn't it effectively be differential?

[robrenz]

Does battery power make something immune to general EMI just because there are no ground loop issues?

[c4757p]

No, but wouldn't the diff amp be equally prone to that, at least, after the differential input stage? That's why it would all have to be well shielded.

I can get a reading on the meter stable all the way to the last LSB with no special precautions, so I don't think I'm really operating in Voodoo Land here.

[robrenz]

I was thinking you were using this for a scope.  I think the low bandwidth of the meters ac measurement is masking all the noise that you would see when measuring on a scope. jahonen has a nice build here for a scope preamp.

I assume you are doing this to get into the range where your meters ac specs are valid?

BTW I am way out of my league here, I am the last person that should be giving you advice

[c4757p]

I assume you are doing this to get into the range where your meters ac specs are valid?

Well, at the low-frequency end, I will sample DC rapidly over GPIB to get a measurement and then compute RMS in software. But 5uV is too low even for DC, at least at faster acquisition rates.

[robrenz]

Aren't you going to get bogus info for your RMS calculation at such low DC sampling rates. Unless you use 100NPLC you are going to have noise that is higher than the levels you are trying to measure. At 100NPLC you are only getting a reading every 1.66 seconds at best. That is a 0.6Hz sample rate.

[c4757p]

Possibly. I don't know off the top of my head how fast I can pump data out of the Fluke or the HP, and I'm a couple hours away from my bench right now, so really I'm more or less publicly brainstorming. Perhaps I could chain two amplifiers for higher gain and feed that into a DSO? I could also experiment with my P6046 for this, though it's in need of a (simple) repair first, and I don't know what its noise specs are.

Clearly I have no experience with low-voltage near-DC measurements

[Andreas]

it looks like it should work for a 0.1Hz to 1kHz* range.

The TL072 is optimized for high source impedances and audio frequencies.

The lowest noise OP-Amps for above 10 Hz and source impedances < 300 Ohms are the LT1028 and the AD797.

http://cds.linear.com/docs/en/design-note/DN355f.pdf

For other source impedances and frequencies you might have to choose other OP-Amps.
Below 10 Hz the optimum is a chopper stabilized OP like the LTC2057.

The question is also which minimum frequency can be handled by your RMS-Instrument.
Usually you measure 0.1 - 10 Hz as peak-peak on the oscilloscope with one filter amplifier and > 10 Hz - 1, 10 or 100kHz as true RMS value with another band width limited amplifier.

With best regards

Andreas

[c4757p]

Those look good. How about AD8638 (auto-zero) for the low end and AD8597 for the high end? The LT part looks good, but Mouser doesn't carry LT and DigiKey has crap prices on a lot of the things I want to order at the same time. Looks like AD8597 is similar to 797 - higher noise, but still well within my limits.

[Andreas]

Those look good. How about AD8638 (auto-zero) for the low end and AD8597 for the high end?

The AD8597 looks very good. Forget the AD8638 and use the 8597 for the whole range if you can't get the 2057 for the 0.1 - 10 Hz.

With best regards.

[c4757p]

Very well, it does look excellent. Thank you for your help!

Now, are there any noise-related gotchas with polypropylene film capacitors? That's what I am planning on using to implement the filters.

[SeanB]

Black stripe end is outer foil, connect to low impedance side or ground. Grip lead end when forming so you do not stress the epoxy bond to the schoop end electrode. Try not to burn them with a soldering iron tip while mounting. Otherwise nothing really, clean board preferably without getting the capacitor wet with any solvents that might permeate the casing. If they are high value big units have 2 holes on the board so that you can use some lacing cord to tie them down in a high vibration environment.

[Conrad Hoffman]

This doesn't sound that tough with a decent opamp and filters. There are no noise problems with film caps. I'd be starting with something like OP-27 or OP-37 for their good DC specs, and a 4th order filter. PM or email me if you want a good single opamp 4-pole filter design. Just as a reference point, I find the low pass filter selections on the Tek 1A7A plug-in very well chosen. One thing to watch out for is popcorn noise. It's not that common now, but will drive you nuts in DC work. It's process dependent and now and then you get some parts that have a bit too much.

[c4757p]

Black stripe end is outer foil, connect to low impedance side or ground.

I can't seem to find any with such a mark. What about this method to test? Wrap a wire once or twice around the capacitor, and feed a high-amplitude, high-rise-time square wave through the wire. Probe across the capacitor in both orientations; the orientation which displays more noise has the probe tip on the outer foil.

I'd be starting with something like OP-27 or OP-37 for their good DC specs, and a 4th order filter.

It looks worse than the AD8597 in most ways. Is there something I'm missing?

[blackdog]

Hi c4757p, :-)


Take a look @ the AD8429 instumentation amplifier.
Ask a sample from Analog Devices.

Use it @ gains of about 100x for the lowest noise.
Symetrical amplifiers is the best way, i use a Audio Precision for noise measurements of my PSU designs.
I never want to measure these low levels asymetrical :-)

Please be aware about protection of your measurement inputs.
You need big capacitors for low noise measurements @ the input,
This wil give large charge currents through the input.
But, probably you know this...

The best lownoise PSU i have build with a 22Khz BW, was about 3uV.
It is a hell to measure this, because of al the EM Fields in my lab.

I hope this helps.

Kind regarts,
Blackodg

[Andreas]

It looks worse than the AD8597 in most ways. Is there something I'm missing?

Op 27, 37 (or LT1007, LT1037) are optimized for higher source impedances in the 1K (-10K) range.

The trade off for low voltage noise is high current noise. So with the AD8597 you will have to use low ohmic resistors and (very) large capacitors.

The typical circuit you can find here:
http://tangentsoft.net/elec/lnmp/misc/schematic3.pdf

With best regards

Andreas

[c4757p]

I think I will go with the OP-27.

The typical circuit you can find here:
http://tangentsoft.net/elec/lnmp/misc/schematic3.pdf

It looks like they are using an electrolytic capacitor on the input. How much noise is there to the relatively high leakage current?

[Andreas]

Hello,

typically you will select the capacitors for low leakage current. (e.g. < 20nA)
Use standard 85 degrees capacitors with high nominal voltage for best results.

I use a pre-charging cirquit with a 9V battery to keep the capacitor charged always to minimize noise on my 0.1 Hz - 10 Hz amplifier.

As I have to keep the source impedance to >= 1k Ohms I use the LT1037 (OP37 would also work). Noise with input shorted is around 120nV pp in the 0.1 - 10 Hz range. (either with precharged capacitor or after 24 hours run in phase).

With best regards

Andreas

[Conrad Hoffman]

Thought this was a DC application. That circuit is AC coupled and inverting, with a very low input impedance. Not something I'd want for a general purpose measurement circuit. The first stage should be non-inverting, with as much gain as possible, since that stage will establish the signal to noise ratio. The input should have some protection in the form of a series resistor and some diodes so you don't lose the opamp if you probe something err... unexpected. Maybe a ferrite bead or some RF supression as well. Follow that stage with a lo-pass filter to knock out the HF noise. For a scope this becomes way more effective visually at 3rd order and above, but for a meter it probably doesn't matter as much, unless the meter has a very wide bandwidth.

[c4757p]

It has to be AC-coupled, it's looking at PSU output noise. The gain applied to the actual DC level will just saturate the amplifier.

Desired range is 0.1 Hz to 20 Hz, if I didn't mention.

I've got a decent filter designed, but I'm in my car now so I can't post it. Later

[Conrad Hoffman]

OK, sorry! In that case the low input impedance isn't a problem, other than requiring very large coupling caps. I'd still add some protection diodes and RF suppression due to paranoia if nothing else. If cap leakage is an issue, there's a trick where you use two in series, with a moderately high resistor to ground. That keeps the junction near zero, so leakage in the next cap isn't an issue and the opamp doesn't produce a big offset.

[c4757p]

How's this? Here are the figures:

Bands

LF
55.9dB at the top
54.6dB at 20 Hz
55.4dB at 100 mHz
52.9dB (-3dB) at 23 mHz
52.9dB (-3dB) at 33.5 Hz

HF
55.9dB at the top
55.7dB at 20 Hz
52.9dB (-3dB) at 12 Hz
52.9dB (-3dB) at 1 kHz



Noise

LF
25.4 uV RMS integrated over 100 mHz - 20 Hz:
    Input at 1 uV RMS: 27.8 dB SNR
    Input at 100 uV RMS: 67.8 dB SNR

HF
28.4 uV RMS integrated over 20 Hz - 1 kHz:
    Input at 1 uV RMS: 26.9 dB SNR
    Input at 100 uV RMS: 66.9 dB SNR

The SPICE models used do model both voltage and current noise.

If cap leakage is an issue, there's a trick where you use two in series, with a moderately high resistor to ground.

Good idea.

[SeanB]

The band may not be very visible on some, even can just be a line on the one side of the value or just in a datasheet that outer foil is to the left when reading the value. You will likely be using 47uF ( about the biggest value in a wire ended package) 63V units, and those are very large. You will find you need a big PCB that will have very low component density on it around the filter. Do not bother using SMT unit resistors there, use through hole and DIP packages to make assembly easier.