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

[Conrad Hoffman]

Good point about resistors. Especially don't use any thick films- stick with ordinary metal films. They say wire wound can be better but I doubt the change would be noticeable. I see you're using the nice 4-pole filter from the old Burr-Brown book by Wong, one of my favorites.

[C]

You state that you want to measure the low level noise on a power supply.
Your ltspice circuit is shorting a bunch of different noise signal sources to your common ground.

In the real world there three signals you do not show in your ltspice circuit. One is below VI, One is below R10 and one is below R17.

Just connecting your circuit only to the power supply will increase the noise at one or both terminals of the power supply. Due to the UN-balanced ground connection of VI that connection will add a different amount of noise then the other connection.

The same can be said about the two other connections(scope, DMM).
Think of your scope connection. If you just connected one of those old big TEK medal boxes with no power connection. You just added a big medal plate( part of a cap) that also acts like an antenna. Plugging in the power cord will increase/decrease the noise but not remove it. If you touch the scope, you have changed the noise.

Here you are shorting what exists at one terminal on the power supply to what exists at the ground connection of the scope.

I would think that you would get much better readings by using a fully differential circuit.     

C

[c4757p]

C, do you think this would work?

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

Couple in both + and - with identical capacitive highpass filters, subtract and amplify with an AD8429, then use the same filter configuration to separate the low and high bands (without gain this time)?

[Conrad Hoffman]

Your signal to noise ratio is always established at the first stage. After that you can only make it worse, rather than better. Thus, one puts as much gain as possible right up front. That way you can do whatever summing and processing after without much concern.

IMO, for most power supply work you won't see much difference between single ended and differential- unless you do! FWIW, I haven't, but I work at low (audio to 100 kHz) frequencies most of the time. That said, it's so easy to make the system differential when building up the unit, why not? That way you have it should any doubt arise.

[robrenz]

Looks like we have circled back to reply#1 

[Conrad Hoffman]

Pretty much, but after a certain age maybe we just like to talk a lot! FWIW, my Tek diff scope plug-in gets more use than any other. I use the lo-pass filters all the time, but very rarely the high pass. I also rarely use it in differential mode. Tek sells a similar add-on gadget (ADA400A) for the newer scopes, and it has much better bandwidth than my old 1A7A, but it's less flexible and a bit of a PITA to use. It also only works with scopes having the right power interface connections around the input BNC.

[C]

If you look at Ti, Analog Devices ...   all the big names you will note that when trying to push speed or low noise they use fully differential circuits with next to no connections to ground as you are using it. So in simple terms, for every signal you have going positive, you have a matching signal going negative. You also have a signal center. This matching and inverted signal chain also helps in the power supply paths if you keep it in place in the power paths. Balance, every thing is balanced.

My first choice would be a differential in and differential out opamp as the chip maker has a better chance of reducing the errors. If this does not exist with the spec's you need then matching opamps with one building the signal in the positive direction and the second the negative direction. Again with balanced signal paths.

You want a circuit that has an input with two connections. You want to amplify and connect that to two different devices each with two connections.
Think on it a sec, The reading you want should not change if you were to ground ether terminal of the power supply. It also should also not change if you connected a signal generator to one of the power supply leads and ground. This signal generator would act the same as EMI or any of other external noise sources that you do not want to measure.

Working up to the base idea. If you were to connect two resistors in series across the power supply, That resistor connection junction would be ZERO Power supply noise. This connection will still have some of all the other noise riding on it. You want the AC part with no DC so you need matching cap's. Your input signal is the three connections to the resistors. Will come back to this side in a sec.

You have a scope connection that is single ended. As I tried to say that ground side of the scope connection will have noise riding on it. What you really want to happen is to add your desired signal on top of that noise such that the ground connection noise does not effect the signal. If you had a fully differential scope input this could be easer. At the scope end you need to convert from fully differential to single ended in a way that will let you use the noise that exists on the scope's ground connection to compensate for it's noise & offset.   If your amp's differential output was two 1M resistors in series you would have a differential center zero. If you then paralleled this with the scope and a network matching the scope's input in series. You would then have the + & - signal. An equal to the scope's ground connection with out the noise and the scope's ground connection with the noise.

Your circuit should be a gradual change from the three inputs to the network that drives the scope.
Looking at what you have, you need the negative version of C5, R5 & U1. As you progress across the circuit you need to decide if the junction of +-R5(the noise center) should be connected to the next stage or skip it and how much resistance this connection should be. Remember the three signal paths, the positive signal, the negative signal & the noise center. Looking from the input side the noise center should see resistance to the scope's ground connection the noise center at the scope end of circuit.   
 
With the current circuit, If you were to put resistors inline to U1's power connections, you would see unequal currents flowing. With the resistors inline to the +-U1 setup these currents should match. Keep this idea in mind for the complete circuit. Just after U1 with the balanced signals you would really have five signal paths. The noise center, +-Signal, +-Power. Note NO ground so the noise that exists on ground is not getting in the signal path all over the place.   

So if you were to use two AD8429's, One for the positive and one for the negative the REF input would be a great assist in getting the signal centered properly on the scope's noisy ground connection.   

The sad thing is that the BNC Jack of the scope might be the only part of the scope's internal signal path that is single ended.   

Sorry, Not sure If I stated everything clearly. It actually sounds more complicated to say then the final fully balanced circuit that does this.
C

[robrenz]

Pretty much, but after a certain age maybe we just like to talk a lot! FWIW, my Tek diff scope plug-in gets more use than any other. I use the lo-pass filters all the time, but very rarely the high pass. I also rarely use it in differential mode. Tek sells a similar add-on gadget (ADA400A) for the newer scopes, and it has much better bandwidth than my old 1A7A, but it's less flexible and a bit of a PITA to use. It also only works with scopes having the right power interface connections around the input BNC.

Glad to see you back Conrad   My 7A22 and 7A13 diff amps are my most used also.

[c4757p]

I don't really see how that method differs from doing the differential->single-ended conversion at the beginning rather than the end, except for having more parts. The amplifier would be powered by batteries, so it's not like there would be ground loop issues between it and the scope. It's just whether you subtract the signals at the left side or the right side of the circuit board, as far as I can see

[c4757p]

This is what I have now.

The part number of the instrumentation amp is a mistake, it's of course AD8429.

I decreased R16 to 30k to eliminate the peak in the HF curve.

[C]

You need to look at all the different current loops or signal loops.

With single ended your signal path return is changing from the positive rail to the negative rail depending of which direction the signal is changing.
All the current differences in single ended design have effects on other parts of the circuit. Current demands are not balanced and opposite so they can not be localized as well as with balanced circuits.   
The ground connection of R8 is changing based on what all the other connections to ground are doing.
With a balanced circuit any change to the positive power rail is matched by an equal but opposite change to the negative power rail.
R8 no longer needs a connection to ground as it is connected to it's matching negative half. There fore no noise or errors from the rest of the circuit effects the R7 & R8 divider.

Need to Look closer as it not just more parts for the sake of adding parts.
There are some down sides to balanced but if single ended is so great then balanced would not exist.

Do a simple thing, in LTSpice put in some resistance to equal the copper resistance you would have on a PC board. Add the other real world things in and check the results.
Then really go to town with a fully balanced version of the circuit and compare the results.

C   

[c4757p]

I'll give it a try tomorrow. Going to bed now.

I don't doubt that it would work. I just suspect it's unnecessary. Remember, I only want a resolution of 5uV (that's not a 5uV signal, that's a 5uV "LSB").

[robrenz]

I got a HP 3410 for measuring PS ripple thread here.  I have not had time to restore it yet but it looks promising.

[c4757p]

OK, I built this. I know there were a good few changes suggested, but I was relatively confident it would work (perhaps with a couple tweaks) and really wanted to get on with it...

It seems to work quite well. More testing will follow (and I've only posted screenshots of the low frequency output, though the high frequency one works just as well).

Couple things -
 - I of course missed the input bias current of the instrumentation amp, which at 250nA on 470k and a gain of 200 would have the thing shooting for 24V... so I removed the 33k differential lowpass resistor and changed both 470k bias resistors to 33k.

- Added a switch to DC-null the inputs. Not really necessary, but it does stop the amplifier from saturating after power is applied, which allowed me to actually watch the settling time.

- RC time constant of the first input coupling stage is 1034 seconds, and 73 seconds for the second stage. Yes, this takes a long time to settle (plot attached).

- No need for batteries. The noise floor is indiscernible even powered externally.

- Clearly, from the layout of the panel parts (jacks and battery clips) I didn't plan the housing very well

Pictures:

- Inside the box. Yes, I built it in an electrical junction box. Works nicely.

- Closeup of the circuitry.

- LF noise floor with both inputs terminated.

- Output noise of my Power Designs TP343A.

- Settling time.

- Schematic.

The only thing I think I'd change if I redesigned it would be to add an offset nulling control to the LF output, so I don't have to use the oscilloscope to do it.