Designing a preamplifer for low noise measurement with RIGOL DS1054Z

[MaxSimmonds]

Hey Everyone,

I've been asked to perform a shorted input noise measurement on a current transducer I designed. At home, I have a Rigol DS1054Z. It only have mV resolution. Suppose the noise I'd like to measure is in the order of uV, I'm thinking I need a preamplifier.

My thought process was the following:

1) I want to measure noise of this current transducer, it could be uV range.
2) My scope has mV resolution
3) What is the noise floor of my scope, if it's higher than the noise of the current transducer, then this won't work
4) Design a preamplifer such that the noise floor of the scope is below the current transducer, by amplifying the signal.

Is this the correct thinking? I'm thinking that I need a 1000X gain preamplifier, such that 1uV = 1mV. I should then be able to resolve any noise from the current transducer card. I shorted the input of my oscilloscope, and then performed an FFT to measure the noise floor, it's worse nearing DC:



At DC, it's ~-86dBV, so that's 0.05mV of noise? I'm assuming that means that, without a preamp, any signals below 0.05mV will be hidden in the noise? Plus, since the smallest volts per division I can go is 1mV/div, it's going to be hard to measure anything in the order of 0.1mV?

So, my preamp needs to have a noise floor less than -86dBV, AND amplify the input around 1000 times, so that it would be useful for the Rigol? Is that the right thinking? Unfortunatly, whilst writing this, it's been decided to use a eval board (specifically this one https://www.mouser.co.uk/ProductDetail/Texas-Instruments/LMH6629SDEVAL-NOPB?qs=7lkVKPoqpbZbNcznfTrd9g== ) instead, but I still want to design my own, if you haven't guessed yet, I really don't know a lot about noise and measuring it.

[Fungus]

Is this the correct thinking?

Yes.

This video might be helpful:

[mawyatt]

This is a really good video! Those old LT app notes, especially the ones by Jim Williams, should be required reading for every EE grad going into analogish electronics

Best,

[David Hess]

I have done what you are attempting many times with operational amplifier based x1000 amplifiers although not with a true current source where a transimpedance amplifier would be more suitable.

[Kleinstein]

One may not need to get the noise all the way down that it is much lower than the scopes noise. A factor if 3 usually good enough. The noise source add as power and 3² is allready enough to essentially ignore the smaller noise source.

For builing the amplifier it is important to also look at the frequency range and also at the source impedance.  For the high impedance (some 10 M) and the large BW the scope amplifiers are not that bad. The main reason why an extra low noise amplifier can do better is usually either a lower BW or a degin for a lower impedance. The noise level of the scope is also depending on the BW and with the FFT with the length of the data set used.

The video is really nice. The LT apply note with the parallel transistors only looks at voltage noise and only works well with a really low impedance source. With the couling cap at the input this can add noise for the lower frequency range.  Parallel amplifiers reduce the votlage noise, but also increase the current noise. This makes the amplifier more suitebale for lower impedance sources.

[MaxSimmonds]

Hey thanks guys! All super useful, especially that video. I also found one made by Marco Reps, it's the first I've seen of him and now i'm hooked!

https://youtu.be/XpbDMo8an5w

I think we're now going for using the Agilent 3458A dmm, connected directly to the output of the current transducer board I designed. I'm unsure if this will work, but I can't find a valid reason why it shouldn't.

Would this work? An 8.5 digit multimeter, to measure the RMS noise?

[Kleinstein]

One can use a DMM and measure RMS voltage from there. Just be aware of the frequency range covered. With the 3458 one has a lot of integration times to chose from and also the choice auto zero or non auto zero.  For the higher frequency part there is the AC mode of the DMM.

The DMM input is made for high impedance, a little like the scope. The noise level may not be much lower than with the scope - just the lower BW helps to get more resolution.
So one may still want a preamplifier.

[MaxSimmonds]

Okay, then I guess it's possible. If I wanted to calculate the spectral density (the nV/root hertz value) would I do the following:

Assuming the system will have a bandwidth of 10KHz (current frequencies above this are attenuated by eddy currents induced within a stainless steel vacuum vessel, this is for a particle accelerator application), and a roll off of approximately 20 dB/decade (https://accelconf.web.cern.ch/pac2009/papers/th5pfp083.pdf link for those interested):

1) Measure the Vrms noise using the dmm, with an integration time of 0.1ms (10KHz)
2) Use the following equation: Vrms / (root(10000)) = Spectral Density

Would this be correct? And also, secondary question, why is the spectral noise higher at lower frequencies? I would have thought that at lower frequencies (higher integration times) the noise floor would be better?

Thanks again for helping out a noise noob!




EDIT// Dave to the rescue - I guess the noise is high at lower frequency simply because it's not spread out as much, am I confusing the integration time with bandwidth?!?

[Kleinstein]

The spectral noise density goes up at low frequencies (e.g. below some 1-100 Hz depending on the circuit) because of 1/f noise and thermal fluctuations.
The noise density of the ADC in the DMM does not change very much with integration time, at least not for the middle part. It can go up for shorter integration time as the resultion may hit a limit with this DMM. The details depend on the DMM used. Anway in the low votlage range (100 mV) the limiting part is more the amplifier, for the 10 V range it is often the ADC and maybe the reference.

The bandwidth is usually half the sampling rate - though depending on the details there can be also noise contribution from outside this band. So it depends on the details (e.g. using AZ mode or not).
The noise bandwidth for a single sample is 1/ (2 * integration time) - so they are linked, but the frequency response is more complicated.
The simple formular of RMS noise divided by square root of the BW is OK for white noise or same average, e.g. for a first estimate.

[MaxSimmonds]

Okay I think that makes sense, at least for the most part.

So if I want to know the noise density value at 10k, I need to use an integration time of 0.05s? Could you explain this a bit more, is it literally die to Nyquist theory?

Looking at Dave's video, maybe I want to use a signal analyser instead? Or is the DMM approach (measuring RMS noise and calculating the noise spectral density) also valid?

Thanks very much for your time btw, really appreciate it!

[egonotto]

Hello,

perhaps you can read:
Art Kay Operational Amplifier Noise
to get more information.

In the meantime you can evaluating the ADA4523-1 36 V, Low Noise, Zero Drift, Operational Amplifier.
This is the OP that Marco Reps used in his great video. There is a eval board EVAL-ADA4523-1ARMZ
https://www.mouser.de/ProductDetail/Analog-Devices/EVAL-ADA4523-1ARMZ?qs=vmHwEFxEFR%2FOGv2wT%252BhMrw==
for about 27€ 

Best regards
egonotto