Measuring amplifier noise with scope?

[aneevuser]

However, it looks a little too good to be true - in my setup, won't I still need a resistance to ground at the inputs of the instrumentation amp, to satisfy input bias current requirements? If so, doesn't that undermine the whole argument that you've made?

That is right; a DC connection to common is still required at the input to supply the input bias currents however it is the source impedance which determines input noise.

I'll have to come back to all this at a later time, but from what you've said, I'm guessing that the effective noise resistance is the source resistance || bias current resistance (or something similar - not sure of the details)? So I can in fact make the bias current resistances quite large without degrading noise performance too much?

[David Hess]

I'll have to come back to all this at a later time, but from what you've said, I'm guessing that the effective noise resistance is the source resistance || bias current resistance (or something similar - not sure of the details)? So I can in fact make the bias current resistances quite large without degrading noise performance too much?

That is right.  And it is the differential source resistance which matters because the instrumentation amplifier rejects common mode noise.

[aneevuser]

I'll have to come back to all this at a later time, but from what you've said, I'm guessing that the effective noise resistance is the source resistance || bias current resistance (or something similar - not sure of the details)? So I can in fact make the bias current resistances quite large without degrading noise performance too much?

That is right.  And it is the differential source resistance which matters because the instrumentation amplifier rejects common mode noise.

I'm not sure what you mean by "differential source resistance" here. Does this just mean the source resistance across the diff. inputs i.e. in my case the 220R?

Anyway, this looks incredibly sweet - I'll put it together tomorrow and see if I can measure any improvement in the noise. Unfortunately I'll have to use a 411 for the "backend" amp - only 3-4 MHz GBW -  but I think that'll be good enough given my bandwidth requirements, and the higher noise shouldn't matter there.

Oh and I've just noticed another benefit - controllable gain via a single pot. - I don't think that was possible with my original diff. amp. arrangement.

[David Hess]

I'm not sure what you mean by "differential source resistance" here. Does this just mean the source resistance across the diff. inputs i.e. in my case the 220R?

Yes.

Oh and I've just noticed another benefit - controllable gain via a single pot. - I don't think that was possible with my original diff. amp. arrangement.

There are ways to do it but it adds complexity.

Douglas Self wrote a book called "Small Signal Audio Design" which has a chapter on microphone input amplifiers.  Analog Devices has some application notes on the subject.  And check out Walter Jung's book "Op Amp Applications Handbook" which is freely available online.

[aneevuser]

OK, after a torrid 24 hours of nothing at all working properly (breadboard eff ups, probably), I've put together an operational version of the following in-amp-topology preamp:



with:

R_i = 4K7
R_f=560R || 10 pF
R_g=100R pot.
R=1K

I've used hand tested 2% resistors to ensure the best matching with my somewhat limited stock of components, and it's once again a Heath Robinson breadboard affair, with no attempt at keeping leads short. In the first version, I was getting lots of oscillation with a 1 kHz i/p signal (and some bizarre o/p clipping at 1V) so I've added some more decoupling after rebuilding and the problems have now gone away. It's powered off 2 9V batteries so I don't have to worry about power supply noise.

With the measurement setup as described in the OP, I measure 2.7 mV(rms) self noise for the amp with a 220R source resistance - a smallish but useful improvement from the original 3.5 mV(rms). However, I also now have a measured CMRR of -105 dB (! is appropriate, I think) and I think that makes a more significant difference to the the audible noise. (The CMRR of the purely differential amp was about -60 dB).

With the gain set to 70 and with the mic. given a bit of adhoc shock mounting, the noise is now pretty acceptable in recordings - much less noticeable rumble.

So, to summarise: instrumentation amplifier topologies for mic. preamps is now very much the bag I'm into, baby. I'm fairly impressed that a clueless noob such as myself can throw something together with such good CMRR from discrete components.

If anyone wants to point out any obvious improvements that can be made, feel free to do so.

BTW, I have in the mean time discovered the INA163 which is aimed at these sort of applications - would I be likely to see a noticeable improvement if I were to use one of those instead?

[magic]

With the gain set to 70 and with the mic. given a bit of adhoc shock mounting, the noise is now pretty acceptable in recordings - much less noticeable rumble.

With the measurement setup as described in the OP, I measure 2.7 mV(rms) self noise for the amp with a 220R source resistance - a smallish but useful improvement from the original 3.5 mV(rms).

This isn't a valid comparison if you haven't preserved the original 15kHz bandwidth limit.

I recommend taking a few seconds of recorded noise and feeding it into some spectrum analysis software for comparison. This will show noise density vs frequency, in some unspecified dB units which stay consistent as long as recording volume and analysis software settings are the same. This is important, don't fiddle with any FFT options between measurements.

With a bit of calibration, one can even produce numbers in absolute units of nV/rtHz or dBV/rtHz. For example, an NE5532 with 1Ω/100Ω gain network and the input shorted to ground makes a ballpark accurate 5nV/rtHz input-referred noise reference because Johnson and current noise are made negligible. More accurate calibration is also possible from first principles, but it's a subtle business as shown by example of one NwAvGuy whose noise measurements are complete bollocks if naively interpreted in units of dBV/rtHz.

[aneevuser]

With the measurement setup as described in the OP, I measure 2.7 mV(rms) self noise for the amp with a 220R source resistance - a smallish but useful improvement from the original 3.5 mV(rms).

This isn't a valid comparison if you haven't preserved the original 15kHz bandwidth limit.

Hmm. Are you talking about the high frequency roll-off here? If so, yes, it looks like I've screwed up - doubly in fact. There's a typo in the info above - should be 10 nF not 10 pF, but in fact the 10 nF is wrong as it doesn't give the ~15 kHz 3db point. Damn - I'd better look at this again. Thanks. However, it's wrong in the right direction - this should double the noise bandwidth, and if that's good enough, then it can only get better when I fix it.

I recommend taking a few seconds of recorded noise and feeding it into some spectrum analysis software for comparison. This will show noise density vs frequency, in some unspecified dB units which stay consistent as long as recording volume and analysis software settings are the same. This is important, don't fiddle with any FFT options between measurements.

Sounds like a good idea - does Audacity do this? If it does it more accurately than the scope, that would be ideal.

[magic]

I think it has that option, check the menus.

One thing I forgot: I'm not entirely sure if the whole noise floor won't move up or down if you increase/decrease the length of file. Yes, you read that right

Weird problems exist when measuring noise because spectrum analysis divides the whole spectrum in buckets and reports signal energy per bucket. With concrete tones, each tone falls in one bucket and this bucket reports the whole energy of the tone correctly. With noise, bucket width (measurement resolution) determines how much energy falls in each individual bucket. The results may vary with software settings in bizarre ways.

[magic]

INA163

These things have inherent noise advantage in that they use a very simple current feedback input stage with only one transistor per cold/hot input. This gives them noise performance on par with a normal opamp made of the same low noise transistors, but it would take two such opamps to build an equivalent circuit. Conceptual schematic of a similar part from AD is attached for reference; I suppose BB uses similar input circuitry.

But for reasons not entirely clear to me, their differential to single-ended stage is oftentimes horribly noisy. When you look at the plot of input referred noise vs frequency and gain for INA163, you see that it only reaches its full potential at gains greater than 100x. At 100x or less it would be beaten by quiet opamps like LT1028, AD797, OPA1612, maybe even NJM2122. At gains of 10x and less it's on par with what your current NE5532 circuit should deliver.

[aneevuser]

INA163

These things have inherent noise advantage in that they use a very simple current feedback input stage with only one transistor per cold/hot input. This gives them noise performance on par with a normal opamp made of the same low noise transistors, but it would take two such opamps to build an equivalent circuit. Conceptual schematic of a similar part from AD is attached for reference; I suppose BB uses similar input circuitry.

Thanks for this. I think that this has in fact answered another question that I was going to ask. David Hess mentioned the Self audio book - I have a copy of this in fact, though I havent looked at it much. Self has some preamps frontended with a discrete transistor stage, whose topology confused me - they look like they have an almost-long-tailed-pair, with emitters shorted by a resistor - this is simply a discrete transistor in-amp, isn't it?

But for reasons not entirely clear to me, their differential to single-ended stage is oftentimes horribly noisy.

They're using 5K resistors there - that seems large to me - could that be part of the reason?

BTW, what determines how small the resistors can be chosen on the o/p stage? The o/p impedance of the front end amps will be negligible, so presumably it's not that?

When you look at the plot of input referred noise vs frequency and gain for INA163, you see that it only reaches its full potential at gains greater than 100x. At 100x or less it would be beaten by quiet opamps like LT1028, AD797, OPA1612, maybe even NJM2122. At gains of 10x and less it's on par with what your current NE5532 circuit should deliver.

Well, I'm looking for gains of 70-100 with my current mic. - maybe less if I get a condenser mic - so it should be a good match from what you've said.

Anyway, I'm still basking in the glow of -105 dB CMRR from a breadboard lash up - eat ya heart out, IC designers!!!!

[aneevuser]

Two other points:

1) It's just dawned on me that since the noise model for a resistor is a noisy AC voltage source in series with an ideal resistor, then if you want to analyse the noise behaviour of a circuit properly, you need to do it on the AC equivalent circuit. This is probably blindingly obvious to anyone who isn't a noob.

2) I'm now beginning to see why the weird audiophile guys get so weird - it's very easy to get painfully involved with trying to hear small improvements in noise, or whatever, and to keep tweaking things to get the best possible result. Anyway, that's enough for now - my delivery of oxygen free copper wire has just arrived - can't wait to see what difference that makes.

[magic]

Self has some preamps frontended with a discrete transistor stage, whose topology confused me - they look like they have an almost-long-tailed-pair, with emitters shorted by a resistor - this is simply a discrete transistor in-amp, isn't it?

Obviously, only if the amplifier has very high open loop gain from each input to the output and feedback is arranged in the same manner as here to set a desired closed loop gain.
Such resistor arrangement could also be used in any other kind of amplifier to reduce and linearize the differential gain of an LTP stage.

oxygen free copper

Have a look at the JRC MUSES03. It's a rare opamp with OFC pins and it has JFET inputs so you can parallel pretty much as many of them as you want to average out the noise. Bummer that the starting noise is relatively high.

Maybe TI will pick up the fad and offer OPA827 with OFC? So far I have only seen one TI analog manager rating on DIYAudio how unfair it is that "a competitor" sells snake oil chips made with 1980s technology and makes money on that.

Those guys have no clue

[aneevuser]

Have a look at the JRC MUSES03. It's a rare opamp with OFC pins and it has JFET inputs so you can parallel pretty much as many of them as you want to average out the noise.

Thank you but sadly, since Mouser seems to be asking for about 60 of the finest English pounds each (inc. VAT - they're generous like that) in quantities < 10, I suspect that I'll stick with my discrete-in-amp-o-matic solution, humble as it may be.

[exe]

My friends, I see discussion of correcting offset voltage, but it seems I missed the problem statement -- why do you think offset is a problem? I believe many well-designed audio circuits don't need it. If one wants, there are many precision opamps with very low offset voltage.

It is also possible to externally correct offset voltage, or even build a something like a chopper amp, or with some servo circuitry. But before going that way I'd say it needs to be proven that this is necessary.

Update: wrong thread

[magic]

I see discussion of correcting offset voltage

I think you meant this post for some other thread?

But for reasons not entirely clear to me, their differential to single-ended stage is oftentimes horribly noisy.

They're using 5K resistors there - that seems large to me - could that be part of the reason?

You can calculate how much of a reason it is and how it compares to the second stage noise spec from their datasheet. Hint: their second stage is almost identical to the circuit you started with

BTW, what determines how small the resistors can be chosen on the o/p stage? The o/p impedance of the front end amps will be negligible, so presumably it's not that?

There are other concerns like the size of output stage transistors required to drive those resistor networks (remember the amp runs on up to ±18V rails), power consumption, heating of the die.

[exe]

I see discussion of correcting offset voltage

I think you meant this post for some other thread?

Yes, thanks for telling me, I was wondering what happened to that message.

[David Hess]

Thanks for this. I think that this has in fact answered another question that I was going to ask. David Hess mentioned the Self audio book - I have a copy of this in fact, though I havent looked at it much. Self has some preamps frontended with a discrete transistor stage, whose topology confused me - they look like they have an almost-long-tailed-pair, with emitters shorted by a resistor - this is simply a discrete transistor in-amp, isn't it?

It is the two transistor equivalent to the dual operational amplifier differential preamplifier we discussed earlier.

[aneevuser]

It is the two transistor equivalent to the dual operational amplifier differential preamplifier we discussed earlier.

This confuses me a lot. Self seems to have quite a few discrete transistor frontends for his amps, often with some differential aspect. However, I can't see how this could possibly beat an integrated implementation - IC designers can match transistor sizes accurately, use laser trimmed resistors, slap in a current source at will, and use whatever other tricks I'm unaware of, to make a well matched differential amp. How can a discrete implementation possibly compare with that?

[aneevuser]

For the record, NE5532 has over 30MHz GBW at frequencies that matter and for NE5534 it's over 60MHz and perhaps up to 80MHz from some manufacturers, IIRC.

The trick? Gain falls faster that 6dB/octave starting at a few hundred kHz.

I now have time to come back to this. I'm not seeing the trick - can you elaborate? In fact, I'm not sure that I understand what you're saying - is it that effective GBW is a function of overall amp bandwidth, rather than the constant 10 MHz or whatever quoted in the datasheet?

[aneevuser]

NJM2122 is not much pricier and very low noise.

This looks like a promising replacement for the NE5532 - same pinout, low price - but...this datasheet:

https://www.njr.com/semicon/PDF/NJM2122_E.pdf

quotes only equivalent input noise voltage - which is very good - but doesn't mention input noise current. I'm suspicious that it's appallingly bad. Anyone know what's going on here - AoE doesn't list it so no joy there.

[magic]

So you discovered the joy of Japanese datasheets. JRC, NEC or Mitsubishi, all that stuff was marketed for audio products and usually only specifies noise in terms of total noise output from some RIAA preamp or other "typical" audio circuit. Oftentimes A-weighted.
 

That being said, this one has a "noise versus source resistance" plot. Of course it's the total noise over some unspecified bandwidth (lemme guess, 20-20k) with a few weighting schemes to choose from, but you can see the general pattern: it's about 1.5x worse at 200Ω than at 10Ω. Since Johnson noise of 200Ω is 1.8nV/rtHz, it fully explains the increase and no third noise source is apparent.

For the record, the minimum limit of current noise is the shot noise of the input bias current. Good bipolar opamps achieve this limit, at least above some tens of Hz. Shot noise density formula due to one Schottky: √(3.2e-19·I_B), so 1pA/rtHz for 3.6µA.

I actually bought one of those chips, but haven't played with it yet.

I now have time to come back to this. I'm not seeing the trick - can you elaborate? In fact, I'm not sure that I understand what you're saying - is it that effective GBW is a function of overall amp bandwidth, rather than the constant 10 MHz or whatever quoted in the datasheet?

Okay, I dug out my old notes. Actually, NE5532 had only some 20~25MHz GBW at audio frequencies. The 30MHz figure was for a fake 5532 from an auction site, I have no idea what's inside of that one but (surprisingly) it has to be some good shit

It means that open loop gain at 10MHz is 1x and at 1MHz is 10x, but at 100kHz it's about 230x, and then at 10kHz it's 2300x again.

[David Hess]

It is the two transistor equivalent to the dual operational amplifier differential preamplifier we discussed earlier.

This confuses me a lot. Self seems to have quite a few discrete transistor frontends for his amps, often with some differential aspect. However, I can't see how this could possibly beat an integrated implementation - IC designers can match transistor sizes accurately, use laser trimmed resistors, slap in a current source at will, and use whatever other tricks I'm unaware of, to make a well matched differential amp. How can a discrete implementation possibly compare with that?

The two transistor version is a differential tranconductance amplifier with a constant resistance load to convert the output currents into a differential voltage.  The differential voltage gain is the collector resistance divided by the emitter resistance and this configuration is commonly used in true instrumentation amplifiers.  Check out figure 6 in the Analog Devices MAT-04 datasheet which could work as a pretty good microphone amplifier.

The matching of the transistors is important in DC precision applications but less so in AC applications.

The operational amplifier version works considerably differently but produces the same result.