Voltage noise in opamp inputs.. how much is too much?

As I chug along on one of my many projects, I came across a fork in the road. Go with the normal impedence converter consisting of discrete BJTs as the frontend to an opamp, or use the inputs of an opamp straight from the source?

As I figure, the opamp already has BJTs or FETs on it's inputs so why series these parts? why not find an opamp with low noise inputs?

I came across some $$L schemos and behold.. the 9k uses opamps straight on the buss with no BJTs before it. I guess I was on the right track here.

Searching through datasheets and info I see that the BJT front should be low noise, but how low is good enough?

the mighty MAT04 that I was planning on using previously has a voltage noise of 2.5nV/rthz. the AD829 that $$L uses are 1.7nV/rtHz!. the AD812 is 3.5 nV/rthz. The 5532 is 5nV/rthz.

besides other qualifications like slew and drive capability, how low should the noise be? what is the threshold of tolerable?

:thumb:

..5534AN is 3.5 nV/rthz as well..

As a rough rule-of-thumb, you can gain some 6dB of noisefigure at 600Ohms by optimizing using esoteric techniques, compared to 5534.

Jakob E.

sure.

Lets take a look at the well known lm394, 1.8nV/rthz, followed then by 5534s.. it's used in the $$L4k mixamp whereas the ad829 takes it's place in the 9k, again followed by 5534s..

I don't want to take a "shoot for the lowest noise" approach without knowing why I'm doing so. there are other opamps that I would choose over this one if I knew what acceptable noise was.

thanks!

:thumb:

If I knew what acceptable noise was.

Click to expand...

Who is the customer: you or me? It looks to me as if it weren't me so it must be your turn to say what acceptable noise is. :wink:

As a guideline:
* specify the dynamic range you are looking for for a given number of summing channels (say 100 dB for 24 channels)
* check the maximum level you get with the given supply rails (say +20 dBu)
* that'll give you the maximum output noise (-80 dBu)
* set up an Excel spreadsheet that calculates the output noise as a function of source and feedback network impedance as well as opamp voltage and current noise
* play around with your design 'till you like the result

To say it loud again: you cannot just look at one figure (i.e. voltage noise); noise is a function of source and feedback network impedance as well as opamp voltage and current noise! Most low voltage-noise amplifier have high current noise and vice versa. If you don't know how to choose a suitable summing network impedance a "low-noise" opamp is not going to help you.

Let's take a look at the well known LM394, 1.8 nV/sqrt(Hz)

Click to expand...

I guess you meant 0.8 nV/sqrt(Hz).

Samuel

Who is the customer: you or me? It looks to me as if it weren't me so it must be your turn to say what acceptable noise is. Wink

Click to expand...

true. I am just wondering what would be the average acceptable noise in the design community.

To say it loud again: you cannot just look at one figure (i.e. voltage noise); noise is a function of source and feedback network impedance as well as opamp voltage and current noise

Click to expand...

I appreciate your concern but I understand this completely. I was sure that I mentioned that I am only asking about this and not other charachteristics at this time but i see that I did not state that clearly, sorry. I am trying to get a firm understanding of each piece of the puzzle as I go.

I guess you meant 0.8 nV/sqrt(Hz).

Click to expand...

Nope. the National datasheet shows 1.8nV/sqrtHZ:

http://cache.national.com/ds/LM/LM194.pdf (bottom of page 2)

Nope. The National datasheet shows 1.8 nV/sqrt(Hz).

Click to expand...

Well, that's at 100 uA. With higher currents it get's down to 0.8 nV/sqrt(Hz).

I am just wondering what would be the average acceptable noise in the design community.

Click to expand...

I think you really need to specify more: how many (min/max) channels? What summing resistor value? Balanced/unbalanced network? Active/passive network?

Of course I can just spite out a number and tell you that 10 nV/sqrt(Hz) is good enough--now you go and build a 128 channel summer with 1k summing resistors and a few OPA2604s, listen to it and notice that it is much too noisy.

Or I could tell you that you need 0.5 nV/sqrt(Hz) for "true low noise"--now you go and build a 2 channel summer with 100k summing resistors and a couple of LM394s, listen to it and notice that it is much too noisy as well. :grin:

Samuel

Well, that's at 100 uA. With higher currents it get's down to 0.8 nV/sqrt(Hz).

Click to expand...

Ah I see.

I think you really need to specify more: how many (min/max) channels? What summing resistor value? Balanced/unbalanced network? Active/passive network?

Click to expand...

24 channels max is the plan for now. I am working with 3k summing right now, active network.

I would suggest roughing out a design without specifying the opamp. Figure out how much noise it will generate on its own (thermal noise). Then look at various opamps (both voltage noise and current noise) with your design and look for the one that adds the least noise to what's already there, or at least a small enough amount of noise that you consider it acceptable.

Peace,
Paul

24 channels max is the plan for now. I am working with 3k summing right now, active network.

Click to expand...

In this case I would say that 5 nV/sqrt(Hz) or lower is OK for practical purposes. It might measure better with 1.5 nV/sqrt(Hz), but I don't think that the advantage counts much in practice.

I quickly calculated a few numbers for an unbalanced 24 channel active summer, 20 kHz BW:
* JH-990: -103 dBu
* LT1028: -102 dBu
* NE5534: -95 dBu
* OPA604: -86 dBu

Samuel

I also plan on using the AD812 if possible. superfast but primarily currentfeedback and a dual opamp. I liked $$L's usage of the ad829 but I am thinking I could get away with the AD812 with some tweaking and use full currenfeedback instead of the pseudo currentfeedback they used.

In a situation like this would you rather use an inverting/virtual ground setup or a direct non-inverting driver? Would you rather use the interfacing opamps as buffers with gain of 1 and use another stage for gain or use the interfacing opamps for gain and use the next stage for low gain driving? Assume the use of a instrumentation opamp setup.

oh and thanks for the advice Samuel and Paul!

In a situation like this would you rather use an inverting/virtual ground setup or a direct non-inverting driver? Would you rather use the interfacing opamps as buffers with gain of 1 and use another stage for gain or use the interfacing opamps for gain and use the next stage for low gain driving?

Click to expand...

Always put at least 20dB of gain in your first stage so that the first stage governs the noise performance. This is best achieved using the instrumentation configuration. You can then design for drive and power bandwidth in the output stage.

Inverting input stages have their attractions, particularly as the opamp does not have to deal with common-mode signals, but the source impedances become part of the gain equation, and so are more suited to situations such as mixing busses where the impedances can be controlled rather than for microphone input stages.

True balanced inverting electronic inputs, i.e. no transformer, can be realised using the superbalanced configuration using two opamps, but the gain is still dependent on the source impedance.

Assume the use of a instrumentation opamp setup.

Click to expand...

already on it!

:thumb:

Thanks!

And so are more suited to situations such as mixing busses where the impedances can be controlled rather than for microphone input stages.

Click to expand...

Er, we are talking about mixing amps, aren't we? :?

And use full current feedback.

Click to expand...

I guess you are aware that CF amplifiers have very high current noise at the inverting input.

Samuel

Er, we are talking about mixing amps, aren't we? Confused

Click to expand...

:shock: I hope so, at least that is what I was talking about..?

I guess you are aware that CF amplifiers have very high current noise at the inverting input.

Click to expand...

Yes, I am aware of this although I have not gotten to a full understanding of this yet. However I am aware of the lack of current noise specs for a lot of opamps! Makes life a little more difficult.

:thumb:

Just wanted to say this is a very interesting thread! Thanks guys...

Hmmmmm.....
What's the problem? If we are talking about virtual earth mixing, then when you get to 24 channels or thereabouts, the mixing amplifier noise is well swamped by the combine thermal noise of the mix resistors multiplied by the current gain of the amplifier. The lower the value of mix resistor, the lower the noise.
The limitations are the drive capability of the channel amplifiers, and the increasing problems with ground paths as the impedances reduce.

My solution for some years has been to use balanced mix buses, eliminating the ground path problem and giving 6dB more headroom.
:grin:

With mix resistors at 2-3Kohm, noise is less of a problem than other factors like gain/bandwidth, and transient stability.... and the dreaded ground path. :shock:

When you get to 24 channels or thereabouts, the mixing amplifier noise is well swamped by the combine thermal noise of the mix resistors multiplied by the current gain of the amplifier.

Click to expand...

Not at all--look at the figures I posted earlier in this thread. Or do you have a hint what I could have done wrong while calculating them?

It looks to me as if a 24 channel active summer with 6k8 resistors behaves (noise wise) like an inverting amplifier with a 283 and 6k8 resistor--and it is hard to argue that voltage noise doesn't matter here...

Samuel

24 channels is probably too few for this suggestion, but what about summing in chunks of, say, 8 channels, then "summing the summers"? In theory the final noise flow will be a bit lower, plus you have the advantage of short mix bus circuit lines and each summer running at a lower noise gain.

Bri

Yes, that works well Brian..... We used to do that in the 70s on very large PA systems in London theatres. We mixed in batches of 16, and then mixed the mixes; it was the only way to get acceptable noise figures on a 100+ channel desk!

My "lab notes" are long-ago missing, but I recall a breadboard test I did 30 years ago.

Going from what my "then current" thoughts probably were for a 32 channel mix bus...

10K buildout R's into an unbalanced summer/ACN.

Test circuit #1 was 10K/32, or approx 312 Ohms, with one end grounded and the other into the inverting input of the opamp, with a 10K feedback R.

Test circuit #2 was 10K/8, or 1250 Ohms with the same hookup and feedback R, and then built times four. The output of each 8 channel summer/emulation was then summed into a 4-port summer. I seem to recall a 6 dB "noise advantage" with this configuration using 5534's.

My main interest was the much shorter bus lengths (12") , which otherwise on a desk with 1.5" wide modules would be at least 48" long for 32 channels.

Bri