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

[TedF]

And even more important Brian.... those dreaded ground paths!!!
Manageable length buses with some heavy ground strips linked just from those channels to their mix-amp and the system can stay nice and clean. But when it comes to complex multi-track routing, the answer has to be balanced buses.

 

[PRR]

Forget noise voltage and current.

Forget impedance.

Resistor noise POWER is CONSTANT.

Doesn't matter if the bus is 1 ohm or 1 Meg. Same noise power.

Once you have defined the number of inputs, mix-loss is also defined.

Then your ultimate signal/noise ratio (for thermal noise) is only determined by how much power you can extract from your sources (channel amps or pan-pots).

And some piddly details like impedance, but in theory these can be resolved with transformers.

For decades we were stuck with +/-15V chips good for maybe 100mW, which we often split 4 to 8 ways. Actual input power was on the order of 10mW or +10dBm.

A 24-input resistive mixer has 28dB loss. We have -18dBm at the bus.

Over a hifi audio bandwidth, resistor noise will be about -126dBm. (Not -126dBu unless the resistor happens to be 600 ohms.) Ultimate S/N is 108dB.

Assuming you can actually push your channel amps to +10dBm cleanly, this will generally be ample. In fact thermal noise is often not a problem.

As TedF says, in modern multi-routing, and with class AB output stages, you get a crap-level on the bus and ground that may be far higher than thermal noise. Simple crosstalk may be acceptable. Junk from AB outputs often sounds nothing like any of the inputs, and may be quite annoying. So high-performance multi-routing boards usually have to gulp the cost and go to balanced (ground-ignoring) mixing.

I suspect TedF and Bri will confirm: a large console can not be designed on paper. Do your homework, but then you have to build it. You will find instability, radio reception, mystery crosstalk and crap, and all sorts of other evils that don't live on the drafting board.

As for key values: plagiarize. 5K to 22K mix resistors are generally fine. A 24-in mix stage has enough to do without actually providing gain (if you really mix 24 hot channels, you may need a loss in the mixamp). Heroics like +/-50V supplies or 75-ohm sources do buy S/N, but at very high cost (especially since everything should be doubled-up before you go for heroics).

The choice of mix-amp details can perhaps be left for last. Build the bus and populate the channels. Then build a 5534 mixamp, an AD812, a 5534+LM194 mixamp, a dozen-6DJ8 mixamp, and listen. Sorting them for noise at bus impedance may be easy (or not, since they give different noise spectra). But mixing multiple music signals can really confuse an amp, and some will be more euphonic than others. And the mix-amp module is something you can retro-fit, in development or even in the customer's lap. (You could even swap mixamps depending on the job: a TL072 for techno-pop, a fat-tube mixamp for an acoustic orchestral project.)

 

[TedF]

PRR, why does your avatar look just like Steve Dove??

Rah Rah!!! yes, of course noise power is constant, but that fact is still confusing to most.... But the explanation that the more power you can squeeze out of the channel (with lower R values) the lower the noise will be, is easier to understand.

And yes, those different mixamp designs will provide different flavours; and those flavours are various sorts of distortion depending on the various imperfect factors in the amplifiers. Trying to measure audio distortion to get meaningful figures is very like looking at a Rembrandt and copying it in a 'painting by numbers' kit.
I agree absolutely with PRR.... be a real artist and listen.... but be constantly on guard against snake oil! :shock:

 

[Brian Roth]

As PRR mentioned, getting the mix buses "right" can be a mind boggling experience.

When I built that desk back in 1977, I had everything nailed down on paper/prototyped EXCEPT the mix buses.

I fought all sorts of problems in the final unit, which required various bizarre tweaks to make the buses stable/quiet.

Hence my interest in the "short bus" concept.

Bri

 

[Svart]

good advice!

I am just trying to bring the cream to the top first, avoiding as many of the "snake oil" designs as possibly with careful learning.

 

[Brian Roth]

One other issue to ponder....

I'm gonna assume a 24 channel desk for discussion...24 input modules into 24 channel mix buses *and* 24 input channels into a stereo mixdown bus.

There are two ways to connect the 24 channel busing switches on each input module via a SPDT switch (assuming unbalanced signal onto the buses). In no particular order:

Method "A" feeds the signal to the N.O. contacts of the switches. The N.C. switch contact is grounded. Each switch wiper then connects to the appropriate bus via a summing resistor.

Method "B" feeds the input channel's signal via a summing resistor into the wiper of the switch. The N.C. contact is grounded, the N.O. contact ties to the mix bus.

"A" has the advantage (?) of ensuring a constant resistance onto the mix bus, particularly important with a "passive" mix amp (ie, a mic preamp sort of stage like was used in Years Past, and now popular again with devices like the Folcrom).

"B" has the advantage that the noise gain from the ACN is optimized since no more than the required number of inputs for the project are ever hanging onto the mix bus.

Now let's ponder one channel of the stereo mix bus. In that case, all 24 channels HAVE to be connected onto the mix bus.

Hence, we now have 24 mix R's hanging onto the bus, thus raising the noise gain of an ACN, or lowering the levels onto the bus with a "passive" summing amp. In either case, the noise floor HAS to be higher with 24 channels on the bus vs., say, 2 channels as might be typical in a multitracking example.

Now, in the case of 24 inputs with 24 channels of possible bus destinations, we run into another snag. Method "B" requires that the stage driving into the mix switching circuit can sucessfully drive 24 bus resistors which are now in parallel either to ground or onto the ACN (virtual ground) bus. Assuming 10K resistors, that means the driving stage has to provide enough current into 416.6666 Ohms.

Under typical circumstances (one channel feeding a couple of multitrack buses), Method "A" alleviates the loading problem in each channel at the expense that the ACN is always running "worst case" noise gain, or that the "passive" bus is running at the worst case loss.

TANSTAAFL.

Bri

 

[TedF]

Hi Bri,
Instead of your method 'A', the real ideal as far as noise and overload, is to forget the ground on the switch wiper. If the series R is kept lowish, then the crosstalk penalty should be bearable.... it's a question of PC layout! :?
Does the method onto the stereo bus really matter? The channels are connected most of the time anyway.
But really, most of these problems are eliminated when you use balanced buses.... and I know I keep banging on about it, but the biggest problem by far in unbalanced mixing systems is GROUND PATH. :guinness:

 

[Kit]

Allright, thats it....this topic is getting bookmarked.

Lotsa great info. :thumb:

BTW, maybe this should go to the "Mixers Meta"?

 

[Svart]

i was designing these busses to go into an existing mixer but I think I am going to design them for the new channel strips I am designing and use balanced busses.

 

[Svart]

yeah I had those spec'd on an earlier design. I figured that if I could find an opamp with low enough noise I could just do without with careful design.

 

[Brian Roth]

[quote author="TedF"]Hi Bri,
Instead of your method 'A', the real ideal as far as noise and overload, is to forget the ground on the switch wiper. If the series R is kept lowish, then the crosstalk penalty should be bearable.... it's a question of PC layout! :?
Does the method onto the stereo bus really matter? The channels are connected most of the time anyway.
But really, most of these problems are eliminated when you use balanced buses.... and I know I keep banging on about it, but the biggest problem by far in unbalanced mixing systems is GROUND PATH. :guinness:[/quote]

Hey, Ted. Ummmm...neither of the "classic wiring schemes" that I listed involved a grounded wiper. I mentioned both "A" and "B" since they are found in desks made the last 30-ish years, which is NOT to say that anyone had the optimal solution.

"A" would seem to be mandatory if implementing a "passive" mix bus, particularly with an input transformer on the summing amp, in order to keep the impedance on the bus constant. Hence, the summing R's need to connect to either the signal source (assumed to be a very low Z) or to ground.

However, I can certainly see that a balanced implementation of "A" could be concocted whereby the "input side" of the summing resistor pair was shorted together in the "off" condition to reduce crosstalk while still maintaining a constant impedance on the mix bus. But, that would ensure that the summing amp was always running at worst-case noise gain (or makeup gain in the case of a passive summer..which is the only reason I can see to use "A" when a transformer coupled input is required).

Dammit Ted <g>, you're making me think "outside of the box" now!

I still would believe that a balanced version of "B" would be preferable with an ACN to ensure low levels of crosstalk, especially if the "small" sort of pushbuttons (ALPS, etc) are used for the assign switches. I can visualize picofarads across the switch contacts.....

But I am probably wrong/living in the past....I will keep pondering all of this!

Bri

 

[TedF]

Let's all keep pondering....

I have tried transistor front ends on mixamps just to see if my arithmetic was right or wrong, and found that there was no real advantage, after connecting 3 or 4 inputs the dominant noise is from the channels or the mix resistors (unless you are using cheap chips as the mixamp!)

But don't lose sight of what the whole thing sounds like.

 

[Brian Roth]

Ted..OK, cool.

I should draw up the "classic'" A and B (unbalanced) schemos so that everyone can see the designs.

From there, we can mull over balanced busing concepts.

I am certainly interested!!

Bri

 

[TedF]

My point is that the internal noise of the mix amp is swamped by the noise of the channels/mix resistors.
If you have a mixamp with say 1nV/rtHz, is it significantly quieter than one of 1.5nV/rtHz when is running at 20dB gain and when it's 'seeing' the noise of say 24 10K resistors? I think not.

 

[Svart]

his was the motivation at SSL for the quasi-balanced 4K mods

That is on the 4k, the 9k uses the opamp directly after a series resistor of ~5k6 directly from the buss with no BJTs or other devices between.

 

[Samuel Groner]

My point is that the internal noise of the mix amp is swamped by the noise of the channels/mix resistors.

It is surely swamped by channel noise on most designs, but hardly by resistor noise (again, look at the numbers I posted). Your argumentation is sound but nonetheless I believe that it is worth spending a thought on how amplifier noise influences things, especially as the original post is about a summing box and not a full mixer where you don't have channel noise.

It's the same thing with microphone preamplifier noise--in about any practical circumstance the (by far) most important source of noise is ambient room noise. Nonetheless most designers don't accept an EIN of -120 dBu, altough it hardly matters if it's 5 dB better.

Samuel

 

[Svart]

my copy of the book is on it's way as we type.

 

[bcarso]

I recall a not-too-bad discussion of noise gain is also in Pease's Troubleshooting Analog Circuits book.

 

[TedF]

You guys jogged my memory... and I decided it was none too good, so this afternoon I did a little experiment..... I set up a test rig using a selection of dual chips as a balanced dual virtual earth amplifier with low value series R inputs (330 + 330 ohms) and measured the output noise with the 'channel' side of the resistors shorted together.
To make it meaningful, the results are quoted as effective 'dB below input' filtered 20Hz to 20KHz.
The chips were: LM33078, NE5532, TL072, and RAY4559.
the figures were: 126.8, 125.7, 114.9 and 119.5.
While in a mixer there are a host of other factors, this does give a rough guide to the sort of noise to expect..... using 3K3 feedback resistors and 10 channels, the noise attributable to the mix amp would be -106dB (using NE33078)... or have I got my sums wrong? it could be 6dB better than that. :?
I will let someone else point out that the 126.8 figure is mighty close to the thermal noise in the input resistors!

 

[Svart]

the 33078 is interesting and good but for some reason it just doesn't sound as good as the 5532 to me.. :?