inverting summing amp noise calculation / reduction

[mitsos]

I want to use an API-type ACA (active cobnining amp) like classic API's ACA-bo in a box with a few preamps, but also connected to a pair of DB25's for more inputs.  I read on ESP's site that because inverting summing amplifies noise more than the signal, it is best to disconnect unused channels.  I'm wondering how much noise 16 channels' input R's would induce (assuming 47K input R and 28K feedback R).   

Sorry to be asking this when I could just set it up and test, but I really don't want to wire up the DB25's only to find they are unusably noisy without some sort of way to switch them out. 

Speaking of switching them out, I figure I could use a toggle switch for each stereo pair, so 8 total, anyone have a better idea?

thanks for the help!

 

[joaquins]

With this config the noise gain is the number of channels plus one times the gain of a channel. So, for 16 channels it's 17 times the gain, so 28/47=0.6 so about 10x or 20dB. Then the actual S/N ratio depends on the noise at the non inverting input of the amp, which depends on the layout of the summing bus.

In any case you should consider the 17 times, not the 10 times figure, since your S/N ratio would be affected by that 17, not the 10, since your channels are attenuated those 4dB, so your noise gain/signal gain is 24dB. Always the N+1, no matter how many channels or how much gain the amp has.

JS

 

[mitsos]

Thanks neighbor!

I'm still unsure of whether to try this or not.  I have some DB25 PCBs from someone here, but I'd have to run signal to the switch then to the summing amp. Simpler to use switching TRS jacks but not enough space. 

BTW, non-inverting goes to ground through 100K in parallel with .01uF.  Not sure how this affects noise, I'd appreciate the enlightenment. 

Your numbers look like about what I was expecting/fearing. 

As for disconnecting the inputs, can I ground the "out"side of the input resistor, or is it better to disconnect it at the opamp side? If grounding is (about) the same, would a volume pot before the input resistor work OK?  Like an Aux send?

thanks!

 

[joaquins]

For a few channels (let's say 8) most of the time is better to switch the "input" of the resistor to ground so it doesn't act as an antena, it will still be affecting the noise gain of the summing amp but 9 times is something you can live with. The best way to do it is switching at the "output" of the resistor (the VE summing node) but you should tie the resistor to ground and do it without traveling a long way to the switch, usually electronic switches are the best way to go here, and as they are switching between  a VE point and ground, always 0V and ~10k summing resistors are much bigger than the resistance of the switch, which makes them quite good for the application, since no problem of rails limits or whatever. You don't need 15V audio switches, them capable of handling  1V would be good enough since they won't have any voltage at any time.

The resistor at the non inverting input is to compensate the resistance at both sides of the summing opamp to reduce offset, this kind of works in this application but not eliminates it since you are changing the resistance at the other input. I usually pick a value at 50% to 70% channels connected, this way is a good compromise to a usual application. The cap is there to reduce the thermal noise of the resistor. If you keep all the channels connected all the time or switch the "input" of the resistor to ground you should pick a value just like the resistance seen by the inverting input at this point, this will cancel the offset much better which could be quite good if you are trying to avoid the output coupling cap.

JS

PS: I used " when mentioning the "input" or "output" of the resistors so no one slap us on the face, but we get what we are trying to say...

 

[mitsos]

PS: I used " when mentioning the "input" or "output" of the resistors so no one slap us on the face, but we get what we are trying to say...

Gotcha.

I probably won't be adding relays, and 9 sounds better than 17!  I should have been more specific, there are L and R summing amps, so 8 channels per amp, so 9 is what we're talking about right?

The summing R and feedback R, I can change these of course.  But sticking with the 47K as an example, to pick a shunt R for the non-inverting input at 50-70% of the inverting R, you mean the resistance of 4-6 paralleled 47K resistors? four gives me about 11K, and 6 is about 7.8K, so something between those if going with 47K? 

BTW, is there any noise advantage with using lower (say 10K) input resistors, vs 47K,  assuming lower value R's have lower thermal noise?  Any disadvantage?

Thanks for the help!

 

[joaquins]

So yes, with 8 channels in each summing amp you have 9 as noise gain/signal gain no matter what resistors values you are using as summing and feedback or gain. I wasn't talking about relays but electronic IC switches, cheaper and smaller. You could go for some maxim ones which comes in a lot of shapes and colors, or many other brands out there.

About the non inverting resistor you got it right.

Yes, lower value resistors are inherently less noisy, as general rule higher the power lower the thermal noise but usually higher distortion. In a big mixer you have to make a compromise, you need to drive all the resistors for all the different bus from one opamp usually, and you have a lot of channels adding a lot of noise. In a simple summing board you have only one bus and only 8 channels, so both, noise and driving capability is much less critical. Usually 10k is no problem for any source, big enough so the source impedance won't affect the gain much, you may have few hundreds of ohms at most from your source, noise is low enough compared to line level signal and then you don't have to worry. If you have to drive 24 bus 96 channel mixer you have a problem and you probably want to split the driving amps and the summing amps to lower bus/channel count so you can reduce both, noise and distortion.

JS

 

[PRR]

> test, but I really don't want to wire up ....only to find

No work. You propose sixteen 10K resistors? Put one 625 Ohm To ground and listen.

> DB25's only to find .... some sort of way to switch them out.

Un-plug this DB25?

 

[PRR]

> lower value resistors are inherently less noisy

All resistors have the same hiss power.

Lo-R is lower hiss voltage. Hi-R is lower hiss current.

Which matters more depends on how you use them.

 

[mitsos]

> test, but I really don't want to wire up ....only to find

No work. You propose sixteen 10K resistors? Put one 625 Ohm To ground and listen.

Thanks! I needed a simple method!
 

> DB25's only to find .... some sort of way to switch them out.

Un-plug this DB25?

of course, but then someone has to reach behind a rack or whatever. I'd like to leave these connected but they won't always all get used.  Maybe going through a patchbay, but then you'd still have the "antenna"  from the patchbay to the back of the unit. 

Probably going to go with toggles in the end, the more I think about the pot idea, the more I think it will be detrimental to the panning done ITB. And I don't know that I can find IC switches easily around here...

> lower value resistors are inherently less noisy

All resistors have the same hiss power.

Lo-R is lower hiss voltage. Hi-R is lower hiss current.

Which matters more depends on how you use them.

I was thinking of johnson noise, which should increase with the R value, right?  But that's not what you're talking about I think.  What you say makes sense at an instinctive level though I don't know that I could explain it properly (noise is related to power dissipated?). 

Since we are talking about series resistors going to the inverting input of an opamp as "inverting summing amp", which one should concern me more?  My first guess would have been current noise since we are not dropping volts, but we are going to a "virtual earth" so in a sense we are, but even so, we are only talking about line level AC signals, not hundreds of volts, so...

Am I overthinking this?

 

[ruffrecords]

You have several sources of noise to consider:

1. The inherent noise of the amp
2. The noise in  the resistors
3. The noise on your 0V line
4. The noise from the input sources

The first three are multiplied by the VE noise gain. If the feedback and input resistors are the same then this is N+1 where N is the number of connected inputs. 1. is down to op amp selection. If it could not be got low enough not to worry about then nobody would use op amp VE mixers. 3. Is down to good construction and grounding techniques. 2. is never as bad as you might expect. The reason is  as you add more resistors the voltage gain goes up but the resistor noise voltage goes down because all the resistors are in parallel as far as their noise voltage is concerned. Nobody ever seems to mention 4. but it is often the worst culprit and one of the best reasons for disconnection VE feed resistors. Most sources will have a noise floor around -90dbBu (the VE 10K feed resistor has a noise voltage of less than -110dBu). The noise from summed sources goes up 3dB for every power of two so sixteen connected sources will add 12dB of noise and bring the ouput noise up to -78dBu.

Bottom line is that in most practical systems the noise from the inputs themselves is greater than that introduced by the feed resistors and the VE op amp so that is a very good reason to disconnect them.

Cheers

Ian

 

[mitsos]

Thanks Ian, like I said I tend to overthink things, but reading your post confirms that I should disconnect the unused channels if the DB25 will remain permanently connected to the DA outputs (or whatever other sources I guess).

Thanks!