Line Input and Channel Gain module for summing Amp

[joaquins]

In class B the spikes should be simultaneous so there may be some help and in class A no spikes at all which is great, which may be a major advantage on class A for maintain cleaner grounds.

JS

Power grounds are like a sewer so not part of the audio signal. 

JR

But aren't we referring the output to this? Of course a separate dirty ground will be used for LEDs and relays, but the dirt generated by the sagging current on the rails from class AB opamps will be present in the decoupling caps to the output reference we are talking...

JS

 

[JohnRoberts]

Power grounds are like a sewer so not part of the audio signal. 

JR

But aren't we referring the output to this? Of course a separate dirty ground will be used for LEDs and relays, but the dirt generated by the sagging current on the rails from class AB opamps will be present in the decoupling caps to the output reference we are talking...

JS

I'm not sure what you are referring to...

If a ground or more correctly a 0V node is used as one of two audio lines,  differential circuitry can make the audio hot be relative to any voltage riding on the  0V node, that way a later differential receiver will subtract out any 0V error.

IMO differential circuits are very useful.

JR

 

[joaquins]

I don't know if I'm confortable calling it 0V, since we know there aren't 0V but some noise is present across the node, then calling it ground makes more sense since it's easier to get the idea there aren't 0V there, or reference instead may be a better name.

Those spikes or any noise will appear as common mode, it won't be seen by the receiver but will be present, perfect CMRR isn't possible so at the some point some noise may appear in the signal. Also, in the case of a mix bus where many signals are converging, then many ground references are converging too, there is a problem, many different ground noises mixes together and interfere with each other, then crosstalk could occur in a reference path instead of in a hot signal path. I'm making the comparison between having a reference as clean as possible with as low current as possible in order to get minimal noise inserted there with this approach which treats the ground as a signal so any noise is in CM but there are some deliberated noise inserted to the ground.  It's all good in single signal path but I can see a problem in a mixer where you have to mix grounds and hot signals together. Using a balanced summing amp could be an option but you would be having 3dB more noise from the architecture when trying to cancel out the noise you've created on the grounds. I don't know what's the best approach for that, maybe summing the grounds with smaller resistors to the non inverting input of the summing amp and starting a new path for the signal and it's reference from there. Then attaching them together in a different point of the chain, while avoiding ground loops with this smaller resistors (smaller than the summing, so johnson noise isn't a problem).

I follow you in the differential approach, it's very interesting and clever, some clear example is the 4 wire going to the fader you have mentioned many times, I'm just not sure what happens with this in certain cases, like interfacing the outside world or summing stages.

JS

 

[PRR]

> Reduced pot to 1K impedance (INA134 is more than capable of driving)

Why? The INA134 already has a 7uV or *245K* noise resistance. The pot can be anything under 40K and do "nearly no" added hiss.

I don't think 5V max output is "puny" for in-studio lines. What is the true dynamic range of the converters? (It's never as much as the number of bits implies.) 120dB? Then converter hiss is 5uV. It is likely worse (but I have not kept-up with the latest designs). If it is truly 117dB, then converter hiss equals INA hiss and noise figure at the INA is 6dB. Old data I have found suggests that 100dB is doing good, converter hiss nearer 50uV, so INA and pot hiss is total non-issue.

You are getting lots of Deep Thoughts here, but I think most of it is beyond what the job calls for. If the converters are next to the mixer in a benign room, I'd radically K.I.S.S. The converter "balanced" outputs are probably single-ended plus an inverter. Figure out which is the direct feed, and take it single-ended thru a short wire to the mixer. Go right to a 10K pot, 5K mix resistors, and a mix-amp with gain. You save three opamps (one in the DAC, the INA, and per-channel). Not that I care about your budget, but less monkey-motion is less signal corruption. "EMC filtering" should not be needed if working in one benign room. (Sure is when working portable.) With just one gain-stage, decoupling and ground questions are simpler.

Aside from a question whether INA134 can drive 1K, the low value throws heaps of current into the "ground". For any practical ground-bus, LF cross-talk rises.

Straining for gnats is the enemy of making music. Some pots and a 5532, you can have excellent technical quality and be making kewl toonz by Wednesday, instead of being chained to a soldering iron for weeks.

 

[bjoneson]

Wow... whole lot of information here. I agree with PRR on the fact that the simplest and perhaps even technically superior design that would fulfill my requirements would be a direct feed to a pot and bus resistors with a 5532 based ACN. That said, being chained to a soldering iron for a few weeks in the garage might be a nice reprieve from the chaos of 4 kids running around the house.

But in all seriousness, I'm eyes wide open on the fact I'm completely over engineering for the use case. But it's not unlike a lot of other things in life. This DIY business is an outlet. Some people remodel their kitchens with granite counter tops and gas fired stainless steel stoves, when in reality some formica and George Foreman grill will get the job done more than adequately. (I'm not saying a 5532 is a George Foreman by any stretch, but maybe Formica? It's technically superior, just not as sexy?). All of this touchy feely subjectivism definitely competes against the empirical engineer in all of us, but at least I'm aware of it. 

All that being said, I did have a question still lingering about the pot value. I was inclined to lower the value based on what I had read in Doug Self's book "Small Signal Audio Design":

" The resistance value of a volume control should be as low as
possible, given the loading/distortion characteristics of the stage driving it. This is sometimes
called ‘low-impedance design’. Lower resistances mean:
1. less Johnson noise from the track resistance;
2. less noise from the current-noise component of the following stage;
3. less likelihood of capacitive crosstalk from neighboring circuitry;
4. less likelihood of hum and noise pickup."

I agree that #1 can be tossed aside since it's easily swamped by other noise. You mentioned the amount of current being dumped to ground on these low impedances, risking LF crosstalk. It seems in contradiction to #3, would you mind expanding further?

@Abbey eluded to this as well.

 

[joaquins]

The drawbacks are missing there, more current, so more crosstalk from it, plus more distortion, which is mentioned but not explicitly as a drawback. The distortion is not only the distortion on the amp stage driving that source but also due to the noise in the rails//ground if layout isn't managed properly.

About current noise, dividing voltage noise by current noise will give the optimal input resistance, in the case of most BJT input opamps is few kΩ, in the case of JFET are more close to MΩ, that tells you you are not benefitting from going much lower than that figure, where current and voltage noise will have the same impact, if you have lower resistance current may give a little bit less but never get much better performance. The other thing is picking up noise, well, if the environment isn't a problem which shouldn't inside a metal case in a friendly room, and if you are considering going unbalanced as PRR recommendation the noise won't be picked uo in the fader, but in the line before it and there's nothing you can do about it. I forgot the one you actually asked, between two conductors separated by a dielectric (two traces from adjacent channels and the board or just air) there is some capacitance, this could cross couple HF between one channel and the other producing crosstalk, if your impedance is lower the capacitor will be still coupling both traces but at a higher frequency, so in band coupling will be less problematic. Still I hardly see this is a problem, 10k fader is still 2k5 maximum output impedance to consider in your current input noise of the next stage, coupling or whatever, I don't see any benefits going for 1k in this case.

I do like 5532, 5534 has less noise and higher slew rate, but needs external compensation for unity gain stability, may be a good choice for the mixer. The thing with DOA is they are designed for specific tasks, 990 has really low voltage input noise which make it applicable for low impedance sources, but it also has really good driving capability compared to ICs, this is because they can pick completely different devices for each transistor in the circuit, in ICs they are all pretty much the same.

JS

 

[bjoneson]

Makes sense, I'm beginning to think 5k might strike a better balance in my case. 10k seems to be a common value these days, when many pots / sends have to be driven, but in my case it's just the one.

 

[joaquins]

The problem is to find 5k faders, if you are using pots you are ok.

JS

 

[bjoneson]

The problem is to find 5k faders, if you are using pots you are ok.

JS

Yes, pots in my case. Lots of 5k log out there from Bourns, Panasonic, etc...

 

[abbey road d enfer]

Doug Self's book "Small Signal Audio Design":

" The resistance value of a volume control should be as low as
possible, given the loading/distortion characteristics of the stage driving it. This is sometimes
called ‘low-impedance design’. Lower resistances mean:
1. less Johnson noise from the track resistance;
2. less noise from the current-noise component of the following stage;
3. less likelihood of capacitive crosstalk from neighboring circuitry;
4. less likelihood of hum and noise pickup."

I agree that #1 can be tossed aside since it's easily swamped by other noise. You mentioned the amount of current being dumped to ground on these low impedances, risking LF crosstalk. It seems in contradiction to #3, would you mind expanding further?

@Abbey eluded to this as well.

I don't think I've eluded; in fact I mentioned the fact that running 15mA peak current was indeed a concern regarding x-talk.

 

[bjoneson]

Doug Self's book "Small Signal Audio Design":

" The resistance value of a volume control should be as low as
possible, given the loading/distortion characteristics of the stage driving it. This is sometimes
called ‘low-impedance design’. Lower resistances mean:
1. less Johnson noise from the track resistance;
2. less noise from the current-noise component of the following stage;
3. less likelihood of capacitive crosstalk from neighboring circuitry;
4. less likelihood of hum and noise pickup."

I agree that #1 can be tossed aside since it's easily swamped by other noise. You mentioned the amount of current being dumped to ground on these low impedances, risking LF crosstalk. It seems in contradiction to #3, would you mind expanding further?

@Abbey eluded to this as well.

I don't think I've eluded; in fact I mentioned the fact that running 15mA peak current was indeed a concern regarding x-talk.

I appologize, again, I'm a work in progress learning. Could you describe where / how the crosstalk occurs?

What's the physical interaction creating crosstalk? Capacitive coupling with adjoining traces? Some type of feedback through the ground system?

I can take you at your word, but clearly there's some tradeoffs happening between the advantages of low impedance, and the disadvantages of high current, and I'm just trying to better understand them so I can make more informed decisions moving forward.

 

[bjoneson]

I did just turn up this SoS article which seems to describe the phenomenon as it relates to shared ground connections in headphones:


http://www.soundonsound.com/sos/jul13/articles/qanda-0713-1.htm

 

[bjoneson]

So, after reading, I think I was able to self educate a bit. It seems the crosstalk is introduced via the ground trace resistance causing a voltage to appear on adjoining circuitry sharing that ground trace. Obviously the higher the current, the greater the voltage per Ohms law.

I suppose this could be mitigated with careful consideration, perhaps even dedicated traces for these high current returns.

 

[abbey road d enfer]

Those spikes or any noise will appear as common mode, it won't be seen by the receiver but will be present,

Will they? Indeed if the opamp was pure class-B, the current in each supply leg would be a rectified sine-wave, but when both currents are combined - by being decoupled to a single point - the resulting current will be a pure sinewave. Now if the load reference is this very same decoupling point, the resulting node current is zero.  This valid as long as the transient current is absorbeb by the decoupling caps, so at LF, the current spikes have to be absorbed by te rails; that's why large value electrolytics are necessary, whose decoupling point must be chosen as not interfering with this beautiful scheme. 

Also, in the case of a mix bus where many signals are converging, then many ground references are converging too, there is a problem, many different ground noises mixes together and interfere with each other, then crosstalk could occur in a reference path instead of in a hot signal path.

That's right. That's the real art of designing a mixer, combining solid understanding of parasitic current circulation, use of brute force for the ground reference, taking advantage of differential structures and all the various "tricks" that have been discussed here.

  Using a balanced summing amp could be an option but you would be having 3dB more noise from the architecture when trying to cancel out the noise you've created on the grounds.

I've always found that the advantage of getting rid of most of the longitudinal noise largely surpasses the increase of noise due to the increased bus impedance. [/quote] I don't know what's the best approach for that, maybe summing the grounds with smaller resistors to the non inverting input of the summing amp [/quote] Indeed, they should be of lower value in order to maintain about the same level of noise as when unbalanced. The only limit there is that these resistors must be significantly larger than the parasitic resistance between the summing amp input and teh "ground" of any channel. In particular when the mixer uses ribbon cable, this resistance can be significant. A 48-channel mixer has looms about 4 m long, with a resulting resistance of about 0.5 ohm, so using 100 ohm resistors there offers the possibility of reducing longitudinal noise by about 40dB, which is a lot. Contrary to Johnson, longitudinal noise doesn't have to exist, so in theory it could be completely cancelled. In fact it's very often the dominant noise so it's worth taking the time and measures to improve it.

and starting a new path for the signal and it's reference from there.

The preferred arrangement here is to have a resitor of identical value tied to a stiff ground that will be the new reference point.

I'm just not sure what happens with this in certain cases, like interfacing the outside world or summing stages.

That's when you have to analyse the system, using the same approach than with your "parasitic resistance" example.

 

[abbey road d enfer]

So, after reading, I think I was able to self educate a bit. It seems the crosstalk is introduced via the ground trace resistance causing a voltage to appear on adjoining circuitry sharing that ground trace. Obviously the higher the current, the greater the voltage per Ohms law.

I suppose this could be mitigated with careful consideration, perhaps even dedicated traces for these high current returns.

That's exactly what I'm exposing here. The physical arrangement of "ground" tracks must make sure that output return currents are separated from the inputs.

 

[joaquins]

Those spikes or any noise will appear as common mode, it won't be seen by the receiver but will be present,

Will they? Indeed if the opamp was pure class-B, the current in each supply leg would be a rectified sine-wave, but when both currents are combined - by being decoupled to a single point - the resulting current will be a pure sinewave. Now if the load reference is this very same decoupling point, the resulting node current is zero.  This valid as long as the transient current is absorbeb by the decoupling caps, so at LF, the current spikes have to be absorbed by te rails; that's why large value electrolytics are necessary, whose decoupling point must be chosen as not interfering with this beautiful scheme. 

In the scheme I draw they kind of do appear as CM since rG2 is much smaller, and even could be eliminated with some brains in the layout. Then the signal is referenced to it, the output takes as reference that voltage on the decoupling point and the next stage is taking that same voltage as reference, so effectively they are CM at the output.

As you mentioned the current going to ground in a perfect class B should make a copy of the signal. In a class A there shouldn't be current going through the caps. The major problem I see is with class AB amps, where the spikes occur at different times and may introduce some HF current even with LF signals, the problem comes in when the CMRR of the next stage is not so good, and since class AB are the most common amps in most applications I see this as a thing to give more attention.

I don't get the last part, how they will interfere? The caps I draw should be this big caps and smaller ones in parallel if that helps for something right?

I've always found that the advantage of getting rid of most of the longitudinal noise largely surpasses the increase of noise due to the increased bus impedance.

The 3dB I was taking was due to add another opamp as dual summing, then make the difference, like truely balanced summing amp, not the Johnson's noise but the EIN of the other opamp in the summing configuration. I can see how the Johnson's noise of the smaller resistors shouldn't be a problem at all, and then I see how this makes a better approach.

The preferred arrangement here is to have a resitor of identical value tied to a stiff ground that will be the new reference point.

A 100Ω in your example, from the input reference to the decoupling point of the summing amp and take this point as reference for the next stage?
Once I have all this where do I join all my grounds? Close to the bus seems right, like, the last point from each stage before the ground resistors to the ground bus and the new ground reference of the summing amp where they decoupling caps are attached seems logic, but maybe is too close and could interfere with the summing structure providing a shorter path (without the resistors) for the current to flow to the summing amp and make a mess again... Using inputs and output points looks an even worse case making the noise at the summing points higher even if it should get cancelled out with this structure.

I think is time to redraw the boards for my mixer, since I already build only a few boards of the channel amps as prototypes this shouldn't be a problem, the summing amps already have a separate pin for the bus ground with a jumper so they can remain as they are, but still should modify a few things about them when the time is right. About the summing/bus boards I have a bigger problem since I already have most if not all of them but is a 1m 16 channels and they are all small simple boards so I could start again with those once the thing is working. I didn't include the sensing ground for each bus when I first build it. Maybe summing/bus boards and summing amp could be redesigned together. In any case I will make this analysis when I have the time with all the parasitic resistors for the summing stage. As this project is kind of in the freezer and this summer is not going to get any better for what I see I have plenty of time to review the design, and if I want to work on it I have a lot of other stuff to do like building a lot of DOA, I maybe design my own before I finish the mixer, that would be nice.

Thanks for the help Abbey.

JS

 

[bjoneson]

Hey just a heads up. I got pretty overwhelmed by information once we got into the "grounding" discussion. I'm learning to more appropriately consider "return currents" as opposed to a "grounding system".

I'm still digesting information, and trying to learn, but I did find an article that really opened my eyes and better understand what's going on, and thought I'd share:

http://www.hottconsultants.com/pdf_files/ground.pdf

 

[bjoneson]

Alright... I'm at that point where I'm cramming more information into my head than I can make use of right now. I did go ahead and put together a simplified schematic showing some of the system level considerations. I've identified all of the "ground" nodes.

Trying to determine the best way to get all of the return currents and references connected to minimize crosstalk.

Absent from the diagram is the PSU module, providing the power rails and a power common return to the power transformer center tap. I'm assuming a connection from the center tap to chasis / earth ground. Beyond that I'm looking for advice / suggestions on routing all of these nodes.

Based on my reading... I "think" I understand a bit about where the current wants to flow, but hoping some folks can help fill in some blanks...

ND1 (Input Module)... the source of any current here I believe is both the connected device (which would want to return via chasis ground?), and the bias current of the input opamp which is sourced from power common.

ND2 is the "ground point" between the bypass caps, I assume this node is providing a return path to power common.

ND3 / ND4 are interesting, because since the source is the line input opamp, I believe current here wants to return to ND2 at the bypass caps?

ND5 I believe has bias current from the opamp wanting to return to power common, but it also has current from the feedback loop at the output which I assume again wants to return the bypass caps (ND6).

ND6 - Another bypass cap "ground point" I assume wants to direct current back to power common.

ND7 - The intent here is to create a ground point to increase "offness" from capacative coupling within the switch.  Again, also looks like a current return to the bypass caps of the channel gain amp, but should be a zero volt reference.

I think I'm going to stop there for a moment and see if I'm even on the right track.

Many thanks for any help!

 

[joaquins]

I kind of lost it, with all the diagrams and that, but for the input channel looks like those nodes connected in the order they are numbered to a ground trace makes sense, then from the last node you should take the reference for the next stage. That bus ground connects to the ground on the summing amp and from there the other two nodes on that. One detail we discussed is to put small (~100Ω) resistors between the ground bus and the nodes you mentioned, also another resistor from the first node on the summing amp to the second. Then you connect all your channels and mix amps through a common node where are all the same, I think I'd choose for this the last decoupling stage of the input channels and the first one of the summing amps, but that was just my first guess and might not be optimal.

JS

 

[abbey road d enfer]

Alright... I'm at that point where I'm cramming more information into my head than I can make use of right now. I did go ahead and put together a simplified schematic showing some of the system level considerations. I've identified all of the "ground" nodes.

Trying to determine the best way to get all of the return currents and references connected to minimize crosstalk.

Absent from the diagram is the PSU module, providing the power rails and a power common return to the power transformer center tap. I'm assuming a connection from the center tap to chasis / earth ground.

No. The "common return" is to the reference point of the PSU, which is the point the regulators (or the last smoothing cap if there's no regs) are referenced to. The xfmr's center-tap is a point that is best left floating at the end of an otherwise unconnected antenna (more in the original sense of "appendage" than in the radio-electrical sense).

ND1 (Input Module)... the source of any current here I believe is both the connected device (which would want to return via chasis ground?), and the bias current of the input opamp which is sourced from power common.

ND2 is the "ground point" between the bypass caps, I assume this node is providing a return path to power common.

ND3 / ND4 are interesting, because since the source is the line input opamp, I believe current here wants to return to ND2 at the bypass caps?

ND5 I believe has bias current from the opamp wanting to return to power common, but it also has current from the feedback loop at the output which I assume again wants to return the bypass caps (ND6).

ND6 - Another bypass cap "ground point" I assume wants to direct current back to power common.

ND7 - The intent here is to create a ground point to increase "offness" from capacative coupling within the switch.  Again, also looks like a current return to the bypass caps of the channel gain amp, but should be a zero volt reference.

You need to connect all theses "grounds" to each other with resistors (it offers many possibilities) and look at the consequences.
Remember that some of these points are inducing current, some are just receiving and some are both simulatneously.
ND1 is a receiver (forget about bias current), ND2 is a "transmitter" (assuming PSRR is large enough) and ND3 is both (it transmits the current flowing through the fader AND receives part of the parasitic voltage that is imprinted to it. ND7 is just a "transmitter", and ND5 is typically both; that shows the importance of not making the NFB network values too low because it increases the parasitic current.
When you establish connections between the various "grounds, you will see that the hierarchical arrangement is the one that creates the less cross-talk between the various paths. And indeed ND3 should be as close as possible to ND2, same for ND6 & 7.