Best Console Summing amp/Master Buss comp Ideas....

[leigh]

The schematic I was talking about is the schematic those distortion measurements were made with. An op amp will have X % THD operated open loop, and when you wrap NF around it that open loop distortion is reduced by the loop gain margin (ratio between open loop and closed gain.) I will ASSume the distortion numbers were made with comparable circuits.

Ah, my misunderstanding. On pg 25 of the LME49710 datasheet they describe how they did the THD measurement on that, using a resistor to extend "the resolution of the distortion measurement equipment". I don't have any similar info on how the NE5534 THD measurements were done.

While I don't like to critique other console designs, I have a couple quick observations.  The 12k input resistors and 4.7k  feedback is reducing the worst case closed loop gain several dB which is "improving" the loop gain margin by those same several dB.  So this helps reduce phase shift and distortion in the first stage if/when a lot of stems are assigned. The following stage with 4.7k input and 12k feedback restores the gain to nominal unity and is relatively easy lifting for a decent opamp. There is no free lunch  so there is diminishing return from putting too much gain in the second stage, but for modest amounts this makes sense, especially for old school op amps. The phase shift and distortion added back from the second gain stage will be less than the amount of improvement from running the first stage with less gain. .

OK, that was my guess, that running the first stage at a reduced gain (about -8dB if I'm doing the math right) gave the VE summing a little more breathing room. And then the second stage swaps those resistor values, and it's +8dB back up.

This gain sharing or distribution could make a difference especially if the first stage op amp is marginal.  Note: the 5534 is only stable down to a gain of 10 dB (3x) so if those 12k input resistors are not always assigned a 5534 could be unstable without the extra compensation cap. The 49710 is unity gain stable so that could be useful in that circuit configuration.

The original design was done with all TL071's. The 12k's aren't always assigned, although R141 provides the summing amp input at least a 20k path to ground.

If dropping in a 5534, neither the -8dB nor the +8dB stages would be stable, then...? (Since both are below a gain of 10db). Those compensation caps are the stock values still, 100p for the 1st stage, 33p for the 2nd stage. Along with the respective feedback R values of 4k7 and 12k, that makes the bandwidth for those stages about 300K and 400K, yah? I'm still too much of a dummy to know if that would be sufficient to tame 5534 oscillations. I'd have to just try it and check it with a scope.

 

[joaquins]

Is not the signal gain what you are interested in when analyzing stability, but the noise gain, think at the summing stage as being a non inverting stage and the "noise" input is the non inverting input, what gain would it be taking all the resistors connected to the inverting input from the sources as gain resistors connected to ground, in parallel, compared to the feedback resistor. This will give you the noise gain, which should be greater than 10dB for NE5534 stability, in other case probably a value as low as 22pF is enough to make this opamp unity gain stable, and as you can't make the noise gain less than unity in this configuration you should be good to go. Checking stability with a scope is always a good idea, but you should know what you are looking for, but if there is an oscillation it would be under 10MHz in a non compensated 5534. As has been mentioned before, it's more important a good implementation in PCB routing, with decoupling, reference ground and noise management than the actual opamp used in it, at least till once you are over the specifications of your opamp with your PCB optimization.

JS

 

[leigh]

And just to add to Guru JR's comments, you need the highest standards of Decoupling, Grounding & Layout to get anywhere near the potential of 5534.

I'm not surprised that "uber OPAs" would require such a high level of art to design around, but I thought that the 5534 was a fairly tameable beast, given that a few caveats are heeded.

Again, quoting from the Douglas Self article I brought up earlier in this thread:

The 5532 and 5534 type op-amps require adequate supply decoupling if they are to remain stable, otherwise they appear to be subject to some sort of internal oscillation that degrades linearity without being visible on a normal oscilloscope. The essential requirement is that the positive and negative rails should be decoupled with a 100 nF capacitor between them, at a distance of not more than a few millimeters from the op-amp; normally one such capacitor is fitted per package as close to it as possible.

It is not necessary, and often not desirable, to have two capacitors going to ground; every capacitor between a supply rail and ground carries the risk of injecting rail noise into the ground.

That addresses the decoupling only, obviously – but for an audio-limited bandwidth (and, say, a comfortable frequency margin above 20k), why do you say that "grounding & layout" for the 5534 would be so much trickier than for other opamps?

If you are versed in the art, you will look at the performance you get now from 5534 and compare with what your have calculated. If they are far out, the problem isn't to do with 5534 ... and using an uber OPA will likely make things worse.

Can you actually measure the THD of your Trident to these levels?  And do serious work on stability?

To clarify, I actually have never had a 5534 in this console. Stock, it was originally all TL071's. By the time I got it, it had been through some Jim Williams mods, and the summing amps were already swapped for LT1357's. So I have compared those, the AD843, and the LME49710 as summing amps in the board.

But, to answer your questions, I have been using the FuzzMeasure application to measure THD (and freq response, and phase shift). It sends a swept sine tone through a circuit to do all these measurements. And thus far, THD measurements are around the 0.002% mark - however, that is me sending the test tones in through an input channel, and out the mix outputs. In order to see THD for the summing amp circuit alone, I could inject the test tones directly onto the mix bus, and then take the output from the master insert send (before it hits the balanced output stage).

I have a scope to look for oscillation problems, and a bunch of pico-range caps to add as compensation caps where they are needed (as they were in the Aux output summing amps, for example). Not sure what other "serious work on stability" you might have in mind, but if there are other things I should be checking for, please do tell. I'm learning every day in this process...

 

[joaquins]

Is good practice to decouple the capacitance in the bus with an inductor paralleled with a resistor, or at least the resistor but without the inductor will degrade crosstalk and noise against the inductor paralleled or nothing at all. NE5534 needs more current to work and is able to work with lower impedance loads than TL071, which may bring stability issues against the TL071 for example, that's why stronger stability cares are needed, also the higher unity gain bandwidth may bring some considerations together with higher open loop gain. If you already have 0.002% THD in the complete loop, why do you want to make it even lower, can you hear it? or you are having other troubles with the stage?

JS

 

[JohnRoberts]

The schematic I was talking about is the schematic those distortion measurements were made with. An op amp will have X % THD operated open loop, and when you wrap NF around it that open loop distortion is reduced by the loop gain margin (ratio between open loop and closed gain.) I will ASSume the distortion numbers were made with comparable circuits.

Ah, my misunderstanding. On pg 25 of the LME49710 datasheet they describe how they did the THD measurement on that, using a resistor to extend "the resolution of the distortion measurement equipment". I don't have any similar info on how the NE5534 THD measurements were done.

don't worry about that, just take such puffery with a grain of salt.

While I don't like to critique other console designs, I have a couple quick observations.  The 12k input resistors and 4.7k  feedback is reducing the worst case closed loop gain several dB which is "improving" the loop gain margin by those same several dB.  So this helps reduce phase shift and distortion in the first stage if/when a lot of stems are assigned. The following stage with 4.7k input and 12k feedback restores the gain to nominal unity and is relatively easy lifting for a decent opamp. There is no free lunch  so there is diminishing return from putting too much gain in the second stage, but for modest amounts this makes sense, especially for old school op amps. The phase shift and distortion added back from the second gain stage will be less than the amount of improvement from running the first stage with less gain. .

OK, that was my guess, that running the first stage at a reduced gain (about -8dB if I'm doing the math right) gave the VE summing a little more breathing room. And then the second stage swaps those resistor values, and it's +8dB back up.

It is logical.. but obscure without rigorous inspection.  I like it.  8) Back in my 1980 console article I talked about using distributed summing where sub groups of say n/4 channels were pre-summed, then those sums were combined later in a by 4 summer to keep the noise gain of each sum amp modest.

This gain sharing or distribution could make a difference especially if the first stage op amp is marginal.  Note: the 5534 is only stable down to a gain of 10 dB (3x) so if those 12k input resistors are not always assigned a 5534 could be unstable without the extra compensation cap. The 49710 is unity gain stable so that could be useful in that circuit configuration.

The original design was done with all TL071's. The 12k's aren't always assigned, although R141 provides the summing amp input at least a 20k path to ground.

If dropping in a 5534, neither the -8dB nor the +8dB stages would be stable, then...? (Since both are below a gain of 10db). Those compensation caps are the stock values still, 100p for the 1st stage, 33p for the 2nd stage. Along with the respective feedback R values of 4k7 and 12k, that makes the bandwidth for those stages about 300K and 400K, yah? I'm still too much of a dummy to know if that would be sufficient to tame 5534 oscillations. I'd have to just try it and check it with a scope.

I just checked my calendar and it is 2014... Use the better part...

JR

 

[ricardo]

And just to add to Guru JR's comments, you need the highest standards of Decoupling, Grounding & Layout to get anywhere near the potential of 5534.

I'm not surprised that "uber OPAs" would require such a high level of art to design around, but I thought that the 5534 was a fairly tameable beast, given that a few caveats are heeded.

The 5532 and 5534 type op-amps require adequate supply decoupling if they are to remain stable, otherwise they appear to be subject to some sort of internal oscillation that degrades linearity without being visible on a normal oscilloscope. The essential requirement is that the positive and negative rails should be decoupled with a 100 nF capacitor between them, at a distance of not more than a few millimeters from the op-amp; normally one such capacitor is fitted per package as close to it as possible.

It is not necessary, and often not desirable, to have two capacitors going to ground; every capacitor between a supply rail and ground carries the risk of injecting rail noise into the ground.

Though Self has written some good stuff, this bit is rubbish.

To get BEST performance from 5534 you MUST decouple both rails to an 'earth' with 100u electrolytics as close to the OPA as possible.  His 100n from rail to rail is insufficient for low THD and may give instability under certain conditions.

But if you decouple to an 'earth', you make it Dirty by "injecting rail noise into ground" .. so you can't use it as a reference for your feedback & signal paths.  You need a separate Clean earth.  You need a clear idea of your grounding scheme and its NOT trivial to come up with a good one on a big system like a large mixer.

Have a look at Kingston's http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

Note he also concludes ... when you have done everything properly .. ALL the good OPAs sound the same.

 

[leigh]

Though Self has written some good stuff, this bit is rubbish.

To get BEST performance from 5534 you MUST decouple both rails to an 'earth' with 100u electrolytics as close to the OPA as possible.  His 100n from rail to rail is insufficient for low THD and may give instability under certain conditions.

Under which conditions is too low a decoupling capacitance likely to cause this instability and higher THD? I'd like to test it (and the 49710 as well) to see if I'm running them with enough decoupling.

Also, this may be a semantic issue only, but I thought that 100uF was considered less in the "decoupling" range and more in the "bulk" or "reservoir" cap range.

What isn't a semantic issue is that a 100uF cap is going to have at least a couple ohms ESR, right? And so an extra inch of PCB trace isn't going to matter, proportionally, the way it does with tiny ceramic caps you'd want right up against the power pins of your chip. So, get them close, sure, but it wouldn't have to be the primary layout concern, yes?

But if you decouple to an 'earth', you make it Dirty by "injecting rail noise into ground" .. so you can't use it as a reference for your feedback & signal paths.  You need a separate Clean earth.  You need a clear idea of your grounding scheme and its NOT trivial to come up with a good one on a big system like a large mixer.

Have a look at Kingston's http://www.groupdiy.com/index.php?topic=37307.80

Many true gurus chime in.  It proves how OPA rolling takes a VERY poor second place to correct earthing, layout & decoupling.  It’s a long thread but read the whole thing from #41 to find pearls of wisdom.

Note he also concludes ... when you have done everything properly .. ALL the good OPAs sound the same.

I'm inclined to agree, the more I dig into this, that OPA rolling is 99% a waste of time. Although I can't deny seeing the noise floor drop between a TL071 and more modern chips.

I will check out that thread of Kingston's for sure.

About decoupling power caps to an 'earth', yes, at some point I started to get an idea of how that could be done cleanly. But I also started to realize that I was never going to get that on a retrofit/mod project of my old Trident board. This application note was helpful in thinking about running separate "signal common" and "power common" lines:
http://www.analog.com/static/imported-files/application_notes/AN-202.pdf

 

[ricardo]

Under which conditions is too low a decoupling capacitance likely to cause this instability and higher THD? I'd like to test it (and the 49710 as well) to see if I'm running them with enough decoupling.

Too much distance between the OPA & the decoupling caps to 'earth'.  For some uber OPAs, an extra inch is too much.  The load the OPA drives.  Poor layout.  Poor 'earthing'.

Also, this may be a semantic issue only, but I thought that 100uF was considered less in the "decoupling" range and more in the "bulk" or "reservoir" cap range.

What isn't a semantic issue is that a 100uF cap is going to have at least a couple ohms ESR, right? And so an extra inch of PCB trace isn't going to matter, proportionally, the way it does with tiny ceramic caps you'd want right up against the power pins of your chip. So, get them close, sure, but it wouldn't have to be the primary layout concern, yes?

The ESR of Aluminium Electrolytics is precisely the reason why they are important.  It's not the resistance of the PCB tracks but their INDUCTANCE.  The ESR of Electrolytics is just what's required to damp the rails.

Once you have the Electrolytics, you can add any other Golden Pinnae decoupling scheme you favour.  But DON'T LEAVE OUT THE ELECTROLYTICS TO 'EARTH' NEAR THE OPAS.

Kingston's thread has more details.

Just try local decoupling with Electrolytics to a local Dirty Earth and measure THD before & after.  If you are using Self's flawed method, you should see substantial improvement in THD20k.

Most of the THD improvement is cos w/o the Electrolytics, the OPA is marginally unstable.  Even if it doesn't break into oscillation, operating conditions inside are upset ... so THD is higher

Whether all this is a 'primary layout concern' depends on your expected level of performance.  VERY FEW people can do this well enough to do justice to 5534.

 

[JohnRoberts]

About decoupling power caps to an 'earth', yes, at some point I started to get an idea of how that could be done cleanly. But I also started to realize that I was never going to get that on a retrofit/mod project of my old Trident board. This application note was helpful in thinking about running separate "signal common" and "power common" lines:
http://www.analog.com/static/imported-files/application_notes/AN-202.pdf

I didn't read that complete analog devices note, but it appears comprehensive.

I do have some anecdotal experience to share about downsides from keeping power and signal ground too separate. Back in the late '70s early 80s, I had the misfortune of having a recording console installed in the direct beam of a large AM radio station antenna. The compliance between my separate power and signal ground systems caused a weird interaction where the entire console had a sympathetic oscillation at 960 kHz excited by the AM carrier with a couple volts of signal swing between the two grounds. 

I ultimately stabilized the console with HF shunt caps between the two grounds on every channel, but that was not a fun service call.  8) When looking at grounds I try to visualize the complete path that sundry currents follow. They all must ultimately find their way back to the PS, and even if not passing 1 MHz signals you must anticipate exposure and unintended consequences.   

JR

 

[leigh]

The ESR of Aluminium Electrolytics is precisely the reason why they are important.  It's not the resistance of the PCB tracks but their INDUCTANCE.  The ESR of Electrolytics is just what's required to damp the rails.

Talk to me like I'm a caveman here. When you say "it's not the resistance, it's the inductance", what I hear is: it's not the DC resistance, but the resistance for high frequency AC. Is that indeed an equivalent statement?

And, if you're saying that PCB track inductance is bad, I'm not sure how it is that the ESR of an electrolytic is simultaneously good – since I thought that ESR rises with frequency as well. In other words, wouldn't both an electrolytic's ESR and PCB track resistance/inductance act to "damp the rails"?

 

[leigh]

If you already have 0.002% THD in the complete loop, why do you want to make it even lower, can you hear it? or you are having other troubles with the stage?

Thanks for your thoughts here, JS... I am not focused on getting the THD any lower per se, I just want to know where it is now, and make sure that any other changes I make don't cause it to jump back up. My focus lately has been getting the noise floor lower, which has included trying out different power decoupling schemes for the summing amps - there's a whole big thread about that here: http://groupdiy.com/index.php?topic=57113.0

The summing amp performance, including self-noise, is probably at an acceptable point now for this project, and lately I have been working on improving i/o and mix bus grounding. The summing amps installed are the aforementioned LME49710, which overall seem more stable and less cranky than the NE5534. But I do want to ensure that those "hidden oscillations" (that don't show up on IC output pins, but internally raise the THD) that can plague the NE5534 aren't going to show up as a problem for the LME49710.

 

[ricardo]

Talk to me like I'm a caveman here. When you say "it's not the resistance, it's the inductance", what I hear is: it's not the DC resistance, but the resistance for high frequency AC. Is that indeed an equivalent statement?

No.  They are NOT equivalent.  Inductance & Capacitance are Reactive and what gives rise to oscillation and other evils.  Resistance provides the damping in a resonant circuit.  Insufficient damping and the circuit oscillates.  Its Electronic Eng. 101

If you use ONLY ceramics or polyesters to decouple, you introduce more Reactance and the possibility of oscillation and higher THD cos these have little or no damping.  A good Aluminium Electrolytic however, has serendipitously just the right amount of ESR to damp amplifier & OPA power rails if placed AT the device.

And, if you're saying that PCB track inductance is bad, I'm not sure how it is that the ESR of an electrolytic is simultaneously good – since I thought that ESR rises with frequency as well. In other words, wouldn't both an electrolytic's ESR and PCB track resistance/inductance act to "damp the rails"?

PCB track inductance is ALWAYS bad if you don't have electrolytics AT the OPA.

Yes Electrolytic ESR 'rises' with frequency but it is still resistive.

If you can measure THD below 0.01% at 20kHz, just try it on something that uses Self's flawed decoupling.

Kingston's thread has a lot more practical evidence & recommendations.

 

[JohnRoberts]

Real capacitors have ESR (equivalent series resistance) and ESL (equivalent series inductance). I do not expect the ESR to change while the impedance plot of a cap wrt frequency will first show a falling impedance slope dominated by capacitance, then a minimum impedance dip dominated by ESR, followed by a rising impedance slope dominated by ESL. 

JR

 

[leigh]

No.  They are NOT equivalent.  Inductance & Capacitance are Reactive and what gives rise to oscillation and other evils.

Ok, remembering that I'm a caveman, Reactive-with-a-capital-R to what?

 

[JohnRoberts]

No.  They are NOT equivalent.  Inductance & Capacitance are Reactive and what gives rise to oscillation and other evils.

Ok, remembering that I'm a caveman, Reactive-with-a-capital-R to what?

"Reactive" is the general term for impedances that change with frequency (like inductance or capacitance), as compared to resistive, that does not change impedance with frequency.

JR

 

[leigh]

No.  They are NOT equivalent.  Inductance & Capacitance are Reactive and what gives rise to oscillation and other evils.

Ok, remembering that I'm a caveman, Reactive-with-a-capital-R to what?

"Reactive" is the general term for impedances that change with frequency (like inductance or capacitance), as compared to resistive, that does not change impedance with frequency.

OK, so "Reactive" means changing impedance with regard to frequency. Right.

When Ricardo said "It's not the resistance of the PCB tracks but their INDUCTANCE", I asked him, "so, it's not the DC resistance, but the resistance for high frequency AC?" And he answered, no, that's not equivalent.

In my mind, those are equivalent, because I still think impedance = "resistance for a certain frequency of AC signal". Is that a backwards way of thinking about it?

 

[JohnRoberts]

OK, so "Reactive" means changing impedance with regard to frequency. Right.

yes

When Ricardo said "It's not the resistance of the PCB tracks but their INDUCTANCE", I asked him, "so, it's not the DC resistance, but the resistance for high frequency AC?" And he answered, no, that's not equivalent.

In my mind, those are equivalent, because I still think impedance = "resistance for a certain frequency of AC signal". Is that a backwards way of thinking about it?

Are you a lawyer?

If it helps you to think of it that way go ahead, but inductance or impedance is the more correct terminology.

Resistance implies a simple relationship between voltage and current and power dissipation. Reactance (like inductance) has a more complex relationship between voltage and current and actually stores energy instead of dissipating it as heat like a resistor.

JR

 

[leigh]

Are you a lawyer?

No. As previously mentioned, I am a caveman. I understand simple concepts, and try to understand complicated concepts by stringing together a bunch of simple concepts. I understand resistance intuitively. Impedance, which has more than one variable at a time to track, has always been a bit out of my intuitive grasp.

Resistance implies a simple relationship between voltage and current and power dissipation. Reactance (like inductance) has a more complex relationship between voltage and current and actually stores energy instead of dissipating it as heat like a resistor.

Thank you, that helps a bit with a more intuitive view. Specifically, the storing energy part of it. And that therefore it is, in some sense, time-based. Which has something to do with why Wikipedia says impedance has both magnitude and phase, whereas resistance has only magnitude.

 

[joaquins]

...
"Reactive" is the general term for impedances that change with frequency (like inductance or capacitance), as compared to resistive, that does not change impedance with frequency.

JR

I'd like to comment on this, resistance also changes with frequency, because the skin effect for example and core loses in case of inductors and transformers parasite resistance. Reactive means current and voltage aren't in phase but 90 deg out, no power is dissipated as you explained later. But resistance can change with frequency and it does, maybe the effect is not so big in our case but its there, in high current or high frequency, and it's definitely there for us in core losses.

In my design I'm using 1 pair of electrolytic caps for two DOA between them wich is as close as it could get, 100u. And DOA already have ceramics on them, 2x100n.

JS

 

[JohnRoberts]

...
"Reactive" is the general term for impedances that change with frequency (like inductance or capacitance), as compared to resistive, that does not change impedance with frequency.

JR

I'd like to comment on this, resistance also changes with frequency, because the skin effect for example and core loses in case of inductors and transformers parasite resistance. Reactive means current and voltage aren't in phase but 90 deg out, no power is dissipated as you explained later. But resistance can change with frequency and it does, maybe the effect is not so big in our case but its there, in high current or high frequency, and it's definitely there for us in core losses.

In my design I'm using 1 pair of electrolytic caps for two DOA between them wich is as close as it could get, 100u. And DOA already have ceramics on them, 2x100n.

JS

In  case you did not notice, I was trying to keep it simple and relatively easy to understand, while already not as simple as the OP wanted. 

Skin effect is a phenomenon related to how current flows in and around wire conductors (it's a long story, but the charge of electrons actually repel each other, causing higher current density on surfaces of wires and less current flowing in the center of wires), arguably this can affect the apparent resistance for some specific examples, none of which I've encountered while designing a console. 

It is the nature of the internet for people to share everything they know on a subject even if it does not contribute to understanding in the short term.  Years ago I came up with an idea for an electronics book where every chapter started with very basic rudimentary information, and the deeper you read into each chapter the more complex and detailed the information.  A beginner could read only the beginning of each chapter to get a basic understanding, and revisit earlier chapters that were skipped over later if a deeper knowledge on some subject was needed.  I hypothesized about this book, because I wanted to buy one.  ;D ;D Being self taught I often found myself thrust into the deep end of the pool trying to understand complex topics from advanced text books or papers. With my easy book, I could start reading back where I understood then progress as deep as I needed to go.

JR

PS: I also knew I could not write this book myself. I might be able to write the beginnings of all the chapters, but I would defer to multiple real experts from say a capacitor company to write about advanced capacitor technology, etc.  For my sub-chapter on skin-effect I'd probably get some electric power distribution design engineer to write about the practical effects in high current power distribution, or maybe an old school litz-wire (Litzendraht) guy to discuss the HF phenomenon .