Designing a modern mixing console - Part 1 (and introduction): Channel Input

[Brian Roth]

I also was as dumb as a Box of Rocks "back in the days".....yet, I allowed a VERY simple signal path from the mic preeamp out to the multi-track  "feeder' if desired.

EQ , busing, etc was optional...hard switched, and NOT with CMOS.

http://www.brianroth.com/projects/m77/m77.html

Today, I'D use relays designed for telecom....OR spend a lot of time mucking with FETs.

Bri

 

[JohnRoberts]

Just some more two cents.....in several parts.

I'd have to dig back through 20-ish years-ago notes to be sure the circuit is accurate (OK...I wasn't as organized back then...LOL), but I drew a quick n' dirty sketch of a FET switch node which I used for the monitor switcher I mentioned in previous messages, and I've attached it.  Sorry it's not done in  "PrettyCAD", but I am doing this more as an exercise/share the idea,  than a way to make money.  ;-)

Yup, that is the concept I used to electronically switch a bank of console channels between monitor and mix down modes. In a LOFT console back in the late '70s/early '80s.  This was before Audio Precision even existed, but on SOTA Sound Technology bench test gear, we basically measured opamp residual, just like you did.

That circuit was a way to turn on/off a signal path using a 4053, with the 4053 inside of the negative feedback loop.  R3 keeps the opamp from going nuts when the 4053 is in switch transition.  R2 and R3 are in parallel when the 4053 is in the "audio path is ON", and is unity gain with the values shown.  The CMOS switch section is inside of the negative feedback loop when the 4053 section is "ON".  I didn't have an Audio Precision test set back then (nor now....just an aging Amber 3501 test set), but the numbers I came up with were essentially the same numbers from the 353 opamp by itself.

OTOH, I've also seen the 4053 used in a similar configuration with R2 omitted, and R3 as a 10K.      shrug

Yes, the open loop (r2 omitted), virtual earth inverter/summer is widely published in app notes, and used in many commercial products.

I drew R4 as a method to minimize input DC offsets with the 353 opamp, but to be honest, I think the non-inverting opamp input was merely connected to ground in my switcher gizmo.

Isn't the 353 a JFET input? Bias current should be insignificant, and r4 wouldn't generate any DC offset.  I don't have any documentation on my old work but IIRC I used BIFETs and cap coupled between the TG and opamp - input to manage clicks. 

That circuit has been part of a larger monitor switching system (12 of them.....part of a "stereo-in/six stereo destinations out") for 20 years.  No AUDIBLE noise when switching, and no complaints.

Should be as clean as any practical audio path, so sound as good as that path.

One cool thing I recall playing-with was stacking multiples of the 4053's into the same summing node.  IIRC, it worked well, but I don't recall how good the crosstalk numbers were.  Things like stray capacitance coupling will mess with crosstalk.

Crosstalk inside the TGs is well managed by shunting the "off" input nodes to ground. I suspect PCB layout will be the dominant factor defining crosstalk figures, since that shunted current needs to be kept out of the final opamp + input.

So, Pick Your Poison.....somewhat complexities with the FET switches, or simpler (???) mechanical Panasonic relays as what Amek/Neve used on their 9098 desks!

Best,

Bri

I went with CMOS for cost and PS current concerns. For a one-off design, YMMV.

JR

 

[Nishmaster]

what's the purpose of R12 (250k)?

It's just to establish the 5V reference of the switch input and limit current draw at that node. Since all switches will be off the same 5V rail, I want to make sure they're all isolated.

Doug Self suggests a cutoff frequency of 0.3Hz or lower in a circuit using coupling capacitors. You might also want to search out Stephen Groner's work on reducing distortion in electrolytics by using anti-parallel connection. It's on this forum.

I do remember his work from a few years back on the anti-parallel configuration. That was a cool idea, I had forgotten all about it until you just mentioned it again. Thanks Paul.

I drew a quick n' dirty sketch of a FET switch node which I used for the monitor switcher I mentioned in previous messages, and I've attached it.

Ah yes, I hadn't thought of paralleling two feedback resistors on either side of the switch like that. Very cool!

The OP asked about power supplies...

These desks used many PSU's:

Wow, that is a serious power system there. 1A per module is rather hugely beefy. Then again my desk will be several orders of magnitude less complex, so I hope also several orders of magnitude less thirsty! Time will tell. 1/10th of that per module would be more reasonable here.

EQ , busing, etc was optional...hard switched, and NOT with CMOS.

Yeah, I've been mulling over my options with bus switching and otherwise. At the top of the channel will be the bus input selector switches, 16 of them. Then, there will be the insert in/out switch and pre/post eq switch. At the EQ, there will be an in/out switch. Each send will have an in/out switch for maximum attenuation. At the bottom, the bus assignment switches, 8 stereo buses plus 1 for L/R. Then there will be a solo button (solo in place, not PFL, which I've never really cared for), solo safe, and mute. My initial feeling is to just hard switch the simple bus/eq/aux/insert in/out switches and to hell with clicks/pops there. It's not like you change bus assignments while the mix is going to tape.

Solo and mute, though, I definitely will want to be silent, since those will absolutely be used a ton, so need to be silent running. If I stick with CMOS for critical only, the parts I have selected will be just fine. However, looking across the whole channel, I'll need at least 35-40 of the switch units. At $3.30 a pop, that's just not going to work as either relays or Maxim parts.

So, as always, we'll see what compromise shakes out.

-Matt

 

[boji]

Hello Nish,

Concerning shared power for relays and switch leds, why 5v and not 12v?

 

[Brian Roth]

OK  so 1/2 amp fuses in case something shits the bed,,,,

Bri

 

[ej_whyte]

You could make IC1A an inverting stage and wire the "Input trim" as a rheostat (+end resistor) to vary the gain instead. That means you can move the CMOS-switch to the IC1A input and get rid of IC1B.

That is a possibility, although that limits me to a unity gain configuration at the lowest. It would be nice to have some attenuation. Perhaps the real answer in this case is to move the trim and gain block to after the switch buffer, thus allowing trim of the bus input also. The problem I have with that solution, though, is that it encourages poor gain staging. If the bus is too hot it makes more sense to pull your faders down instead of reducing the bus input trim to the channel.

I thought inverting configurations allowed for negative gain/attenuation?

Thanks JR for the explanation of best practice for electrolytics as well.

Cheers
Elliott

 

[Nishmaster]

Concerning shared power for relays and switch leds, why 5v and not 12v?

It really could be any convenient voltage. I'm going with 5v as most logic stuff is either 3.3v or 5v, which leaves me the option of upgrading the desk to more advanced logic some time in the future should I want to.

OK  so 1/2 amp fuses in case something sh*ts the bed,,,,

Roger that. I like the fuse idea.

I thought inverting configurations allowed for negative gain/attenuation?

Inverting stages are by definition negative gain, i.e. gain with a sign change. What you are referring to are gains less than 1, which are possible in an inverting configuration, but with a 5532, I'm not really too trusting of the stability of the amp.

 

[JohnRoberts]

OK  so 1/2 amp fuses in case something sh*ts the bed,,,,

Roger that. I like the fuse idea.

or flame proof resistors...

I thought inverting configurations allowed for negative gain/attenuation?

Inverting stages are by definition negative gain, i.e. gain with a sign change. What you are referring to are gains less than 1, which are possible in an inverting configuration, but with a 5532, I'm not really too trusting of the stability of the amp.

ne5532* is unity gain stable so will work with any feedback configuration. Unity gain inverting, is actually a noise gain of 2x so "more" stable than unity gain non-inverting.  The only stability concern I would add might be from using unusually high resistance feedback resistors, that could be impacted by pin and PCB stray capacitance to ground. The 5532 is old school so not up to modern performance for input bias and such, but generally a good and stable performer for audio paths.

JR

* the ne5534 is not unity gain stable so requires more consideration for stability.

 

[ej_whyte]

Oh yeh my bad on the negative gain, you knew what I meant though 

 

[Nishmaster]

Revision 4.

With this revision, we should be control click/scratch/pop free. Replacing the full range pot at Gareth's suggestion is a Baxandall style gain trim with about +/-10dB of control. This trim control now affects both sets of inputs, which is much more intuitive from a user interface standpoint. CMOS switch has the shunt legs now correctly added. One nice bonus of the active trim stage is that our inversion from the switch buffer is now eliminated.

Question: Should I include drain resistors after C7 & C8 (the coupling caps of the THAT chips)? I plan for those to be film types to establish my time constant, then the rest can sink into sub-hertz land someplace.

or flame proof resistors...

C'mon, letting a little smoke out never hurt anybody, right?

ne5532* is unity gain stable so will work with any feedback configuration. Unity gain inverting, is actually a noise gain of 2x so "more" stable than unity gain non-inverting.

I know the 5532 is internally compensated, and thus stable at unity gain unlike the 5534. What I wasn't sure of was whether or not that extended to sub-1 gains. Either way, looks like I'll be able to avoid that.

Console block diagram forthcoming.

-Matt

 

[JohnRoberts]

UL likes flame proof resistors in locations that might catch on fire during a common fault, like a shorted IC.

=======

OK here is opamp stability 101... if you understand the basic mechanism that causes opamps to oscillate, the minimum gain advice and compensation caps will make more sense.

The typical opamp has a ton of open loop gain, 100dB or more. This gain basically drives the output positive or negative 100dB times the difference between the + and - inputs. By connecting negative feedback from the output to the - input we can harness this huge gain to make very well defined modest lower "closed loop" gains.  However there is a problem in NF paradise. It takes a tiny finite amount of time for the input signal to make it to the output, and back to the - input. At some very high frequency, this time delay is enough to result in 180' of phase shift, so the nice NF that makes the - input follow the + input is now positive feedback that tells the output to go in the wrong direction.

The solution for this, is to add a simple one-pole integrator roll off in the middle of the opamp forward gain path so at higher frequency the gain is reduced. As long as the gain (and negative feedback) drops below unity by the frequency where the delay turns the NF positive, the opamp will be stable.

A unity gain compensated opamp, starts the roll off low enough that the open loop gain has dropped low enough, that even 100% NF doesn't cause oscillation.  A decompensated opamp, like the 5534, uses a smaller compensation cap, so the open loop gain rolls off less, and relies upon a pad in the NF network to attenuate the open loop gain and feedback to less than unity in time to prevent oscillation.

This smaller cap in the decompensated 5534 is why the 5534 has a slight faster slew rate than the unity gain stable 5532.

OK, now to bring this around to inverting feedback configurations,  the NF from the opamp output still forms a pad with the input resistor (presumably connected to a low impedance), so a unity gain inverter  actually has a noise gain of 2x and is more stable that unity gain non-inverting. 

A gain of less than unity inverting, will look like something between 50% and 100% NF. An inverting gain of -100x still looks like 99% NF.

I hope this helps...

JR

 

[Brian Roth]

If you look between the two Euro I/O connectors towards the left of this this pic:

http://www.brianroth.com/pix/aaron/9098-nov2010/IMG_0117.JPG

...you can see fuses FS1 for the +V audio rail and FS2 for the -V audio rail.  They were physically quite close to the incoming PSU power connector so that (hopefully) any gross failure downstream in the module wouldn't result in smoked PCB foil traces.  Similarly, another fuse (not visible in the pic) was installed for the incoming 5VDC rail. 

In addition, Amek then "branched out" the audio +/- rails into "circuit chunks" of maybe a half dozen or more opamp chips through flameproof 5R6 resistors in series with each rail into the "chunk".  Thus, a failure/short in a "chunk" would "blow" the 5R6 feeding that section, while the rest of the module continued to function.

EDIT:  Once I became familiar with the desks and how they did the module power distro, it became relatively easy to track down failures related to a blown series resistor into a chunk.

The fuses near the power inlet connectors served as a "last resort".  All those precautions were important, since each 9098 module could in theory draw the "full wrath" from the available 12 Amps of the audio PSU for that section of the desk, which would smoke foil inside the module OR the motherboard foils feeding the module.

I'm glad Amek did it that way, since the desk has had failures of the 0.1 uF "mono ceramic" rail bypass caps in various modules.  I think I may have replaced ONE of the fuses at the power inlet within the two 9098 desks I've worked on, but also replaced more than a few of the 5R6 flameproof resistors which fed a rail into a "chunk".

Best,

Bri

 

[Brian Roth]

I realized I have a page extracted from the 9098i desk schemos which shows some of the module power distro (and TWO Amp fuses on the rails).

The older 9098 desk broke down the opamps into smaller "chunks".

Bri

 

[pstamler]

Another way of looking at the 5532 inverting <unity gain issue is that the noise gain of an inverting opamp is the same as the non-inverting gain:

Noise gain = -{(Rfeedback/Rin) +1}

Imagine the worst case: Rfeedback = 0 ohms (a short). Then signal gain will equal 0, but noise gain will equal -1. You can't get the noise gain below 1 no matter what you do. Hence the opamp will remain stable, period.

I'd think again about using the 353, though; I've measured some pretty nasty high frequency distortion from it. Check out Samuel Groner's measurements of the TL072, which has similar performance. Really, there are cleaner FET-input opamps out there, like the OPA2134 (dual version) or OPA134 (single). More expensive, but WAY cleaner.

Peace,
Paul

 

[abbey road d enfer]

Revision 4.
Question: Should I include drain resistors after C7 & C8 (the coupling caps of the THAT chips)?

You don't have to, but I would make provision for a 100k res to ground, just in case.

I plan for those to be film types to establish my time constant, then the rest can sink into sub-hertz land someplace.

I'm not sure I agree with rolling off -2dB at 20Hz. This is the domain of tone control. I would put large lytics there (100-220uF) and deal with subsonics in the filters department.

 

[Brian Roth]

As JR suggested, "fix" the low frequency roll-off at one point, then be SURE that every cap up/downstream uses caps WELL outside of that bandwidth.

Otherwise you are going through a trauma to ensure all the jillions of coupling caps don't conspire to result in a LF roll-off at a much higher frequency.

Bri

 

[abbey road d enfer]

The gain trim uses a tad too many components for my taste.

 

[abbey road d enfer]

It takes a tiny finite amount of time for the input signal to make it to the output, and back to the - input. At some very high frequency, this time delay is enough to result in 180' of phase shift, so the nice NF that makes the - input follow the + input is now positive feedback that tells the output to go in the wrong direction.

JR, I'm sorry to disagree with your wording of this explanation. I know you mean well, but you have to be very careful.
This very wording is what gave birth to the silly TIMD crusade, which stands on: since the output signal is always late, NFB cannot correct it properly.
That's the paradox of Zenon of Elea, that says the arrow can't hit the turtle, because everytime it travels any distance closer to the target, the target moves a fraction of this distance. We know the mathematical answer to that paradox, which is that the series is convergent to a finite value.
It doesn't take a tiny finite amount of time for the input signal to make it to the output, the output signal starts to change at the same time the stimulus is applied (indeed, the time it takes for the electrons to move, some fraction of picosecond, but not really our concern -we're not dealing with GHz).
Due to parasitic capacitances mainly, when the input stimulus reaches a certain slew-rate, the output doesn't follow as rapidly, meaning it is not capable of slewing as fast as linear amplification would permit, but the output starts at the same moment the input is hit. It is possible to represent this phenomenon as a delay, by superimposing the input signal and the appropriately-scaled slew-limited output, but the amount of time is not finite, it increases with the slew-rate of the input signal. For any signal below slew-limiting, the input-to-output is zero for all practical purpose.
Please consider I'm not lecturing you, I know you know this well, but many people who are not as well alert may embark on a goose chase. That's what happened when Matti Otala wrote his paper on TIM, which resulted in bannishment of NFB in audiophool circles.

 

[JohnRoberts]

It takes a tiny finite amount of time for the input signal to make it to the output, and back to the - input. At some very high frequency, this time delay is enough to result in 180' of phase shift, so the nice NF that makes the - input follow the + input is now positive feedback that tells the output to go in the wrong direction.

JR, I'm sorry to disagree with your wording of this explanation. I know you mean well, but you have to be very careful.

I'm trying to make stability compensation more understandable to opamp users. I welcome you comments.

This very wording is what gave birth to the silly TIMD crusade, which stands on: since the output signal is always late, NFB cannot correct it properly.

I don't argue with the TIM crowd.. their lack of fundamental understanding about how things work, make finding common ground too difficult. The quirks of NF were known to engineers designing HF radar electronics during WWII, maybe before.

That's the paradox of Zenon of Elea, that says the arrow can't hit the turtle, because everytime it travels any distance closer to the target, the target moves a fraction of this distance. We know the mathematical answer to that paradox, which is that the series is convergent to a finite value.

I try to avoid calculus in my explanations and life in general. 

It doesn't take a tiny finite amount of time for the input signal to make it to the output, the output signal starts to change at the same time the stimulus is applied (indeed, the time it takes for the electrons to move, some fraction of picosecond, but not really our concern -we're not dealing with GHz).

I don't want to follow you down this rabbit hole, since that won't help the others understand (IMO).

The use of dominant pole compensation (an integrator in the forward gain path) MEANS THAT POLE DOMINATES THE OPEN LOOP RESPONSE.. so yes the output responds instantly, but instantly as in a one pole integrator responding instantly. Any incidental circuit delay is insignificant in this context. I only mentioned delay to explain why dominant pole compensation is used in the first place. Sorry if i wasn't clear.

Now before this dominant pole open loop transfer function stirs up new fears of phase shift and lag in the closed loop response, the huge extra gain margin between an opamp's large open loop gain and modest closed loop gain, reduces this 90' of phase lag proportionately based on the loop gain margin or ratio of open loop to closed loop gain.  With good opamps, running at reasonable closed loop gains (i.e. with lots of loop gain margin) you will be heard pressed to measure significant phase shift within the audio passband. OTOH If you try to run a general purpose opamp up at at mic preamp like closed-loop gain (60dB+), the lack of loop gain margin at HF can show up in the top octave as measurable phase shift.  When you run out of open loop gain entirely due to the falling gain with frequency characteristic, the closed loop response reverts to the simple integrator, but this should only occur way above the audio passband in any respectable design.

Due to parasitic capacitances mainly, when the input stimulus reaches a certain slew-rate, the output doesn't follow as rapidly, meaning it is not capable of slewing as fast as linear amplification would permit, but the output starts at the same moment the input is hit. It is possible to represent this phenomenon as a delay, by superimposing the input signal and the appropriately-scaled slew-limited output, but the amount of time is not finite, it increases with the slew-rate of the input signal. For any signal below slew-limiting, the input-to-output is zero for all practical purpose.

I didn't raise the issue of slew-rate other than to note in passing, that the smaller 5534 compensation cap supports faster output rate of change.  Slew rate is another non-issue raised by the arm waving audio phools, since properly designed systems that LPF their inputs, never ask for faster edge rates than the circuits can deliver. (This wasn't always the case decades ago with crapo opamps).

Please consider I'm not lecturing you, I know you know this well, but many people who are not as well alert may embark on a goose chase. That's what happened when Matti Otala wrote his paper on TIM, which resulted in bannishment of NFB in audiophool circles.

I try not to worry too much about audiophools following false oracles. I used to be in that (hifi) business, and escaped decades ago when I personally experienced the lack of cause and effect between actual performance, and product success.

=======
In the interest of clarity and to stay with my intended goal of making this more accessible. The use of the dominant pole compensation is precisely to trump or swamp out the sundry real delay issues between input and output. As long as the open loop gain (and NF)  is attenuated to well below unity, by that time these non-ideal behaviors become significant, who cares what we call them? 

Again I hope this makes sense to a few out there...  I know Abbey already knows all this stuff... so well he wants to edit me.  8)

JR

 

[Nishmaster]

Before continuing I'd once again like to thank everybody for their contributions. The generosity here is quite humbling.

Continuing on:

OK here is opamp stability 101...

Thanks JR (and Paul). I was aware of the mechanism of opamp stability (or lack thereof, depending). Had I actually thought through a little further, it would be plain to see that even gains of less than one cannot by definition have greater than 100% negative feedback.  :-[ I have all these bits of self-taught EE floating around in my head and occasionally they don't all work together yet.

I'd think again about using the 353, though; I've measured some pretty nasty high frequency distortion from it. Check out Samuel Groner's measurements of the TL072, which has similar performance. Really, there are cleaner FET-input opamps out there, like the OPA2134 (dual version) or OPA134 (single). More expensive, but WAY cleaner.

I assume you're talking about Brian's switcher gizmo here. I don't really plan on using much in the way of TL072/LM353 type era devices (although I do realize the 5532/4 are not far behind in age these days).

In addition, Amek then "branched out" the audio +/- rails into "circuit chunks" of maybe a half dozen or more opamp chips through flameproof 5R6 resistors in series with each rail into the "chunk".  Thus, a failure/short in a "chunk" would "blow" the 5R6 feeding that section, while the rest of the module continued to function.

That's a smart feature. What I need to do is branch my power pins for the opamps out into their own sheets but I haven't figured out how to do that yet in Eagle. Clearly smarter folks than I designed that board, but 5R6 seems high at first glance. Doesn't sag become an issue?

Thanks for the sheet, Brian.

You don't have to, but I would make provision for a 100k res to ground, just in case.

Roger that.

I'm not sure I agree with rolling off -2dB at 20Hz. This is the domain of tone control. I would put large lytics there (100-220uF) and deal with subsonics in the filters department.

I thought -2 there seemed somewhat sensible. Do I really want subsonics eating up headroom throughout? Or does it even matter?

As JR suggested, "fix" the low frequency roll-off at one point, then be SURE that every cap up/downstream uses caps WELL outside of that bandwidth.

I calculated them to be quite low, but it won't hurt to bury them even further. I suppose what I think is low is probably not low enough over the course of a whole console.

The gain trim uses a tad too many components for my taste.

It is a little parts-verbose, true. What I do like about it is that the gain is not tied in any way to the pot value. 20% pot tolerance here still means a 0% variance in total gain range. I also like the lack of high value resistors this soon in the audio path. I will probably sim both to see where the noise performance ends up. I have not been able to find much on the noise contribution of feedback resistors aside from the general advise to use smaller values when possible, up to a point, of course.

As for TIM, I never understood how it could possibly exist in the form often presented. If the output acted in the way described by its proponents, a measurable static phase shift would be present from DC to infinity in addition to the gradual phase shift with slew limit, which we know is not the case.

-Matt