Line Input and Channel Gain module for summing Amp

[bjoneson]

me> The INA2134 is full of 25K resistors. Without untangling the topology, I would scratch-mark that as a 25K source resistance.

The rated output hiss of INA is 7 microVolts. Counting on thumbs, that is like a 240K resistor. Makes any diddling of 10K 4k7 1k kinda moot.

There are better diff-amp topologies and implementations.

Side-note-- why the DOA 2520? Fancy opamps have use at very low levels (low hiss) and at very high levels (favorable distortions). Here it seems to have no particular advantage. Fader up, the mix-amp clips first, and INA hiss drowns DOA hiss. Fader down, the INA2134 clips first.

Agreed on the point about the INA hiss making a lot of the subsequent stuff moot. There's stuff with better specs out there, but I've got a pile of these things around.

As far as the 2520... I'm actually likely to run the mix amps on a larger rail +/- 24V, which should allow me to drive the channel amps at the fader a bit. Point taken on bringing the noise  from the INA along with it.

It may be naive, but I'm interested in the potential unique sonic characteristics of DOAs, and the challenge of putting them together. To your point, drawing out unique sonic characteristics requires pushing the envelope of the device in some way.

 

[PRR]

> +/- 18V, and +/- 24V. .... additional headroom on the bus.

2.5dB does not begin to account for 6dB-18dB summing build-up in large mixers and mixes.

> not sure what you mean by "running the faders at -10dB".

Maybe the art of mixing has changed.

 

[bjoneson]

> +/- 18V, and +/- 24V. .... additional headroom on the bus.

2.5dB does not begin to account for 6dB-18dB summing build-up in large mixers and mixes.

> not sure what you mean by "running the faders at -10dB".

Maybe the art of mixing has changed.

My plan is to run 24 channels. I might just be misunderstanding you.

As far as running the faders at -10. I thought you were talking about introducing some kind of fixed attenuation into the circuit itself.

In a large mix with 24 channels running perhaps with similar signals,l. I agree now, it begs the question, why provide 10dB of signal gain on the channel? If you run the channels nominally at -10. You have 10dB of room to unity.

I just needed to digest for a bit before it clicked.

It seems like much of what I've seen to handle the summing buildup is variable gain trim on the summing amps.

I like that, because it seems to provide a little more flexibility.  The 10dB of channel gain is there if needed. That said, the tradeoff is that you always get the 10dB of noise gain to go with it.

 

[JohnRoberts]

> +/- 18V, and +/- 24V. .... additional headroom on the bus.

2.5dB does not begin to account for 6dB-18dB summing build-up in large mixers and mixes.

> not sure what you mean by "running the faders at -10dB".

Maybe the art of mixing has changed.

My plan is to run 24 channels. I might just be misunderstanding you.

As far as running the faders at -10. I thought you were talking about introducing some kind of fixed attenuation into the circuit itself.

While I can't speak to what PRR means, he is probably expecting a reasonable build up of level when you combine multiple stems.  ASSuming incoherent sources 24 stems all at equal loudness (which isn't a normal mix) will combine to roughly 5x or +14 dB.  I have talked about this before but I incorporated a -10dB gain trim into a master sum amp on only one console but mainly because that console had over 100 stems feeding the L-R bus.  In practice this is not much of a problem and why they put meters on the master bus for the occasional hot outlier mix.

In a large mix with 24 channels running perhaps with similar signals,l. I agree now, it begs the question, why provide 10dB of signal gain on the channel? If you run the channels nominally at -10. You have 10dB of room to unity.

Because the customer is always right and the extra 10 dB at the channel fader makes it easier to mix. That said to use that  extra +10 dB without saturating channel electronics requires, running the inputs 10dB cooler, and not using much EQ boost, while this is all manageable with multi-point peak detection in the channels. Some console/mixer makers cheat (IMO) and don't detect for clipping at the channel fader. One well known manufacturer who shall remain nameless, used loss in their master sum stage and no overload detection at the channel fader gain stage while successfully advertising high headroom with many true believers. I would have fun in dealer seminars demonstrating how bad I could make that competing mixer sound with no visible overload indication.  8) 

I just needed to digest for a bit before it clicked.

It seems like much of what I've seen to handle the summing buildup is variable gain trim on the summing amps.

I like that, because it seems to provide a little more flexibility.  The 10dB of channel gain is there if needed. That said, the tradeoff is that you always get the 10dB of noise gain to go with it.

Another tidbit I have talked about, back when I was still designing mixers (mostly for live sound reinforcement) I designed one series optimized for high headroom.  The classic problem I addressed was avoiding that channel fader clipping from fader creep as the gig goes on. In a recording mix it is no big deal to drop down all the other faders when one channel can't go loud enough, but this is not much of an option for say a musician mixing the band from on stage.

My strategy was to continue labeling the channel faders with +10 dB at full up, while in actuality the send to the master bus was unity at full up. -10dB actual at nominal 0 dB channel fader. To keep the overall gain structure as expected, I restored the missing 10 dB of gain in the post master fader gain stage (so +20dB instead of typical +10 dB post sum bus fader gain).  As I repeat all too often the bus noise floor for modern electronics is quieter than we need and in this series of mixers I delivered an extra 10 dB of headroom in the inputs with no apparent noise penalty.  Note I limited this series to 16 inputs but the bus noise for 24 inputs is only something like 3 dB more than 16 inputs. 

I am not suggesting that you use my esoteric high headroom gain structure or a master bus trim for only 24 stems.

KISS. 

JR

 

[bjoneson]

Great insight, JR. Makes perfect sense. This has all given me a bit of pause to assess how this will fit into my overall rig, and what makes sense for gain structure.

Thanks again!

 

[Andy Peters]

A couple things going on here. I actually was planning to run the summing amps on +/- 24V rails. The supply I'm currently using provides both +/- 18V, and +/- 24V. Most of the monolithic stuff caps at 18V rails. I'm considering 990s for summing duty on 24V rails to allow for additional headroom on the bus.

In addition to PRR's comment about the ~2 dB headroom gain from higher bus amp rails not being nearly enough to manage the ~14 dB need from the summing, remember that you'll plug your mix amp's output into something, and will that something have +/-24V rails? Likely not.

You provided a pretty thorough explanation there, but I'm not sure what you mean by "running the faders at -10dB".

Well, you can prevent mix bus overload by mixing the individual inputs in at a lower level, literally set the faders at -10 dB instead of 0 dB.

-a

 

[joaquins]

I usually do something like that, when working mixing live performance with cheap mixers I tend to use master fader almost all te way up and use faders and input gains as low as I can so I never run out of headroom, noise isn't a problem most of the time, the sources already have plenty of noise and the back noise from the audience is sometimes even higher. I've got a few bad sounding using the way they should with faders around 0dB and input gains to get a good level and I never had problems mixing this other way.

  If you are designing the mixer where you will mix you will know it's limitations and know where to add the level you need and where to keep it low, it's only a decision of confort how to label the faders. For the gain structure looks not to have any extra gain in the fader amp makes sense, it won't improve any noise performance but it will make headroom lower. It depends in how you use it, if you want to use it with a lot of spare level over so you can boost there in a final stage it's fine. For the mixer I'm building I'm considering controlling the levels with the gain of an opamp, which will be the fader buffer at the same time, so I never have spare gain there, not even at the lower levels. Same thing for the summing amp, controlling it's gain with the feedback resistor as the fader, it might not give the best fader kill but if you are trying to do so you can always use the mute switch or not to feed signal in there at all. You have to use a unity gain stable opamps or you will have troubles to get to the lower levels which are pretty much unity noise gain and a feedback cap won't be much help since impedance is really low, if you use a cap big enough then at higher levels you will be missing high freq. Also you need to drive this low feedback resistance with your amp, but when you are at lower levels low output signal is expected since the input is limited by the headroom of the stage before which is probably the same than this one, then the output current needs to be the same if using an inverting stage as I am, and you can play around with both feedback resistors to get there.

JS

 

[abbey road d enfer]

The quest for higher headroom must be considered in view of all the other parameters.
The most examplary case for me was when Soundcraft introduced the 500/600 series, where the operating level in the channel was dropped down from the usual -2 dBu to -10.
Nobody praised the extra headroom but everybody complained about the increased noise. In addition, that made the inserts run at too low a level for some outboard equipment.
Headroom is very easy to understand; if it clips, that's because the level is too high. everybody understands that in their guts; no need for arguing.
Bus idle noise is different, it's something that's there even if no signal is routed (in fact it's more audible then), and there is nothing the operator can do about it, except changing the gain structure of the subsequent equipment, which they would be very reluctant to do, understandably. Although you can perhaps prove that the overall S/N ratio of the full mix is as good as it was in the previous generation of the same brand's mixers, the client's opinion is that the product is less good.
This is because nobody took advantage of the increased headroom, to put the master fader down, and there's good reason for it: the master fader should be using the most of the stroke for better control. A bus gain trim would be a useful feature there, but it seems only broadcasters find it convenient.
In the end I had all the 500/600 delivered in France modded and nobody ever complained about noise.

 

[JohnRoberts]

@ Abbey  Back in the '80s while I was writing  my column for a recording magazine I compared the general performance of +4dBu to -10dBV  Mixers. Back in the day the -10dBV gear was lower cost and engineered for use with the small format tape machines also popular back then.  Often the -10dBV gear operated off single rail power supplies internally to reduce cost, with cheaper/slower electronics.

Even with 1/2 the rail voltage the -10dBV mixers were very competitive for headroom. An unexpected benefit from the lower rail was the reduced slew rate requirement.  Even by the '80s the modern electronics were getting quieter.  So for their intended use (supporting -10dBV recorders), the mixers were serviceable and not tragically flawed. 

My conclusion back then still valid, it's harder to manage signal integrity between chassis in a -10dBV environment, and bus noise is more apparent, but still dominated by mic preamps.  Since then op amps have only gotten quieter.

Note: Bus noise is something that customers pay a lot of attention to, often with my disliked "all  inputs muted, output WFO" listening test.  : : Just say no...

JR

 

[abbey road d enfer]

@ Abbey  Back in the '80s while I was writing  my column for a recording magazine I compared the general performance of +4dBu to -10dBV  Mixers. Back in the day the -10dBV gear was lower cost and engineered for use with the small format tape machines also popular back then.  Often the -10dBV gear operated off single rail power supplies internally to reduce cost, with cheaper/slower electronics.

Even with 1/2 the rail voltage the -10dBV mixers were very competitive for headroom. An unexpected benefit from the lower rail was the reduced slew rate requirement.  Even by the '80s the modern electronics were getting quieter.  So for their intended use (supporting -10dBV recorders), the mixers were serviceable and not tragically flawed. 

Indeed, when the equipment the mixer drives is set for -10, operating the mixer internally makes sense in terms of level diagram and headroom, but I think it really challenges the noise performance of a 4558-based summing amp.

My conclusion back then still valid, it's harder to manage signal integrity between chassis in a -10dBV environment, and bus noise is more apparent, but still dominated by mic preamps.  Since then op amps have only gotten quieter.

My conclusions are different. 
In a typical mix, most channels will be set with less that 45 dB gain - a half-decent singer will generate about -40dBu with a close-mouthed SM58, resulting in about 85 dB S/N. There is very seldom more than one or two mics with this gain in a mix. Most of the other dominants sources, such as kick and snare operate with a much lower gain, so their contribution to the overall noise is less significant. So I wouldn't say that the mic tramps dominate, just that they are easier to make close to the theoretical minimum (one has to work very hard to make the EIN higher than -124, and have good reasons for it  :-\ )
In most cases, the optimization of the pre-fader channel level is almost instinctive (with the help of PFL), which cannot be said for the channel faders and master fader settings, so the actual bus noise may vary significantly - of course, the more technically proficient the user is, the better the result.

I've had the opportunity to measure a number of mixers and in most cases, bus noise is around -80dBu.
Manufacturers seem to agree that it is not worth trying bettering this figure.
Only in large mixers with 48+ stems do they employ VLN summing amps (generally BJT/opamp hybrids) or a bucket structure (each bucket of 8 channels has its own summing amps that are sent to a master mixer) but in the end the same overall figure is achieved. Indeed much more remarkable in a 96 channel mixer than a small 16 channel.

This is generally fine, because it is almost impossible and impractical to try and use the whole dynamic range of the resulting media.

Note: Bus noise is something that customers pay a lot of attention to, often with my disliked "all  inputs muted, output WFO"  listening test.  : : Just say no...

JR

There are a number of things that are not strictly necessary for the correct operation of equipment, however they are part of their perceived value, just think how much money the german car makers spend on making their doors sound right.

 

[JohnRoberts]

There are a number of things that are not strictly necessary for the correct operation of equipment, however they are part of their perceived value, just think how much money the german car makers spend on making their doors sound right.

Amen...  For the record I do not advocate a 4558 summing amp while operating at -10dBV makes the circa 4558 slew rate less problematic.

Sorry about the veer... The customer is always right. yadda yadda...

JR

PS: I never made a -10dBV mixer even when making and selling -10dBV cassette recorders.. I like -2 dBu for internal operating levels.

 

[bjoneson]

Abbey / JR - Just a treasure trove of info in that dialog.

My lowly line mixer / summing amp project really pales in comparison to the complexity of a full console, but I love hearing the insight and wisdom of folks who have been there / done that at the highest level.

I'm far from an expert on any of this, and have no formal training. There's a number of great textbooks and I've read many, but the human insight factor, insight from experience is impossible to replace.

I really appreciate your willingness to share those insights, and the patience with which you do. I've probably asked more than one rediculous question, buy at no point have I felt "looked down on" for doing so.

Relatively soon I'll get a more "system level" schematic put together and would love any advice or feedback you're willing to offer.

Many thanks!

-Bob

 

[bjoneson]

Alright... I've gone back over the schematic and taken a lot of the feedback into account.

First some notes on the purpose of this device, and the gain staging topology. This, at the end of the day is a glorified summing amp. It's started to look more like a line mixer, but it's primary purpose is to take 24 outputs from my DA converters and sum them to stereo. Generally speaking all "mixing" is still done in the box.

My current DA converters offer a fairly paltry +16dB max output (@ 0dBFS). Given, that... I have absolutely no danger of clipping the INA134 input stage on 18V rails.  My intent is to set up the "nominal" level of the digital domain such that it matches +4dB at the DA stage. Digital "nominal" with these converters would be -12dBFS.

So, knowing I have +16dB for a maximum input signal. This gives me about 8dB to clipping on the 18V rails I've got. I've set the post fader channel amps at +8dB based on this. Hypothetically there's no way to clip the channel electronics in this configuration. The truth is, there is no "need" for 8dB of gain on the channels, given I'm still "mixing" in the box, the channel amps could actually be bypassed completely (provided the INA's were capable of driving the buses). In fact, this is how API has "The Box" set up. They have a 0dB fader bypass button on each channel when you want to so basic analog summing.

All that said, the nominal level of digitally recorded track can vary greatly depending on who has recorder them and where. At the end if the day, the gain control allows me to "calibrate" the nominal level of each track from down to -20dBFS (with 8 dB of channel gain) -18dBFS is a relatively standard nominal level, so having at least 6dB of channel gain is needed to reach the "proper" analog nominal of +4dBu (again, with my converters).

All this allows me to optimize the DAW faders as they relate to the analog mix bus.

I feel pretty good about the basic topology, and made some revisions in component values to try to optimize things a bit...

- Added 4.7uF cap post fader to keep bias DC current out of pot.
- Reduced pot to 1K impedance (INA134 is more than capable of driving)
- Reduced feeback impedance on 2520 and adjusted values for 8dB of gain
- All coupling caps are sized such that frequency rolloff and phase shift are virtually nonexistent down to 10Hz

I've got a couple of questions....

- Grounding scheme / PCB layout. I'm planning on using a ground plane. Is it worth keeping the supply bypass connections separate (either on their own traces or a separate plane)?
- EMC filtering the input. I've seen some folks suggest a low pass filter on the inputs. I'm worried about monkeying around with the INA's as they're laser trimmed for matched impedance and common mode rejection. Wouldn't most RF show up as common mode signal anyway? I feel like perhaps a low pass filter in front of the INAs would do more harm than good in this case.

As always, any other thoughts or suggestions welcomed / appreciated.

Many thanks,

Bob

 

[joaquins]

It's always good practice to have some filtering at the input, I like to set it about 150kHz with a single pole just to feel right, depending on the environment this may not be needed or not be enough, if you are just under an AM tower probably isn't enough, if your room is a faraday shield isn't needed, any point in the middle needs further investigation. What does the INA data sheet said about it's front end?

Sometimes just a small cap is used between the signals to reject any normal mode HF, and maybe smaller caps between signals and ground to control the common mode signal at some degree. The first one won't affect the CMRR of the INA at all since only loads the normal mode signal. The problem is that CMRR for very high frequencies isn't that great and if you have some trouble you have to deal it somewhere, better sooner than later since the first stage that can't handle it properly (doesn't have enough speed for example) will decode it and for audio circuits with radio signals this happens pretty easily and the first stage is probably the one affected. I guess you know how easy is to pick RF, I'm sure you already heard loud and clear radio stations on a cheap guitar amp, the thing is if you will get interference in your wiring or not. I'd probably add some filtering at the input just to be sure, you could use trim caps to improve HF CMRR if you like, but JR will hit you if you add a trim cap. I think Carlec mixer did use that, better HF CMRR, worse time setting it up properly.

JS

 

[abbey road d enfer]

Alright... I've gone back over the schematic and taken a lot of the feedback into account.

First some notes on the purpose of this device, and the gain staging topology. This, at the end of the day is a glorified summing amp. It's started to look more like a line mixer, but it's primary purpose is to take 24 outputs from my DA converters and sum them to stereo. Generally speaking all "mixing" is still done in the box.

My current DA converters offer a fairly paltry +16dB max output (@ 0dBFS). Given, that... I have absolutely no danger of clipping the INA134 input stage on 18V rails.  My intent is to set up the "nominal" level of the digital domain such that it matches +4dB at the DA stage. Digital "nominal" with these converters would be -12dBFS.

So, knowing I have +16dB for a maximum input signal. This gives me about 8dB to clipping on the 18V rails I've got. I've set the post fader channel amps at +8dB based on this. Hypothetically there's no way to clip the channel electronics in this configuration. The truth is, there is no "need" for 8dB of gain on the channels, given I'm still "mixing" in the box, the channel amps could actually be bypassed completely (provided the INA's were capable of driving the buses). In fact, this is how API has "The Box" set up. They have a 0dB fader bypass button on each channel when you want to so basic analog summing.

All that said, the nominal level of digitally recorded track can vary greatly depending on who has recorder them and where. At the end if the day, the gain control allows me to "calibrate" the nominal level of each track from down to -20dBFS (with 8 dB of channel gain) -18dBFS is a relatively standard nominal level, so having at least 6dB of channel gain is needed to reach the "proper" analog nominal of +4dBu (again, with my converters).

You're right at the heart of compromise (the dirty name for optimization). The ideal solution would be an active gain trim, but is it worth it? I don't think so, unless if the advantage of putting the faders in optimum position has enough value for you.

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

I'm not so sure about this. You would be running 15mA current that may cause problems far worse than Johnson. It is not unmanageable but requires a solid expertise in PCB layout and decoupling.

- Grounding scheme / PCB layout. I'm planning on using a ground plane. Is it worth keeping the supply bypass connections separate (either on their own traces or a separate plane)? 

I think that for audio design hierarchical grounding (and decoupling) is a better approach. The essence is "ground follows signal" and its corollary is "decoupling follows load". The basic principle is making sure that the return currents are not cross-circulating with other ground paths. This is in utter contradiction with the use of a "dead dog" ground plane. A good copper pour, whose contour follows the signal path, is best. The idea is "make the ground trace as large as can be whilst still following signal". 

- EMC filtering the input. I've seen some folks suggest a low pass filter on the inputs. I'm worried about monkeying around with the INA's as they're laser trimmed for matched impedance and common mode rejection. Wouldn't most RF show up as common mode signal anyway? I feel like perhaps a low pass filter in front of the INAs would do more harm than good in this case.

You could use common-mode LC filters. They are commonly available but expensive.
But I think unless you're living at the foot of a transmitter, a 1nF NPOceramic cap connected to the chassis on each external connection should do the trick.

 

[joaquins]

The most convenient way is to have the decoupling caps of each stage going to the next stage reference? Once every signal is routed with it's ground then you have to pick an arbitrary place to tie them all together? Is that right?

I can see why this is a good thing, but I still have one doubt, the current introduced in the ground by decoupling caps will put some voltage at this point, this voltage fluctuation will be only at the ground and maybe at the rails, but not reflected to the output signal since most opamps have good PSRR. This gets compensated since the current at the ground should be equal from decoupling caps and load, but class AB amps will have spikes at the rails which won't appear at the load. Other way to compensate for it is to make the input reference following that, so you are now referenced to the same point the input than the output and the spikes will apear at the ground and the signal. Here is an scheme, the 3 resistors at the bottom rG# are resistance on the ground trace, just for representing what we are talking about, it's an inverting stage but I guess same idea would apply to any other case. I'm getting this right or I'm missing something.

Thanks
JS

 

[abbey road d enfer]

The most convenient way is to have the decoupling caps of each stage going to the next stage reference?

That's the idea.

Once every signal is routed with it's ground then you have to pick an arbitrary place to tie them all together? Is that right?

No. That is a very common mistake, based on a wrong understanding of the star-ground concept. The "grounds" must be joined in hierarchical order with the shortest and strongest connection.
The star-ground concept makes sense only when the system has many inputs and/or ouputs, like in a mixing console, where the hierarchical concept can be applied to each channel or group, but a different system must be used for the longitudinal arrangement.

I can see why this is a good thing, but I still have one doubt, the current introduced in the ground by decoupling caps will put some voltage at this point, this voltage fluctuation will be only at the ground and maybe at the rails, but not reflected to the output signal since most opamps have good PSRR. This gets compensated since the current at the ground should be equal from decoupling caps and load, but class AB amps will have spikes at the rails which won't appear at the load.

These spikes exist on the supply rails and on the decoupling caps; but when the decoupling caps are joined at the same point as the load reference, Kirchoff is our friend, these currents sum to zero. This is an approximation because the rails absorb the bulk of the load, but the transients, which are what creates pollution on the ground, should be absorbed by this careful and wise decoupling scheme.

Other way to compensate for it is to make the input reference following that, so you are now referenced to the same point the input than the output and the spikes will apear at the ground and the signal.

Indeed, we could use the differential properties of a the next stage. Any stage is in essence differential at the input, it's the voltage between base and emitter, grid and cathode, etc that is taken into account, so basically, running the reference point of the subsequent stage to the output reference of the preceding stage cancels out these interferences.

  Here is an scheme, the 3 resistors at the bottom rG# are resistance on the ground trace, just for representing what we are talking about, it's an inverting stage but I guess same idea would apply to any other case. I'm getting this right or I'm missing something.

Yes, that's exactly what I mean. In fact, on your example, everything seems to be more or less contained, because the decoupling caps are tied together and the ground tracks appear to be direct and short. But in the case of a dead-dog ground the return path may be quite convoluted.

 

[JohnRoberts]

Indeed, we could use the differential properties of a the next stage. Any stage is in essence differential at the input, it's the voltage between base and emitter, grid and cathode, etc that is taken into account, so basically, running the reference point of the subsequent stage to the output reference of the preceding stage cancels out these interferences.

yup...

JR

 

[joaquins]

Thanks for the response Abbey, this makes a more comprehensive approach on the topic, I was using separated ground path for clean reference and decoupling.

I have one more doubt on the subject, the caps needs to be big enough to consider LF response to the outputs, since technically you are driving the load from the caps, if that's the case we should use 100µF caps at least when working on 600Ω loads maybe more. Bad news we will always be cap coupled since this is quite hard to avoid...

These spikes exist on the supply rails and on the decoupling caps; but when the decoupling caps are joined at the same point as the load reference, Kirchoff is our friend, these currents sum to zero. This is an approximation because the rails absorb the bulk of the load, but the transients, which are what creates pollution on the ground, should be absorbed by this careful and wise decoupling scheme.

But the current should be from the rail, cap, ground, load. Not rail to rail, and the spikes in AB mode happens at different times, since are at different points of the waveform. 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

 

[JohnRoberts]

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