summing large amounts of channels.

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
Back in the days when component manufacturers were trying to sell them to anyone with money, research labs at RCA/Mullard/Texas instruments//Toshiba? etc would get company engineers to come up with 'projects' to showcase whatever was the 'in thing'. Thus Mullard produced a 'cookbook' of exciting things to make with a couple of (new fangled) silicon transistors, although had done similar with germanium transistors previously. Texas with TIS 92, and the even more exciting TIP range of complimentary power transistors 29 to 35 produced a handy booklet with power amplifier designs from 0.5 Watts to 100 watts with hints on how not to blow them up!
Indeed the Williamson valve amplifier was 'hatched' by DTN Williamson to promote the KT55 or was it KY66 (I think) post WW2 to try to boost the sales of valves, having published a high quality amplifier design in 1943 or thereabouts in Wireless World.
I suppose sin offs from the war in the far east brought forward the tropicalisation of transformers etc in an attempt to make equipment reliable under extremes of heat and humidity. There must be some interesting 'snippets' incorporated in the amateur press of 'handy things to make using an ex war Department radar set'! Using the loudspeaker magnetising coil as a power supply choke which could also part cancel the inevitable hum from the output stage was cunning trick.
 
Every patent I have applied for has been defeated by a soaking wet newbie at the USPTO. Once they get it in their head that they think they know, you're done. Never got one.

For summing, I do it the old fashioned way, with a twist, I sum 8 channels locally and then sum the buckets balances, into the masters. I hit the internal busses with anywhere between -2 and 0, then the bucket busses are at +4. Advantages are very low noise, and very low noise immunity. I also use linear pots for panning, with a 1256 under each side in a +6 configuration. Works like a champ....
 
Brian Gibson:
Thanks! However it is rather superficial.
For summing, I do it the old fashioned way, with a twist, I sum 8 channels locally and then sum the buckets balances, into the masters. I hit the internal busses with anywhere between -2 and 0, then the bucket busses are at +4. Advantages are very low noise, and very low noise immunity. I also use linear pots for panning, with a 1256 under each side in a +6 configuration. Works like a champ....
A good design starts with a good level diagram.
 
Every patent I have applied for has been defeated by a soaking wet newbie at the USPTO. Once they get it in their head that they think they know, you're done. Never got one.

For summing, I do it the old fashioned way, with a twist, I sum 8 channels locally and then sum the buckets balances, into the masters. I hit the internal busses with anywhere between -2 and 0, then the bucket busses are at +4. Advantages are very low noise, and very low noise immunity. I also use linear pots for panning, with a 1256 under each side in a +6 configuration. Works like a champ....

IIRC I described that approach in my 1980 console performance limits article as one of the options used. Indeed by summing in buckets of only 8 channels you can mitigate the limitation of loop gain margin that can be problematic when trying to feed tens of stems into a single bus combining amp. My last "big" mixer had over 100 stems feeding the L/R bus, but I've told that story so many times even I am getting bored by it.

===

Yup patent examiners are just doing on the job training to gain patent experience until they can move across the street and hang up their shingle to be a patent lawyer and earn the big bucks. I am convinced a clever patent lawyer could patent just about anything, but that patent would be worth less than you paid for it. Experienced patent lawyers do a dance with the junior examiners. They will even write up superfluous BS extra claims to give the examiners something to reject for a face saving win. I have even had patent lawyers advise against doing a search, they were just collecting expensive wallpaper chits and didn't want to miss out on the fee.

====

As I have said before I prefer circuit analysis to politics because it is, or should be objective truth. As observed here and from my experience with patent examiners who don't know the difference between currents and current sources, there can be subjectivity even in analysis. At least the patent examiners are supposed to question everything.

JR
 
close but not quite there... 4 of the 5 precision resistors could be from the IC but the 5th added resistor could degrade the potential matching. Years ago I asked Gary Hebert (senior THAT corp engineer) about a custom metallization layer. The basic silicon for that family of parts has all the different resistors and they just connect up the ones they need for the given fixed gains. I could grab a 5th resistor from one of those unused Rs and get superior matching but the project stopper was that THAT corp refused to source ICs without full protection wells for every I/O pin. I needed one more protection well than available on the die and completely new silicon just wasn't practical. The cost of just a metallization layer was probably tens of $k.

====

The buffer I was talking about adding if I was to revisit that old production console design was after the pan pots feeding the current sources. To save op amps (remember there were 72 pairs of these current sources just in the monitor section), I drove the synthesized current sources directly from the pan pots so pan position slightly imbalanced the current source output resistance multiplier. There was a just barely audible difference in master bus noise floor when listening WFO. The lowest noise (highest current source output impedance) was with all monitors hard panned left or right.

JR
Gotcha on the upstream buffer.

wouldn this version with the additional-additional buffer gain you the benefits of the matched resistors on the IC as well? Assume the 9ks are on the that1240 (or ina134)

1632371835431.png
 
Gotcha on the upstream buffer.

wouldn this version with the additional-additional buffer gain you the benefits of the matched resistors on the IC as well? Assume the 9ks are on the that1240 (or ina134)

View attachment 84933
For a high compliance bilateral current source the node marked "bus" would be high impedance. From observation any signal on "bus" will be precisely replicated at output of U1... That topology looks like the figures from the 2020 TI application document.

Hanging a bunch of op amp inputs from the bus is probably workable using modern bifet op amps that have negligible input bias current.

I'm not sure what problem this solves

[edit] it solves the problem of needing a 5th precisions resistor, but the tradeoff is adding another noise source into the audio path. [/edit[

other than using more op amps, an implicit goal of FAE (filed application engineers) writing app notes. The precision differential amp wouldn't suck. I don't think these parts were available back when I was developing this.

JR
 
Last edited:
For what it is worth, Doug Self ran through some advantages of 'devolved summing' for which low noise (hiss) can be one but on my assertion that a 'star earth' is not really practical in a universe that has real physical dimensions, and the true observation that you can't induce current flow into a single point, it all falls down on planet earth especially when summing many sources implies significant physical dimensions. The nearest thing you could get might be a custom made 'small outline chip with maybe 100 resistors in a 1cm square package (or however large is necessary) to be the 'summing point' whereas devolved summing can have a range of other benefits such as not having long, sensitive bus traces wandering about and the possibility (using multiple easily swappable summing 'cards' ) maintenance can be easy.
I may have missed some of the real technical aspects of the previous parts of the discusion but a significant part of my life is in the more 'practical' arena where getting people to put the plugs in correctly features.
 
For a high compliance bilateral current source the node marked "bus" would be high impedance. From observation any signal on "bus" will be precisely replicated at output of U1... That topology looks like the figures from the 2020 TI application document.

Hanging a bunch of op amp inputs from the bus is probably workable using modern bifet op amps that have negligible input bias current.

I'm not sure what problem this solves other than using more op amps, an implicit goal of FAE (filed application engineers) writing app notes. The precision differential amp wouldn't suck. I don't think these parts were available back when I was developing this.

JR
I may have misunderstood. On the example you posted in June you used an improved Howland pump, and splitting the +IN feedback loop between resistors. The overall differential amp was balanced -- R220/R221 = R224 / (R227+R225) . The bus signal is taken from the node between R227 and R225.

The challenge you noted was precision matching the resistors. On an INA124 or THAT1240 you don't have access to the +IN node or a fifth resistor, so you could add two external resistors which puts you back into the question of matching.

I thought the advantage was that by putting a buffer into the positive feedback loop you isolated the effects of the fifth unmatched resistor from the diff amp. But perhaps it would be much easier to use a precision matched pair outside the precision diff amp.
 
I am surely repeating myself from only days ago. I tried to get THAT corp to re-metalize one of their I/O chips to grab all 5 precision resistors, they declined (for reliability reasons).

The resistor matching required to get useful elevated source impedance is not that difficult. The 1% resistors I used can deliver a nominal 100x improvement. In practice 1% resistors from the same production batch will be closer than 1% to each other. Peavey had 0.1% resistors in the system but I didn't feel them worth the expense.

JR
 
Just a general observation here. I tried to read this entire thread and it just got too bewildering. Why all the agonizing over 100 inputs - or even 48? Although I haven't hung out in a studio for 30 years, I just can't imagine that many channels being used simultaneously for mix-down of any kind of music! When I became chief engineer at Quad-Eight (in 1972, replacing Deane Jensen), they were making custom consoles ("boards" or "desks" depending on your locale) with 16 inputs, but in actual use I rarely saw that many used. So I re-designed the virtual ground mixing bus structure so that when a channel was unused (muted), its summing resistor was disconnected from the mix bus. Thus, unused channels don't increase noise gain of the summing amplifier. I thought the idea was rather obvious but customers loved it. I honestly don't know how long it remained in the product line after I left Quad-Eight (the purchase of Quad-Eight by Mitsubishi was the beginning of the end for the company - sadly).

The thread's hand-wringing about outrageous signal levels and currents just makes me smile. Granted, however, that grounding can be an issue - especially at the summing bus. I was only beginning my quest to unravel the mysteries of grounding in my Quad-Eight days, but I knew that some of their "standard" internal grounding practices were inviting problems. For example, the practice of grounding cable shields only at the line receiver end! Often a console would break into ultrasonic oscillation if all the faders were set at max gain - for reasons rather easy to explain. It was easily cured by grounding cable shields only at the line driver end - which seems obvious if you think in terms of current flow and loops.
 
Hi Bill! I'm an Olde Timer in the biz who began with a 16 track Ampex into a 24 channel MCI desk at my first "real" studio gig. Things have escalated over the years that are just insane.

Ignoring all the "modern" ProFools <g> projects that use dozens/hundreds of tracks "in the box", some producers decide to mix "out of the box" from 10000000 tracks <g> recorded on the computer.

I recall a project that I saw as a "fly on the wall" in a studio with 48 analog channels coming from the ProFools system. Now, some tracks were only a couple of tambourine hits on each chorus, etc.

This particular producer used six or eight desk channels just for the lead vocal which were multed from the (composited) vocal tracks. One desk pot was the dry vocal. A second was return from a Dimension-D effects box. Third pot was return from a WW2 vintage "FAA" limiter. Fourth pot was return from a LA-2. Fifth pot was return from ....hell, I don't remember!!! lol

All those returns were in a VCA/moving fader group so one pot on the desk would move that entire lead vocal concoction up and down.

Hence, the lead vocal required six (whatever) feeds into the stereo sum bus. The entire song took up 64 (??? it's been awhile) pots on the desk, with many involved with those "parallel processing" tricks.

I thought it was crazy, but that's how some people operate.

Bri
 
Just a general observation here. I tried to read this entire thread and it just got too bewildering. Why all the agonizing over 100 inputs - or even 48? Although I haven't hung out in a studio for 30 years, I just can't imagine that many channels being used simultaneously for mix-down of any kind of music! When I became chief engineer at Quad-Eight (in 1972, replacing Deane Jensen), they were making custom consoles ("boards" or "desks" depending on your locale) with 16 inputs, but in actual use I rarely saw that many used.
As Brian shared back then there were many virtual channels coming from synthesizers, sometimes tens of extra line level sound sources coming from expansion modules all sync'd together using midi.

For the record I rounded down to 100 stems... My largest configuration a split console with 36 inputs and 24 sub masters. Had two tape returns in each monitor strip addressed to the L/R bus, plus a third expansion L/R input capability accessed via the sub master fader insert jack. Then there were another 8 L/R feeds coming from the Aux/Efx send-return section. I don't recall any customers ever saying that they ran out of inputs, but some filled that puppy up pretty close.

So I re-designed the virtual ground mixing bus structure so that when a channel was unused (muted), its summing resistor was disconnected from the mix bus. Thus, unused channels don't increase noise gain of the summing amplifier.
This is pretty much how every sub master channel assignment switch works or worked. For L/R buses in physically large consoles the input strip and master section could be several feet apart. To minimize potential differences between the channel 0V and the master 0V one successful approach is to use a differential bus where all the local input strip 0Vs feed a separate 0V bus. Alternately connecting and disconnecting the feed resistor would imbalance the differential math for imperfect cancellation of 0V differences. One strategy to maintain differential balance is to back ground the feed resistor so when deselected it is still connected to a low impedance source. Of course this leads to a worst case noise gain, undesirable in large bus structures. This can be mitigated by deselecting both bus feed resistors (using a DPDT switch), so the differential buses remain mathematically correct. Note: there are plenty more bus combining variants beyond the scope of a simple discussion.
I thought the idea was rather obvious but customers loved it. I honestly don't know how long it remained in the product line after I left Quad-Eight (the purchase of Quad-Eight by Mitsubishi was the beginning of the end for the company - sadly).
Obvious perhaps but with other considerations.
The thread's hand-wringing about outrageous signal levels and currents just makes me smile.
Makes me a little sad as this is all moot thanks to near perfect digital combining.
Granted, however, that grounding can be an issue - especially at the summing bus. I was only beginning my quest to unravel the mysteries of grounding in my Quad-Eight days, but I knew that some of their "standard" internal grounding practices were inviting problems. For example, the practice of grounding cable shields only at the line receiver end! Often a console would break into ultrasonic oscillation if all the faders were set at max gain - for reasons rather easy to explain. It was easily cured by grounding cable shields only at the line driver end - which seems obvious if you think in terms of current flow and loops.
As I have shared multiple times over the years maintaining signal integrity when passing signals around inside a console requires accepting the reality that ground is only 0V, until you have a second ground node. The proliferation of low cost high performance op amps made using differential processing between nodes cost effective.

JR
 
Quote:
So I re-designed the virtual ground mixing bus structure so that when a channel was unused (muted), its summing resistor was disconnected from the mix bus.

That is a worthy mission but of course the fun really starts when you try to prevent clicks when channels are routed/derouted from the bus when faced with the odd millivolt or three because the bus amplifier is at a different DC potential the the channel modules at the 'far end' (or ends if the bus takeoff is in the centre).
There is also perhaps the 'Fourier' click when using a 'normal' switch (cutting a signal when it it is not at zero crossing).
 
Back
Top