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.

pucho812

Well-known member
Joined
Oct 4, 2004
Messages
14,830
Location
third stone from the sun
thinking about active summing.

At what point would it become unstable.  I read that desks like the focusrite did the following each bucket of X amount of channels would say sum to stereo for that bucket. Then they would sum  the total of the buckets  to the main outputs.
Our desk at the studio does a different approach where the 48 channels just  sum  to together and that was after it was modded from 36 channels and was apparently hard to keep the noise down. Wondering what would be the best way to approve this.

 
The thing is that the more channels you add in a virtual ground style summer with an opamp, the less bandwidth and the more noise gain you have. This was specially important in limited GBP opamps like the 741, specially if you have a gain higher than 1, for example if you sum 24 channels with a gain of 1 , bandwidth is exactly 20KHz in a 741, 32 channels and you go down to 15.2 KHz, this is less important in newer opamps with much higher GBP. 

Also, you may have a signal gain of 1 but the noise gain of a 32 channel summer is 33, so low noise opamps make a difference here.

 
squarewave said:
This is where it's hard to argue with digital which sums perfectly with no noise or distortion.

If there is something that computers do really well, that is adding numbers.
 
pucho812 said:
thinking about active summing.
I wrote about this in my 1980 article about console design almost 40 years ago.
http://www.johnhroberts.com/des_art_1.pdf


At what point would it become unstable.  I read that desks like the focusrite did the following each bucket of X amount of channels would say sum to stereo for that bucket. Then they would sum  the total of the buckets  to the main outputs.
Our desk at the studio does a different approach where the 48 channels just  to together and that was after it was modded from 36 channels and was apparently ahard to keep the noise down. Wondering what would be the best way to approve this.

The noise of one large sum bus, vs several buckets is similar. Perhaps some small benefit from lower coherent noise gain. The noise gain multiplied noise of the different sum buses will combine incoherently, but the larger benefit from using buckets is improved loop gain margin from the lower noise gain at each sum amp. This reduces THD and phase shift (read the article). 

I explored a couple alternatives in my (old) article...  The transamp VE topology enjoys far improved loop gain margin for lower distortion and lower phase shift.

Of course my favorite is my own invention of current source summing...  In theory summing together the outputs of current sources eliminates the N+1 noise gain of typical virtual earth summing. Of course making true bilateral currents sources with low noise and high linearity is not trivial (even for me). What worked for me was using active op amp based synthesized current sources (the familiar 5 resistor topology). I didn't need to target megohms of source impedance just delivering 10x the impedance of typical summing resistors delivers a 20dB benefit all around.   

Of course this cool idea from 40 years ago is mooted by modern digital summing that is lossless.

=====

For your 48 channels a more practical solution is to drop in the Transamp topology in place of one conventional summing amp. Maybe not an actual Transamp that are now getting long in the tooth, but the Cohen topology is pretty well known. Using that topology with modern low noise transistors should spank whatever you doing now for a sum bus with minimal disruption.

JR
 
well this is basically a discussion on the best way to execute the practice.  In our desk, we have hardy 990's as the summing opamps.
It's fine, I do not want to change that stuff out.

Now suppose we were doing a 2 layer approach  as mentioned by sir Winston, would this be applied to the  multi-track busses and aux sends as well?
 
Hi John, thanks for sharing that article.

For the transamp VE example, how is ROL determined? And.. does the transistor introduce an offset voltage at the output of the summing amp?
 
Back in the day, the basic criteria at Neve was for the noise at the output of the bus amp to be no more than -90dBu. The maximum number of channels on one mix bus that would meet this criteria was 64 calculated as follows.

Bus feed resistors were 15K so bus impedance of 64 sources was 15K/64 = 234 ohms i.e. and ideal source for a 10438 transformer. Bus loss is 1/64 = 36dB.

The 1272 bus amp achieved an EIN of -126dBu so with a gain of 36dB to restore the level its output noise would be:

-126 + 36 = -90dBu

In practice most bus numbers achieved much th same numbers. 24 channel sfor instance would have a bus impedance of

15K/24 = 625 ohms

This is too high for the 10468 transformer so it was usually slugged to 300 ohms thereby increasing the bus loss by 6dB from 27.6dB to -33.6dB. The 1272 output noise would then be:

-126 + 33.6 = -92.4dBu

In practice, the noise is often dominated by the noise from the channel amps not the bus amp. If you have 64 channels connected each with output noise around -90dBu, the total noise they contribute to the bus is 18dB higher at -72dBu - obviously unsatisfactory. If you really need to to be able to route that many channel at once to one bus then you really need to pay attention to the output noise of the amp in the channel module that is driving the bus.

Cheers

Ian
 
user 37518 said:
John, do you have a link to all the articles that you published back then?
I put in a link to that console article... (note: I spelled bus wrong back then)

Sorry I do not archive (or much care about) all my past writing. Besides my multiple construction (kit) articles in popular electronics (70s-80s), I actually wrote a column in RE/P magazine (Audio Mythology) back 1980s.  In addition to that I wrote numerous articles while inside Peavey (mid 80s to 90s typically showcasing Peavey SKUs).

JR

PS: I am repeating myself but bus noise is not the biggest problem from large bus confidgurations... its phase shift and distortion.
 
dogears said:
Hi John, thanks for sharing that article.

For the transamp VE example, how is ROL determined?
you can make it anything you want... IIRC inside a transamp it was something like 2.4k and similar value to Rf. For stability evaluation Rf=Rol results in unity gain feedback (output voltage presented to - input).

The power of that common base topology is that open loop gain gets increased by more stems being summed together.  Loop gain margin tracks relatively constant.
And.. does the transistor introduce an offset voltage at the output of the summing amp?
Of course there is diode drop between base and emitter.  My figure 13 is simplified (and figure is spelled wrong too). Typically for a sum bus you would ground the base and cap couple summing resistors into the emitter. Of course there are multiple ways to deal with DC operating points.  Search out a "cohen topology" schematic for details of glue parts. I used that topology in my P-10 MC phono preamp (but I didn't know it was called "Cohen" until years/decades later.

JR

PS: There are probably trans amp app notes floating around the interwebs, while I don't think Paul Buff ever published the innards of the transamp per se,  there was at least one "vallley audio" preamp schematic published that was pretty close.
 
ruffrecords said:
In practice, the noise is often dominated by the noise from the channel amps not the bus amp.
This is theoretically correct and certainly applies to the Neve desks I know, those using summing amps with input transformers. By using heavy audio reference bus bars, it made them quite impervious to longitudinal noise build-up.
All the mixers I know that use VE summing amps suffer from longitudinal noise, that is often  much in excess over the simple N+1 noise.
So actually, reducing the overall summing noise is not a useless endeavour.
Actually, designing a good analog summing for a large number of channels consists in two things;
-Optimizing the structure (differential bus being my fave) for noise
-Optimizing the GBW of the summing amp; transconductance /hybrid being pproven solutions but some recent opamps seem pretty capable in this respect.
What's more, operational bus noise can be greatly improved by choosing to disconnect the feed resistors of the unrouted channels.
Except for the main bus (and the main reverb bus) the actual number of channels routed to a single bus is amazingly low, so the actual "bus" noise is comparable with the channel noise.
So only a few bus need exceptional massage.
Voltage summing does not lend itself to this, though.
 
Winston O'Boogie said:
Yep.  Good stuff. 
The other thing that would help would be to massage the values of the feed resistors to the lowest value the sum amp  can comfortably drive in the desk including everything else that hangs on its output.  Assuming same value of R fb as R feeds here of course.
Since Pucho's desk uses 990's, this can  be a low value R with no problem at all.
Well, even with 10k feed resistors, the source impedance is only a few hundred ohms, usually making input noise current effects negligible, and resulting in thermal noise at the output of the summing amp that's roughly 20dB below the noise of one single channel.
Too low values make some non-linearities happen, particularly in electrolytic caps.
 
abbey road d enfer said:
This is theoretically correct and certainly applies to the Neve desks I know, those using summing amps with input transformers.
even a perfect noiseless mic preamp will boost the self noise of a microphone, these combine incoherently in a perfect sum amp to establish an "ideal" noise floor, BUT unless you are recording inside your anechoic chamber you will have much higher ambient room noise...

I am repeating myself but the difficulty from overly large bus structures is phase shift and distortion caused by inadequate loop gain margin.
By using heavy audio reference bus bars, it made them quite impervious to longitudinal noise build-up.
All the mixers I know that use VE summing amps suffer from longitudinal noise, that is often  much in excess over the simple N+1 noise.
So actually, reducing the overall summing noise is not a useless endeavour.
better is always better...  :-*
Actually, designing a good analog summing for a large number of channels consists in two things;
-Optimizing the structure (differential bus being my fave) for noise
crazy not to use differential for large conventional VE buses.
-Optimizing the GBW of the summing amp; transconductance /hybrid being pproven solutions but some recent opamps seem pretty capable in this respect.
the transamp/cohen topology delivers superior GBW for that application.

For known large conventional VE bus structures you can under compensate the bus sum amp taking advantage of the free stability margin afforded by high noise gain.
What's more, operational bus noise can be greatly improved by choosing to disconnect the feed resistors of the unrouted channels.
1) you need to disconnect both the bus signal resistor and any local ground bus resistor or the differential bus math becomes unbalanced.
2) if using a decompensated or under compensated bus amp, you need to insure that the noise gain does not rise above stability criteria.  You could back ground the deselected bus send resistor to maintain bus balance but would not enjoy any noise gain reduction. Alternately you could back ground the deselected bus send through a capacitor so it looks connected at HF for stability concerns but open circuit for noise build up (you'd want to cap couple local ground bus resistors too). 
Except for the main bus (and the main reverb bus) the actual number of channels routed to a single bus is amazingly low, so the actual "bus" noise is comparable with the channel noise.
So only a few bus need exceptional massage.
my largest console had well over 100 feeds to L/R bus, I am pretty confident that is not a record. In that console I used current source summing for more than half the feeds, current sources and sum resistors can be used seamlessly together.
Voltage summing does not lend itself to this, though.
Let's not ignore console math... You don't want to undertake even simple changes 48 times.  :eek:

I am repeating myself but in my judgement the simplest and most cost/time effective way to upgrade a conventional VE bus with some 48 stems is to drop in a transamp/cohen topology sum amp.  One (or two) and done.... KISS

If you remain invested in staying with 990 DOAs at least tweak the compensation cap to take advantage of the elevated noise gain (I ASSume they still use dominant pole compensation inside). You can use an old op amp trick to stabilize decompensated op amps operated at lower gain using a series R and C from VE - input to ground or op amp + input.  A little slicker if deselected bus send resistors are open circuit is to back ground through a tiny capacitance to maintain HF stability (but again too much per channel detail work for a simple bus amp upgrade.) 

There may be modern uber opamps that are competitive with (maybe better than?) 990 for ein and open loop gain while I haven't seen an IC with inductors in the input LTP yet.  8)

JR
 
JohnRoberts said:
unless you are recording inside your anechoic chamber you will have much higher ambient room noise...
Ambient noise and "bus" noise have a very different colour. Musicians are quite adept at differentiating and blaming electronics whilst accepting the limitations resulting from human beings and equipment being confined in a room.
 
Winston O'Boogie said:
As far as I  could tell, Pucho (and the folks at the studio)  are happy with the desk as is with 990 op-amps.

Seems counter productive to be recommending a change over to some "better" scheme,  I don't think he's interested in re-engineering the whole desk, but I could be wrong?
Maybe he already lost the will to pick up a soldering iron after reading that digital summing is the way to go. 
   
Wait,,, didn't you just suggest that he change resistor values? My suggestion is far less invasive than modifying every channel.

Of course even easier to do nothing if he realizes that it won't change his actual noise floor. 

JR
 
ruffrecords said:
This is too high for the 10468 transformer so it was usually slugged to 300 ohms thereby increasing the bus loss by 6dB from 27.6dB to -33.6dB. The 1272 output noise would then be:

-126 + 33.6 = -92.4dBu

In practice, the noise is often dominated by the noise from the channel amps not the bus amp. If you have 64 channels connected each with output noise around -90dBu, the total noise they contribute to the bus is 18dB higher at -72dBu - obviously unsatisfactory. If you really need to to be able to route that many channel at once to one bus then you really need to pay attention to the output noise of the amp in the channel module that is driving the bus.

Cheers

Ian

This is a great thread.  A treasure of experience and information here.

Ian, forgive my ignorance; when you say the 10468 was "slugged to 300ohm", does that mean a 300r resistor is put before the 10468 or that the 10468 has a 300r resistor that is tied to audio ground? Or something else?
 
Rocinante said:
This is a great thread.  A treasure of experience and information here.

Ian, forgive my ignorance; when you say the 10468 was "slugged to 300ohm", does that mean a 300r resistor is put before the 10468 or that the 10468 has a 300r resistor that is tied to audio ground? Or something else?

I mean a resistor is added across the primary winding hot and cold such that the source impedance seen by the 10468 is 300 ohms. So if the bus impedance was about 600 ohms, as in the example I gave, then you would need to add about 600 ohms in parallel to reduce the source impedance to 300 ohms.

I have just checked the Neve Technical Notes which cover this topic and the 300 ohms figure is probably wrong. The notes are not terribly clear and were not written by a technically qualified person but it looks like the bus impedance could rise to as much as 600 ohms which is good because it improves the noise by 6dB over 300 ohms.

Cheers

Ian
 
Winston O'Boogie said:
I think slugged might be a Brit term.    I've had a few sceptic tanks look at me funny when I said it.  Dunno?

Yours,
Slug.

Not sure where I heard it, has been part of my language for several decades. 
 

Latest posts

Back
Top