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.
Glad Abbey Road wrote all those answers, saves me the effort!
You also have to consider the devices that were available at the time of design AND the budget that was allowed. Stuffing a 5 cents transistor ahead of a TL072 to obtain at least adequate performance compared to the pair of multi transistors on a chip at what was around 10 Dollars a time was acceptable for SSL and the selling price of a 4000. Depending on how many channels the Midas was trying to sum, it is probably at least acceptable. The low base/emitter impedance of low power power transistors when run at a decent current can (I believe) produce quite a good low impedance 'virtual earth' bus when supported by a decent amount of gain to provide negative feedback. As with all life your 'best' may not be the same as anyone else's. Accessing the collector noded of an op amp to then use external transistors with lower base emitter impedance while biassing the internal pair off was used by a few companies for which chip swapping may be less successful now if the originals were NPN and you try a PNP paired 'wonder' amplifier. You then have to consider if the components around the summing stage are laid out suitably for op amps with much higher high frequency capability AND can remain stable when the incoming buses can have significant and possibly varying capacitances to ground. As with everything in life, 'it depends'!
Matt S
 
I will try to avoid repeating myself... these are well inspected topics here for years.
=====
Loop gain margin (the difference between closed loop and open loop gain) is different than stability margin (phase shift vs gain).

JR
 
For a 5534, it's only 10MHz. When you have 50+ stems, it translates as a -3dB point at about 200kHz, which leaves about only 20dB gain margin 20kHz. Gain margin is important as it is the dominant factor in NFB distortion reduction.
How did you calculate the -3 dB point and the remaining gain margin? Or is my datasheet missing something?
How the currents delivered to the various loads return to ground.
Not something that can be discussed in a few words.
Consider that ground is constituted of a superposition of resistive elements, through which currents develop volyages.
I understand. So it also comes down to low impedance paths and trying to avoid bad superimpositions due to spatial arrangement, to put it in a few words.
To be honest I haven't fully understood the grounding design of my XL200 yet. For example I got a 0 V, MIXREF, MWKGROUND and a GNDREF on the bus. Every channel features an op amp that takes the GNDREF on its input and its output is connected to the actual ground of the channel. Does this serve to actively suppress ground noise (schematics attached)?
Depending on how many channels the Midas was trying to sum, it is probably at least acceptable. The low base/emitter impedance of low power power transistors when run at a decent current can (I believe) produce quite a good low impedance 'virtual earth' bus when supported by a decent amount of gain to provide negative feedback.
Thank you Matt! My console is summing 40 channels.

I think I would really like to try a trans amp for summing in my Midas console. I haven't found any concrete information on how to actually design such a summing amp though. Like where do I set the working point for the transistor? NPN or PNP? Which op amp? Feedback resistor values? I guess the most important thing to consider for the transistor is noise?
 

Attachments

  • midas_gnd1.png
    midas_gnd1.png
    1.2 MB · Views: 42
  • midas_gnd2.png
    midas_gnd2.png
    573.2 KB · Views: 36
How did you calculate the -3 dB point and the remaining gain margin? Or is my datasheet missing something?
50 stems means a noise gain of 51. 10MHz divided by 51 is about 200kHz.
To be honest I haven't fully understood the grounding design of my XL200 yet. For example I got a 0 V, MIXREF, MWKGROUND and a GNDREF on the bus. Every channel features an op amp that takes the GNDREF on its input and its output is connected to the actual ground of the channel. Does this serve to actively suppress ground noise (schematics attached)?
Yes, it's quite difficult to understand since it is necessary to see how these different "grounds" are depatched in the channel.
I think I would really like to try a trans amp for summing in my Midas console. I haven't found any concrete information on how to actually design such a summing amp though. Like where do I set the working point for the transistor?
It is set in order to provide the best OSI for the actual bus impedance.
NPN or PNP?
PNP's are usually preferred because they have a slight noise advantage.
Which op amp?
The one that's in the part bin and is decent for the job. ;)
Feedback resistor values?
This should be determined by the actual gain you want to achieve, which is the ratio between teh NFB resistor(s) and teh bus injection resistors.
I guess the most important thing to consider for the transistor is noise?
Correct, although it can be debated that the transition frequency is to be considered.
 
Yes, it's quite difficult to understand since it is necessary to see how these different "grounds" are depatched in the channel.
I could upload the complete schematic of one channel. Would that be allowed or not? I would be really happy if I could finally fully understand the whole concept.
It is set in order to provide the best OSI for the actual bus impedance.
40 Channels with 10k mixing resistors give 250 Ohms bus impedance, if I'm correct. What does OSI stand for?
This should be determined by the actual gain you want to achieve, which is the ratio between teh NFB resistor(s) and teh bus injection resistors.
Yes, so in the original configuration with only the virtual ground summing op amp the NFB resistor is 10k. But why is R3 2k and R4 1M in the Soundcraft 6000 trans amp? IIRC the 6000 has 4k7 mixing resistors.
 
OSI (probably) means "optimum source impedance". When dealing with ein (equivalent input noise) current and voltage noise analysis the input noise current gets multiplied by the source impedance before combining with input noise voltage and Johnson Noise of the resistors.
===
The Transamps when used as a summing amp looks like a grounded base input stage with the signal (current) fed into the emitter. This topology has the unusual characteristic of open loop gain increasing with the number of inputs for a near constant loop gain margin.

I discussed some of this in my old console performance limits article, but some good basic information (IMO).

1980 console article

JR

PS: Be kind, I wrote this article over 4 decades ago and before I learned the correct spelling of "bus".o_O
 
The Transamps when used as a summing amp looks like a grounded base input stage with the signal (current) fed into the emitter. This topology has the unusual characteristic of open loop gain increasing with the number of inputs for a near constant loop gain margin.

I discussed some of this in my old console performance limits article, but some good basic information (IMO).
Thanks John, I've already seen and read through your article earlier in this thread. That was the trigger for me wanting to try the trans amp as summing stage. I still don't get how to set the optimum bias point. Looking at the PNP transistor as a common base amplifier, the input impedance is r'e = 1/gm since r'e is much smaller than the external emitter resistor. gm = ∂Ic/∂Vbe so how should the bias point influence the input impedance?
 
The transamp will behave well with a range of impedances. Since the Transamp is originally designed as a mic preamp it will be happy with modest/low impedances. Since the base is grounded in the summing topology, zero ohm ternmination will make little noise contribution.

One thing I discovered about improved sum bus technology is that when you remove that noise contribution below significance we still have to deal with mic preamp noise, mic self noise, and room noise.

Good luck.

JR
 
Looking at the PNP transistor as a common base amplifier, the input impedance is r'e = 1/gm since r'e is much smaller than the external emitter resistor. gm = ∂Ic/∂Vbe so how should the bias point influence the input impedance?

The input impedance of a CommonBase stage is ca. 26.10e-3 divided by Iq, so for 1mA quiescent, the input Z is 25 ohms.
But the actual impedance of the circuit with NFB is lower, being Rfb/OLG. At 20kHz, where the 5534 has about 65dB OLG, the input Z with a 10k FB resistor would be about 5 ohms.
Please see attachment for the simulation of input impedance with 2K feedback resistor.
But why is R3 2k and R4 1M in the Soundcraft 6000 trans amp? IIRC the 6000 has 4k7 mixing resistors.
Thse values come from an attempt by a bunch of wet-behind-the-ears designers to show their elders they were capable of bettering them. They came for a virtually impossible to clip gain structure, where the channels operated at about -10 dBu and the summing stages had enough attenuation to resist all-channels cranked situations, and ended up with a mixer that left perplexed users wondering why their mixes were noisy.
 

Attachments

  • input Z transamp.jpg
    input Z transamp.jpg
    43.2 KB · Views: 37
Thse values come from an attempt by a bunch of wet-behind-the-ears designers to show their elders they were capable of bettering them. They came for a virtually impossible to clip gain structure, where the channels operated at about -10 dBu and the summing stages had enough attenuation to resist all-channels cranked situations, and ended up with a mixer that left perplexed users wondering why their mixes were noisy.

Haha :D

I think I would really like to try a trans amp for summing in my Midas console. I haven't found any concrete information on how to actually design such a summing amp though. Like where do I set the working point for the transistor? NPN or PNP? Which op amp? Feedback resistor values? I guess the most important thing to consider for the transistor is noise?

There are always exceptions but, generally speaking, a PNP will have lower noise than its NPN counterpart in this application. I think good current gain is still important here so options are more limited now regarding what's still available but, the old Toshiba 2SA1316BL would be a good one which you can still find.

I don't see anything wrong with using a 5532 as the op-amp.

You say your bus feed resistors are 10K so 10K fb R would be the logical choice. If your channel amplifiers can comfortably drive lower, and you're up for changing them out, going down to 6K8 or 4K7 on the critical L&R bus feed R's (and also on the hybrid bus amp feedback R) would yield some improvement.

Edit: In suggesting you could change the bus feed R's, I'm assuming that they (bus feed R's) are fed from buffer amps.
 
Last edited:
There are always exceptions but, generally speaking, a PNP will have lower rbb' than its NPN counterpart and therefore be quieter in this application. I think good current gain is still important here so options are more limited now regarding what's still available but, the old Toshiba 2SA1316BL would be a good one which you can still find.
ZTX951 seems to be the current choice.
 
I could upload the complete schematic of one channel. Would that be allowed or not?
Please do. Nobody here wants to behringerize a Midas console.
What does OSI stand for?
As JR mentioned, it's the optimum source impedance. Typically, it's equal to the ratio of the noise voltage to the noise current, which are usually spec'd for opamps as noise voltage density and noise current density. A little more complex for transistors, since noise current and noise voltage vary with quiescent current, and base spread resistance (rbb') comes into play.
 
One thing I discovered about improved sum bus technology is that when you remove that noise contribution below significance we still have to deal with mic preamp noise, mic self noise, and room noise.
That's why I said I've never had any problems with noise. To me the console seems to be the least noisy thing in the equation.
I don't see anything wrong with using a 5532 as the op-amp.
But would the trans amp benefit from using a "better" op amp like the LM49710 because it has more open loop gain?
ZTX951 seems to be the current choice.
Thank you! I couldn't find the Toshiba 2SA1316BL anywhere to buy.
A little more complex for transistors, since noise current and noise voltage vary with quiescent current, and base spread resistance (rbb') comes into play.
Oof, no idea how to make a decision on that since the datasheet for the ZTX951 isn't providing the specs to calculate that... Also I'm confused. What are we actually trying to do? Match the impedance of the transistor to the bus or to the op amp? What would the result of an impedance mismatch be? More noise? But can it even get worse than a single 5532? If I calculated correctly the quiescent current is set to approx 2.5 mA with R1 in the original Soundcraft 6000 schematic. So I guess I'll just start from this exact schematic but with the ZTX951 and 10k NFB.
Please do. Nobody here wants to behringerize a Midas console.
I attached the schematic of one mono channel. The virtual ground of the summing amp can be jumpered to GNDREF (standard) or MIXREF. I'm also wondering why there is a 75 Ohms resistor from the non-inverting input of the summing op amp to ground (see post #117)?
 

Attachments

  • Mono Input Sub.JPG
    Mono Input Sub.JPG
    109.3 KB · Views: 31
  • Mono Input.JPG
    Mono Input.JPG
    215.4 KB · Views: 36
That's why I said I've never had any problems with noise. To me the console seems to be the least noisy thing in the equation.
I know it's a slight subject of disagreement with JR. My view is that condenser mic noise and room noise have a noticeably different sonic character than noise due to electronics and many musicians discern this difference and understand that room noise is not the console's fault.
But would the trans amp benefit from using a "better" op amp like the LM49710 because it has more open loop gain?
It would certainly; we're counting on you io report the results of the comparison. :LOL:
Oof, no idea how to make a decision on that since the datasheet for the ZTX951 isn't providing the specs to calculate that...
I suggest you go to the proaudiodesign forum. There's a whole thread there.
Also I'm confused. What are we actually trying to do? Match the impedance of the transistor to the bus or to the op amp?
No. You have to grok the concept of OSI, which has very little to do with impedance matching.
If I calculated correctly the quiescent current is set to approx 2.5 mA with R1 in the original Soundcraft 6000 schematic.
So this makes the impedance of the CB stage about 10 ohms. Actually a little more due to the emitter spread resistance. But it results in a very low noise voltage density.
It's like this 10 ohm resistor was in series with the source, so it should be minimized, until the noise current becomes too high. (noise current increases as base current increases)
I attached the schematic of one mono channel. The virtual ground of the summing amp can be jumpered to GNDREF (standard) or MIXREF.
I can't read any labels in your pic.
Anyway I'm always suspicious about this type of arrangement. Makes me think that the designer was not quite sure he had made the good choice, so he left that open to someone else.
I've seen several mixers that had multiple "grounds", in particular one bus for the summing amps "ground". The most common mistake is forgetting to put a sampling resistor between channel ground and this bus. Anyway it doesn't play well with the most common arrangement that disconnects the feed resistor from the bus on unused channels.
An extreme case would be only one channel routed; with the summing amp ground sampling all the channels, the residual noise of all the channels is still there.
I've often found that I got a better overall performance using this summing amps bus to reinforce the main ground bus.
I'm also wondering why there is a 75 Ohms resistor from the non-inverting input of the summing op amp to ground (see post #117)?
Yes, I noticed that. It may be an attempt at fixing an imaginary offset problem or a stability issue...
 
But would the trans amp benefit from using a "better" op amp like the LM49710 because it has more open loop gain?

I've not used it for anything myself but, all other things being equal, then sure, try it. Measure the 10kHz distortion at your 34dB gain needed to get 40 sources back to unity and see if there's an improvement.


Thank you! I couldn't find the Toshiba 2SA1316BL anywhere to buy.

I don't know where you're located but, if in the EU anywhere, I've bought stuff from these guys:

https://www.donberg.co.uk/descript/2/2sa_1316bl.htm

Oof, no idea how to make a decision on that since the datasheet for the ZTX951 isn't providing the specs to calculate that... Also I'm confused. What are we actually trying to do? Match the impedance of the transistor to the bus or to the op amp? What would the result of an impedance mismatch be? More noise? But can it even get worse than a single 5532? If I calculated correctly the quiescent current is set to approx 2.5 mA with R1 in the original Soundcraft 6000 schematic. So I guess I'll just start from this exact schematic but with the ZTX951 and 10k NFB.

You're looking to match the transistor to the bus in terms of noise. There are no noise curves shown for the ZTX since it wasn't designed with us audio folks in mind but, somewhere around 2.5mA sounds like a good jumping off point.


I attached the schematic of one mono channel. The virtual ground of the summing amp can be jumpered to GNDREF (standard) or MIXREF. I'm also wondering why there is a 75 Ohms resistor from the non-inverting input of the summing op amp to ground (see post #117)?

The resistor from non inverting to ground will be in case you lose ground on a ground cancelling bus.
 
It would certainly; we're counting on you io report the results of the comparison.
Will do! Might take a while though. At the moment I'm still busy recapping some channels and comparing different electrolytic caps. Unforunately I don't have any expensive THD analyzer. I can measure frequency responses though and noticed the low end dropping off on some channels. The console is old and the caps are at the end of their life. So the most important piece of measuring equipment are my ears. I already have the impression that I'm hearing ghosts and will soon go crazy.
Anyways, I love this mixing console, especially the EQs and it's got more features than I actually need. I'm using it in my home studio and I'm feeding the channels with a 32 ch Antelope interface. I like to have some channels left for AUX/FX returns. Most important to me is variable low and high shelf filters, 6 - 8 AUX, 32 ch + at least 6 stereo and having meters for every channel. So far, I have not found any other console that fits my needs better than the Midas. Sure there are larger desks like an SSL, but even though I've already had the opportunity to buy one I'm afraid that I would spend more time with a soldering iron than actually producing music. The XL200 is keeping me pretty busy in terms of service work but nothing dramatic. Overall very good quality ALPS part I'd say.
Just some background information why I'd like to stay with this mixer.

I can't read any labels in your pic.
Sorry, the resolution was unfortunately downgraded during upload. Find the schematics here in full resolution:
Mono-Input
Mono-Input-Sub

I've seen several mixers that had multiple "grounds", in particular one bus for the summing amps "ground". The most common mistake is forgetting to put a sampling resistor between channel ground and this bus. Anyway it doesn't play well with the most common arrangement that disconnects the feed resistor from the bus on unused channels.
An extreme case would be only one channel routed; with the summing amp ground sampling all the channels, the residual noise of all the channels is still there.
I've often found that I got a better overall performance using this summing amps bus to reinforce the main ground bus.
I think they are connected with a resistor. Unused channels are not disconnected in the XL200. All channels are always connected to the busses. So you mean just connecting the busses?
Yes, I noticed that. It may be an attempt at fixing an imaginary offset problem or a stability issue...
But it won't hurt leaving them in, even when I install a trans amp?
You're looking to match the transistor to the bus in terms of noise. There are no noise curves shown for the ZTX since it wasn't designed with us audio folks in mind but, somewhere around 2.5mA sounds like a good jumping off point.
Thank you! I will try that.
I don't know where you're located but, if in the EU anywhere, I've bought stuff from these guys:
Yeah, EU. They got 30€ minimum order value though... But I might try that.
 
Will do! Might take a while though. At the moment I'm still busy recapping some channels and comparing different electrolytic caps. Unforunately I don't have any expensive THD analyzer.
If you have a decent soundcard, you mate it with REW and here's your high performance THD analyzer.
I'm using it in my home studio and I'm feeding the channels with a 32 ch Antelope interface.
I believe you can use REW with it, since it has an ASIO driver.
Sorry, the resolution was unfortunately downgraded during upload. Find the schematics here in full resolution:
Mono-Input
Mono-Input-Sub
I think they are connected with a resistor. Unused channels are not disconnected in the XL200. All channels are always connected to the busses.
That's not what I see. The feed resistors are switched either to the bus or to ground, which is the best solution in terms of noise, but the reference bus are also switched, which is the thing to do, so the ground noise is correctly sampled. I would not have used 10k resistors in that position, because it more than doubles the impedance seen by the summing amps (on account there's only one reference bus for two active bus). I would rather have used 100 ohm resistors.
This makes me even more suspicious as to what the designer understood (or not) about summing arrangements. There is no doubt that, when properly implemented, connecting the summing amp reference to the ground sensing bus is the best solution.
It maybe that the possibility to connect to the general ground would be for factory testing, and not for permanent use.
But it won't hurt leaving them in, even when I install a trans amp?
One more reason to install a transamp. But in order to reap all the noise benefits, you need to replace the ground sensing resistors with a much lower value.
The noise is essentially constituted of Johnson noise in the bus resistance(s) and the input noise of the summing amp.
The former is largely dominant. (more than) doubling the value by having the same value in the ground sensing reference increases the noise by almost 3dB.
 
Back
Top