Balanced Summing

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

ej_whyte

Well-known member
Joined
Nov 12, 2010
Messages
263
Location
Cambridge, UK
Hey,

Long time no visit! Hope things are still as active here as ever.

What do people think of the following schematic. Obviously it is just a rough sketch, things such as RF & overvoltage protection, DC blocking etc are all not included for now. I am mainly concerned with low noise and CMRR. Possibly will need to increase R3, R4, R7 & R8 to accomodate the drive capabilites of the OPA1632, I think it can drive 2K.

Thank you
 

Attachments

  • Input.pdf
    12.1 KB · Views: 175
Looks like a lot of expensive parts for a mono sum bus.

So 2x that per channel for stereo?

If you are asking what I think about the concept, I even have a small problem with that rudimentary schematic. The Vcom pin on the 1632 is deigned to grab an output related CM reference. Your schematic shows all the 0V (grounds) tied together. At a minimum the 0V on the output differential should grab an output reference 0V, likewise I can imagine a local 0V for the input that is different from the master section 0V. If you aren't going to use the Vcom capability you could just use some inverting opamps. 

Note all of these 0V nodes ultimately connect together at the PS common  but the practical layout might benefit from keeping them segregated locally.

JR

PS: The 1632 looks like a good part for driving a premium A/D which is not as trivial as some people think.

 
JohnRoberts said:
Looks like a lot of expensive parts for a mono sum bus.

quite obviously this is about very low noise summing and not about budget. I do know that in most topologies noise will be set by mic / preamp noise floor, but there is a development in ultra low noise & distortion..... good thing. I have the same OPA 1632  on my bench and my mixing bus is prepared for balanced buses for the main LR bus... looking forward to what this might bring.
others pay tons of money on transformers, so hey, let's embrace those low noise opamps and squeeze the last bit of performance out of them, yeeehaa!



- Michael
 
Not sure what the requirement is but the 1632 is redundant.  You can drive R7/8 directly from the 4562 for a tiny bit less THD & noise.

Also, can a single 1632 replace the 2x49990s?

JR, I think you were a fan of Balanced Summing.  Can you show us how you did it in da old days?

We never did at Calrec (early 80s) but discussed it often.
 
ricardo said:
JR, I think you were a fan of Balanced Summing.  Can you show us how you did it in da old days?
Actually no, I never did one either, but it should do no harm if executed well (ie Ov references sorted). 

There is a tiny benefit (just like with differential outputs) from 2x the coherent signal voltage, but inly 2x the incoherent noise, so something like 3dB benefit.

There is a more significant benefit from the signal being sent relative to itself, so local 0V differences are common mode and cancel out in the differential stage following the sum amps.

The hoped for benefit from balanced (equal) impedance bus sends picking up noise the same is pretty much non-existant, inside a grounded metal chassis console noise pickup should be under control. 

I am generally a fan of using one hot leg differential bus structures in cost sensitive designs (where how many opamps matter), while I intentionally imbalance the impedance making the cold 0v send side low impedance resistors to shave a little Johnson noise.

JR

 
audiomixer said:
JohnRoberts said:
Looks like a lot of expensive parts for a mono sum bus.

quite obviously this is about very low noise summing and not about budget. I do know that in most topologies noise will be set by mic / preamp noise floor,
Thank you for saying that so I didn't have to.  8)
but there is a development in ultra low noise & distortion..... good thing.
New modern parts are always nice, but I tend to look at such things agnostic about the specific parts and looking at the pros and cons of different topology. So all things equal is this a better approach or just a better part. (while better parts are OK). 

In my 1980 article about console design I identified the sum bus as one of the several problem areas for console design because of the noise gain associated with combining N stems (typically make-up/noise gain of N+1). 

This "balanced sum bus" is a variant on virtual earth topology so subject to same N+1 noise gain.  (noise gain in VE summer is responsible for N+1 noise, phase shift., and distortion.
I have the same OPA 1632  on my bench and my mixing bus is prepared for balanced buses for the main LR bus... looking forward to what this might bring.
others pay tons of money on transformers, so hey, let's embrace those low noise opamps and squeeze the last bit of performance out of them, yeeehaa!



- Michael

I am not a fan of transformers for audio paths so I expect this (1632) instead of transformers should be an improvement.

From my 30+ YO article I observed two different sum bus topologies better than virtual earth (in my judgement).

One was based on Buff's "Transamp" mic preamp topology, perhaps better known around here as "Cohen". While it too suffered from N+1 noise increase, the beauty of the transamp topology in a virtual earth sum bus topology is that the increased number of stems assigned to the bus increased the open loop gain of the transamp the same amount as the closed loop gain increased so, the phase shift and distortion from even a large number channels summed together was not much worse than an unity gain opamp. The transamp input noise was amplified N+1 though.

My favorite approach (then and now) was to synthesize current sources to take the place of the sum resistors feeding the bus. While the relatively small noise from these active synthetic current sources was accumulated (incoherently) in the bus the virtual earth sum amp did not suffer from any of the typical N+1 noise gain, so noise, and phase shift, and distortion were all quite low even with a large number of stems feeding the bus.

I have designed consoles using both of these approaches. They work.

I wish analog consoles were still relevant so I would have an excuse to apply modern parts to my 20-30 year old designs. My friends who are still trying to make and sell analog consoles have fallen on hard times, as modern digital consoles kick their ass with tons of features for silly low retail prices. Maybe they need to find some cost is no object customers.  There is never enough of them when you really need them.  ;D

JR 

PS: Yes I know there are still some people using analog consoles for recording... problem is there are so many decent old ones for sale used, there isn't much market for new ones.

PPS: I am not trying to rain on EJ's design, get the 0V sorted and build it. let us know how it sounds.
 
JohnRoberts said:
From my 30+ YO article I observed two different sum bus topologies better than virtual earth (in my judgement).

One was based on Buff's "Transamp" mic preamp topology, perhaps better known around here as "Cohen". While it too suffered from N+1 noise increase, the beauty of the transamp topology in a virtual earth sum bus topology is that the increased number of stems assigned to the bus increased the open loop gain of the transamp the same amount as the closed loop gain increased so, the phase shift and distortion from even a large number channels summed together was not much worse than an unity gain opamp. The transamp input noise was amplified N+1 though.
Any links to a tried & tested example of this?

My favorite approach (then and now) was to synthesize current sources to take the place of the sum resistors feeding the bus. While the relatively small noise from these active synthetic current sources was accumulated (incoherently) in the bus the virtual earth sum amp did not suffer from any of the typical N+1 noise gain, so noise, and phase shift, and distortion were all quite low even with a large number of stems feeding the bus.
This was the favourite Calrec (1980s) approach but we did it mainly for automation using current steering VCAs like the Allison.  I think Allison became something else after I left the mixing desk scene.
 
ricardo said:
JohnRoberts said:
From my 30+ YO article I observed two different sum bus topologies better than virtual earth (in my judgement).

One was based on Buff's "Transamp" mic preamp topology, perhaps better known around here as "Cohen". While it too suffered from N+1 noise increase, the beauty of the transamp topology in a virtual earth sum bus topology is that the increased number of stems assigned to the bus increased the open loop gain of the transamp the same amount as the closed loop gain increased so, the phase shift and distortion from even a large number channels summed together was not much worse than an unity gain opamp. The transamp input noise was amplified N+1 though.
Any links to a tried & tested example of this?
I don't have a schematic of the console I did using this (Loft Modular Devices late '70s early '80s). IIRC the transamp app notes had an application  OK see figure 6 in  http://www.ka-electronics.com/images/pdf/valley_people_trans_amp_applications.pdf Wayne has a bunch of old data sheets and notes.
My favorite approach (then and now) was to synthesize current sources to take the place of the sum resistors feeding the bus. While the relatively small noise from these active synthetic current sources was accumulated (incoherently) in the bus the virtual earth sum amp did not suffer from any of the typical N+1 noise gain, so noise, and phase shift, and distortion were all quite low even with a large number of stems feeding the bus.
This was the favourite Calrec (1980s) approach but we did it mainly for automation using current steering VCAs like the Allison.  I think Allison became something else after I left the mixing desk scene.
Yup, VCA current outputs should work (assuming the VCA output capacitance from a bunch in parallel is noit high enough to destabilize the bus amp)... I have invested some thought into doing this with modern lower noise VCAs, but my simple opamp based synthetic current sources are much lower noise than typical VCAs.

I did one Loft console with the current source summing and used a DBX VCA as the master L/R sum amp. Because it was running at low noise gain the noise of the VCA was not an issue. It was a pretty cool console (late '70s). 

JR
 
Thank you for all the replies, very interesting.

Yes obviously this is not aiming for low cost (and will be x2 for stereo as you guessed JR), and to some level it is probably unnecessary as the noise will be dominated by other sources upstream, but I would like to ensure that this is the case and make the mixer perform as well as possible (within the budget). Part of it is also a bit of specmanship, I have already built 1 mixer and I would like this one to have significantly better performance, even if it is just to prove to myself that I can do it!

Good tip on the 0V JR, I will look in to this.

ricardo said:
Not sure what the requirement is but the 1632 is redundant.  You can drive R7/8 directly from the 4562 for a tiny bit less THD & noise.

Also, can a single 1632 replace the 2x49990s?

JR, I think you were a fan of Balanced Summing.  Can you show us how you did it in da old days?

We never did at Calrec (early 80s) but discussed it often.

The input to the pair of LM4562 is actually the input to the mixer as a whole (excluding EMI/RF/Overvoltage gubbins, but no 'channel strip' as such), and so I had the OPA1632 to get rid of CM noise before it went to the mix bus. My thinking being that any imbalance in mix bus resistors and R9/R10 would be greatly amplified due to the gain of U3 and U4, so better to get rid of it before hand. This is why R1-4 are 0.1% but the mix bus onwards are standard 1% (although I know having 0.1% all the way through would improve things further). Knowing this, is the 1632 still redundant and am I barking up the wrong tree?

I did think about the 1632 in place of the LME49990's but wasn't sure how it would work regarding virtual earth on the +ve input of the 1632, I need to look in to this, TI have some nice app notes of FDAs.

Cheers
 
ej_whyte said:
The input to the pair of LM4562 is actually the input to the mixer as a whole (excluding EMI/RF/Overvoltage gubbins, but no 'channel strip' as such), and so I had the OPA1632 to get rid of CM noise before it went to the mix bus. ... is the 1632 still redundant and am I barking up the wrong tree?
Good point though like JR, I find it all a bit too complex.

I did think about the 1632 in place of the LME49990's but wasn't sure how it would work regarding virtual earth on the +ve input of the 1632.
1632 datasheet shows nothing of the internals but I suspect it's very like the Transamps in behaviour.

If you do try 1632 to replace the 2x49990s and it works, please let us know.  You could also ask Rochey who has direct access to the TI application engineers.
 
I don't think both Input legs of a 1632 would form a virtual earth type bus amp if just used instead of the 49990. while digging into this part a while ago I did not find a -1 gain application..... when used in single ended input configuration it is shown in a non-inverting configuration.

when Vcom is not required the datasheet advises to leave the pin unconnected. 0V reference is a non-issue.

one more thing you might want to look at is the summing amps drive current. you need to remind that it must be able to provide the same current then all the channels output opamps combined.... so with eight times 1k the summing opamp has to be able to deliver quite some punch...

- Michael
 
audiomixer said:
I don't think both Input legs of a 1632 would form a virtual earth type bus amp if just used instead of the 49990. while digging into this part a while ago I did not find a -1 gain application..... when used in single ended input configuration it is shown in a non-inverting configuration.

when Vcom is not required the datasheet advises to leave the pin unconnected. 0V reference is a non-issue.
The TI sloa54d has some internal details and some applications.

Without looking too closely, my guess is that it would do as a Balanced summer.  But that's on a very quick glance.  Certainly interesting enough to try.  TI imply its black box behaviour is symmetrical.
 
ricardo said:
1632 datasheet shows nothing of the internals but I suspect it's very like the Transamps in behaviour.

Looking at the data sheet for the 1632 I have no reason to expect it to be anything like the transamp, Other than having inputs and outputs.  The 1632 is designed as a front end for A/D convertors, thus the Vcom input to reference it's output to the A/D. I suspect it also has very low impedance output drive characteristic to deal with A/D input S/H noise.

OTOH the Transamp is designed with asymmetrical high and low input impedance for + and - inputs. Plus the Transamp topology is optimized for high gain like mic preamp or large sum bus.

They look like different (silicon based) critters to me.

JR




 
audiomixer said:
when Vcom is not required the datasheet advises to leave the pin unconnected. 0V reference is a non-issue.

one more thing you might want to look at is the summing amps drive current. you need to remind that it must be able to provide the same current then all the channels output opamps combined.... so with eight times 1k the summing opamp has to be able to deliver quite some punch...

- Michael

Yes I had read that but was wary as I had always thought you shouldn't leave any pins floating, I guess if it tells you to do it in the datasheet it should be OK.

I didn't know that about the summing amps current though, could you elaborate? I assume that you mean it has to be able to provide the same current that the channel opamps are actually provideing, not their max current as spec'd in the datasheet?

Cheers
 
'magic' opamp action says:*

rule 2 - the output will try to keep both inputs the at the same voltage level *

the positive input is invariably tied to the 0V reference. the channels are driven at some voltage, U/R = I so you get a sum of currents from all of the channels at the virtual ground node. the opamp output will drive (sink?) the same current trough the feedback resistor (R9, R10 in your design) so that the voltage level is right at 0V on that node, like on the non-inverting input.

in my understanding you will not have to provide n times (n channels) the full current drive capability of the individual opamps driving 1k of course, due to the nature of the signals you want to sum. but imagine you have the same sine generator on all channels - that would sum right away and you would have 18dB gain. would be fun to know just how much headroom you need for a hot mix...

- Michael

* see eevblog.com episode #600 for a good explanation of opamp action....
 
JohnRoberts said:
Looking at the data sheet for the 1632 I have no reason to expect it to be anything like the transamp, Other than having inputs and outputs.  The 1632 is designed as a front end for A/D convertors, thus the Vcom input to reference it's output to the A/D. I suspect it also has very low impedance output drive characteristic to deal with A/D input S/H noise.

OTOH the Transamp is designed with asymmetrical high and low input impedance for + and - inputs. Plus the Transamp topology is optimized for high gain like mic preamp or large sum bus.

They look like different (silicon based) critters to me.
You're right.  TI sloa54d has a simplified internal circuit and its nothing like Transamp or THAT 1510 etc.

I'm too lazy to try it out in LTspice but the 'fully symmetrical' black box behaviour implies it might work.  Any SPICE gurus willing to have a go .. or better still anyone wanna try it in 'real life'?
_____________________

My favorite approach (then and now) was to synthesize current sources to take the place of the sum resistors feeding the bus. While the relatively small noise from these active synthetic current sources was accumulated (incoherently) in the bus the virtual earth sum amp did not suffer from any of the typical N+1 noise gain, so noise, and phase shift, and distortion were all quite low even with a large number of stems feeding the bus.
May we, the unwashed masses, beg JR for a rough sketch of his

'synthetic current sources' ?
 
ricardo said:
My favorite approach (then and now) was to synthesize current sources to take the place of the sum resistors feeding the bus. While the relatively small noise from these active synthetic current sources was accumulated (incoherently) in the bus the virtual earth sum amp did not suffer from any of the typical N+1 noise gain, so noise, and phase shift, and distortion were all quite low even with a large number of stems feeding the bus.
May we, the unwashed masses, beg JR for a rough sketch of his

'synthetic current sources' ?
I attached a piece of schematic from last (big) console I did. This is how I did it for the Peavey/AMR production series, a large old school "split" configuration. The monitor section (right side of console) had three of these per strip so 72x in the 24 bus model.

I can even tell you what I did wrong. We all make mistakes, or see things after a design is finished that we would improve. In this circuit there were two things. One was that I left the current sources connected to the L/R bus even when not assigned. So even though their individual noise contribution was small, 72x a few uV is audible when all inputs are muted. I had this fixed on the bench (not trivial to switch it out silently) and ready to implement as a rolling production change, but the program was killed before I could do this.  :'( :'(  My second change, would be that the synthesized current sources need to be driven from a low impedance. My implementation of driving the current sources from the pan pot caused a measurable degradation in the output impedance of the current source when the changing source resistance from the pan pot altered the quality of the resistance matching of the two input legs for the current source. Note: As i shared with users of these consoles, when the monitor section is not being used, keep all bus feeds over in monitor section panned hard right or hard left for lowest noise floor.

Anyone familiar with this circuit will notice that my 30.1k r312 does not exactly equal the 15k of r314 + 15k r319, like it should for highest output impedance. In practice with a fully loaded frame, I was measuring on the order of a few hundred K resistance on the bus so while not perfect, significantly better than the expected 73x noise gain of a virtual earth bus. My input side of the console used convention send resistors since it was impractical to assign a single current source to multiple outputs (it could be done but with more cost and complexity).

If I were to revisit this today for a cost is no object implementation, I would use a modern low noise JFET opamp that would improve upon the roughly 3uV output noise of the TL07x. lower noise. I would improve the resistor matching. by using 2 of the same value in series to make the 2x leg.

I even talked to THAT corp engineers about the possibility of re-assigning the precision resistors in one of their precision balanced i/o ICs to make a kick-ass current source, but the fly in that ointment was even though they had enough precision resistors to accomplish this, and it would only take a metallization layer (couple $10k mask), their inputs all have extensive protection circuity to prevent static failures, and my re-map of the resistors would leave some pins unprotected. That, and the fact that I do not have a large market for this killed that project.

Like i said I wish analog consoles were still relevant... I could do analog better,  8) but yawn we live in a digital age. 

JR

PS: The official name for this is a "Wildar" current source. I first saw this in an old national applications manual (1970s), but their proposed circuit used 1M and 2M resistors, which are much too noisy for an audio path and too low current. Scaling the resistor values down gives a more reasonable current output and noise level that was practical.  Like I said it really works.    ;D
 

Attachments

  • currentsource.jpg
    currentsource.jpg
    29.8 KB · Views: 102
JR,

A simple collector current source would be too crappy for this, no? or any other '3 leg' device. a mirror for example.

The sum amp for the current sources is just like any VE or have any difference?

JS
 
Back
Top