Balanced Summing

joaquins said:

JR,

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

Click to expand...

I did a rigorous engineering analysis if this back in the late '70s and if you think about this making bilateral (AC) current sources using transistors is not trivial to make them low noise, low distortion etc. At best you end up with a unity gain current ratioing VCA. But even using today's, several decade later best technology VCA is higher noise and distortion, than my simple 5 precision resistors and an opamp (or should I say Wildar's topology). My secret sauce contribution for taking that well know circuit and applying it to an audio sum bus effectively was scaling down the impedance of the resistors involved to get the noise floor down where it needed to be. 

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

Click to expand...

Yes a standard VE sum amp, but now since that bus amp is running at very low noise gain there is no benefit from heroic measures to use a ultra low noise opamp there. In fact one Loft console I designed back in the 70s/80s used a DBX vca in place of the summing amp. Hows that for headroom?  8) 8)  but yes any common opamp will work for the bus amp, and another benefit of feeding the bus with current sources is that the local ground differences no longer matter. The current send is referenced relative to where it is sent from, and converted back to a voltage where it is received.

JS

Click to expand...

Another tidbit, I tried to patent this back then, and ran out of personal funds to pursue, when the lame patent examiner didn't understand the difference between combining "current sources", and combining "currents". All the textbook explanations for how a virtual earth summing amp works says that the input resistors convert their input voltages to currents that are then combined. but that is far different than summing current sources. I was paying for this out of my pocket, and didn't have the resources and patience to pay to educate the wet behind the ears patent examiner about the difference. This is not that unusual with patent applications, if the invention was that obvious, it would have most likely been done already.

JR

JohnRoberts said:

[...]Wildar[...]

Click to expand...

Are you sure you're not thinking of a Howland current pump? Your circuit looks a lot like it (write-up by Bob Pease).There's also the Widlar current source, which is nothing like your plan.

JD "this word, you keep using it. I don't think it means what you think it means" B.

jdbakker said:

JohnRoberts said:

[...]Wildar[...]

Click to expand...

Are you sure you're not thinking of a Howland current pump? Your circuit looks a lot like it (write-up by Bob Pease).There's also the Widlar current source, which is nothing like your plan.

JD "this word, you keep using it. I don't think it means what you think it means" B.

Click to expand...

Yes, yes yes, figure 5 "Improved Howland",, Not sure why I was calling it Wldar. 

OK, page 10 of this link http://www.ti.com/ww/en/bobpease/assets/AN-31.pdf Looks more like what I first read back in the 1970s when I learned of this.  This link was updated in 2002 to substitute LM394s for some discrete devices, but you can tell from the opamps they used that it is much older than 2002.

I never knew these sundry circuits had names until decades later, but no doubt somebody did it first. I may have even been corrected before (about the name)... Not committed to memory even now.  8)

I do not claim that I invented the (bilateral) current source, I just figured out how to use it in an effective sum bus topology.

JR
 

Maybe I'm as dumb as a patent examiner.

I don't see independent current sources, when you tie many of them all together, into a Virtual Earth mix-node.

If the "Virtual Earth" is good and earthy, the bus voltage will be ZERO. So R319 runs from zero, to +in, to R318, to zero.... you may as well skip the resistors and tie +in to zero.

If the mix node is not perfectly zero, R319 senses an error. But is this its OWN error of current in R314A, or an error of another channel in its R314B(etc), or an error of the summing amp in its Rfb? Seems to me, all errors sum on the one bus, nobody knows its "own" error.

I think it ends up one inverter driving one dumb 15K resistor per channel, plus a lot of "help" reducing ALL errors (which should perhaps be solely the sum-amp's job).

JR, could I ask you to dig up even more memories ..

Do you have a noise voltage at the output of the 72 channel mix?  And can you remember the size of the mixing resistor?

I ask this cos if I can dig up a spec for a Calrec M-series desk, I can get the equivalent noise voltage for it.

When the first M-series was 'commissioned', the BBC (could have been IBA) engineers complained that they could hear hum.  Ken Farrar pointed out that it was 10dB below 20kHz noise and that the only reason they could hear it was the new desk was some 10dB quieter than anything they had.

We solved that by simply (?!!) changing the internal operating level of this 72 channel desk.  I remember this cos it was my choice of Allison VCAs which set the noise level at the operating point. 

The summers used another Allison VCA for the summer gain, perhaps like how you used the DBX, but for feedback.
________________________

For Balanced Summing, here's how Ted Fletcher of Alice did it. http://www.theopiumdenproductions.com/balmix.jpg from http://groupdiy.com/index.php?topic=14512.msg0#new

I wouldn't call the Calrec example, Balanced Summing or even Quasi Balanced Summing.  It's just using 2 tracks on the bus for the virtual earth reference.

PRR said:

Maybe I'm as dumb as a patent examiner.

Click to expand...

Not even close. They are just doing OJT and paying dues to become patent lawyers...

I don't see independent current sources, when you tie many of them all together, into a Virtual Earth mix-node.

If the "Virtual Earth" is good and earthy, the bus voltage will be ZERO. So R319 runs from zero, to +in, to R318, to zero.... you may as well skip the resistors and tie +in to zero.

Click to expand...

Bzzt... Just like ground (my favorite word)  ;D doesn't equal 0V. The - input in a inverting opamp, or virtual earth, doesn't = 0V, even when the + input is tied to 0V.  It will = 0V + opamp output voltage divided by open loop gain. So if we put a scope probe on the - input (and it doesn't oscillate) we will see a tiny replica of the output voltage (phase shifted 90' by the compensation pole). 

If the mix node is not perfectly zero, R319 senses an error. But is this its OWN error of current in R314A, or an error of another channel in its R314B(etc), or an error of the summing amp in its Rfb? Seems to me, all errors sum on the one bus, nobody knows its "own" error.

Click to expand...

The "Howland improved yadda yadda"  current source doesn't care why there is a voltage at the output node. It just uses positive feedback to bootstrap up the output so no extra current flows in/out, other than the current determined by the input voltage difference.  They all do play nice together, and generally ignore AC voltage there.

I think it ends up one inverter driving one dumb 15K resistor per channel, plus a lot of "help" reducing ALL errors (which should perhaps be solely the sum-amp's job).

Click to expand...

Think about it... and trust me, it actually works. I have built many consoles using this approach. Back in the early days while I was bench testing this, I went as far as to feed a voltage source  back into bus, and nothing happens.

The main job of the synthesized current sources is to not draw or source any current due to voltage on the bus... that is literally how noise gain degrades bus amp performance. if there is 5 uV of noise on the - input N resistors to a low impedance will draw N x 5uV/R which then must be supplied by the feedback R for N+1x 5uV. With N current sources instead of  N resistors we only get the 5 uV of noise.

In production with those values I measured roughly 250-300k impedance to ground at the bus so tens of dB improvement over conventional resistors.

JR

PS: I used a related trick to sum signals without insertion loss for cascading multiple automatic mixers. Each 8 channel automatic mixer fed the two important signals through "HOWLAND" current sources with a rear panel termination resistor jumpered on a barrier strip. To combine any number of these automatic mixers, just wire the current outputs together so the currents sum, but remove all the termination resistors but one so the currents convert to same voltage as single mixer, no matter how many mixers were combined.  That eliminated a bunch of different make up gains for different system configurations.

ricardo said:

JR, could I ask you to dig up even more memories ..

Click to expand...

you can ask...

Do you have a noise voltage at the output of the 72 channel mix?  And can you remember the size of the mixing resistor?

Click to expand...

Actually it was 72 current sources for just the monitor section on the 24 bus board, and 36 conventional resistor assignment matrixes for the input strips,  and something like 4 dual effects return modules with assignment matrixes, so total something like 116 stems to the L/R mix.

Sorry no I have no idea what we spec'd the noise at. 

I ask this cos if I can dig up a spec for a Calrec M-series desk, I can get the equivalent noise voltage for it.

When the first M-series was 'commissioned', the BBC (could have been IBA) engineers complained that they could hear hum.  Ken Farrar pointed out that it was 10dB below 20kHz noise and that the only reason they could hear it was the new desk was some 10dB quieter than anything they had.

We solved that by simply (?!!) changing the internal operating level of this 72 channel desk.  I remember this cos it was my choice of Allison VCAs which set the noise level at the operating point. 

Click to expand...

I had a -10dB gain trim across the L/R sum amp so board could be operated with 10dB of loss to the L/R bus.

The summers used another Allison VCA for the summer gain, perhaps like how you used the DBX, but for feedback.

Click to expand...

Yup, I only did that for one Loft console but it was slick... I built an automatic fade feature into that master, and when the VCA was commanding attenuation the headroom of the bus actually increased.

I only used the Buff VCA in one Loft compressor, but Paul did good stuff. 

_______________________

For Balanced Summing, here's how Ted Fletcher of Alice did it. http://www.theopiumdenproductions.com/balmix.jpg from http://groupdiy.com/index.php?topic=14512.msg0#new

I wouldn't call the Calrec example, Balanced Summing or even Quasi Balanced Summing.  It's just using 2 tracks on the bus for the virtual earth reference.

Click to expand...

Different ways to skin the cat... just have to reference the send signal to the input 0V. and receive bus to master 0V. My favorite "budget" approach was pretty simple differential... the 100+ stem AMR was a hybrid mixed topology bus, and i regret I never got do a second generation... all the boys and girls were infatuated with Mackie 8 buses... so I was in the wrong price point for the time. 

JR

Lots to read through here and I will try to get my head round the Howland circuit.

Back to my original schematic for now, I just had a quick thought that off the top of my head the differential summing may be a false economy in this case. As far as I know, differential summing has 3 main advantages: 1. Reduced effect of potential differences along the ground bus 2. Reduced effect of EMI and other common mode signals 3. Reduced noise.

I believe in this box advantages 1 and 2 are negligible(?) as it is a physically small circuit. The OPA1632 has a max load impedance of 2k, which means R1-4 & R7-8 would have to be 4k each at a minimum. Now if I just went with single ended summing, there is a much wider choice of opamps and the minimum load could be much lower. This means that for example, if you used 2k resistors, they would be 3dB quieter already and you have just made up for the 3dB you could potentially gain from diff summing.

I need to look at this a lot closer as there is probably something I am missing here, what do people think?

Thanks

Just to be clear what I think if as "differential" summing can be performed with a simple single opamp for left and single opamp for right sum amps.

Imagine 3 buses;  + L-bus, + R bus, and - 0V bus.

So a simple virtual earth sum amp, could use 10k input and 10k feedback resistors, for left and right sum amps. At the same time as we feed +R and +L to their respective buses, we feed the 0V bus (signal ground?) from each input 0V through say a 100 ohm resistor to the 0V bus. In the master section, connect this 0V bus to the + inputs of the virtual earth opamps. To keep the ground references math accurate you need to connect one 100 ohm resistor from the master section local 0V to this same 0V bus.

This way ground errors or differences between the sundry input local 0v references, get divided down like a resistive pad in the 0V bus before getting multiplied by the N+1 noise gain, but feeding the + input this is opposite polarity from the normal signal feeding the - inputs so subtracts out. 

It may be easier to visualize the math if you used 10k resistors for the 0V bus too, but making them 100 ohm instead of 10k delivers the same math, but lower impedance grounds. (I've actually used lower than 100 ohm resistors, but it won't work with 0 ohm shorts.).

This only costs one more bus wire and a few resistors but works. Especially for a small mixer.

JR

Thanks for this JR.  I'd never considered this point of view.  I grovel at your feat in awe 

Yes thank you JR, how would that work in terms of CMRR though? The original reason for my OPA1632 was to remove any common mode signals that came in the input of the device before they got to the sum amps. Would any imbalances between the + and - busses not go through to be amplified by the summing amps?

Cheers

All,

What JR described a couple of posts ago has been discussed in this thread: http://groupdiy.com/index.php?topic=27351.msg328411#msg328411

The schematic in that thread comes from Fred Forssell's website http://www.forsselltech.com/media/attachments/Summing_Buss2.pdf

It is also discussed in Douglas Self's book "Small signal audio design"; Self uses the term "ground canceling summing systems" (pg 451, chapter Mixer Subsystems).

BR,

Rogy

Rogy said:

All,

What JR described a couple of posts ago has been discussed in this thread: http://groupdiy.com/index.php?topic=27351.msg328411#msg328411

The schematic in that thread comes from Fred Forssell's website http://www.forsselltech.com/media/attachments/Summing_Buss2.pdf

It is also discussed in Douglas Self's book "Small signal audio design"; Self uses the term "ground canceling summing systems" (pg 451, chapter Mixer Subsystems).

BR,

Rogy

Click to expand...

Yes, thanks, that Forsell schematic is exactly what I was talking about, less the servo. For simple mixer.. I am inclined to use smaller than 1k resistors for the 0V bus, but enough 1k in parallel will not contribute much noise. While I wouldn't expect it to be significant noise even using 1k resistors, the Johnson noise of this 0v bus gets added to the other noise sources associated with the sum amp input before being multiplied by N+1. Getting down around tens of ohms the Ov bus resembles connecting the signal grounds together, while still accommodating the differential subtraction math.

Looks like Fred still doesn't know how to spell bus.  ;D ;D

JR

ej_whyte said:

Yes thank you JR, how would that work in terms of CMRR though? The original reason for my OPA1632 was to remove any common mode signals that came in the input of the device before they got to the sum amps. Would any imbalances between the + and - busses not go through to be amplified by the summing amps?

Cheers

Click to expand...

in general you would use a balanced/differential receiver at the input to the mixer channel to properly reference the external audio signal to the local channel 0V. Then in the channel you do your level and pan and eq or whatever manipulations relative to that local 0V reference. Then when sending to the sum amp a L/R signal bus and 0V ground bus, will keep the ground difference between the 0V at the channel and 0V in the master common mode, since any voltage on the channel's local 0V is also superimposed on top of the L/R output voltage so cancels in the sum amp differential. 

If you mixer is not doing any manipulations at the channel you could simply send signal + to the L/R bus and signal - to the 0V bus. This would present an unbalanced impedance termination to the outside world so perhaps for feeding a sum bus from the outside world use equal value resistors in the 0V bus to be closer to balanced impedance.

JR