balanced vs unbalanced mixer

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
Have done the level diagram already,and thanks, I know what that is. what I have not had the chance to test is the interaction (real music) the tracks will have an how much summing will up the level. If you consider something like six sources mixed at similar level, what will happen to the sum? I will try that with me test channels and maybe eight inputs from a mix, but I did digress a little, as so many times before with this project.

My project is just about not mixing in the DAW, so that is ruled out, and yes, I know you could have faders to talk with the DAW, but hey, who wants a HUI mixer if one can get a real console? And I like the idea of having analogue summing right in the box.

Th scope thing is really the next step, I wanted one for years and even had a 10 Meg Russian 'micro' scope but that was really just junk.

I haven't checked right what happens when I switch off the channels as they are switched off via two fets like in the Studer 990. but I am aware of the issue.
I will have 12 Mono channels, 4 stereo channels go to the mix so this is a limited issue, 20 channels to mix is not a giant board. I am considering using multi level partial summing even it might seem as an overkill,  I have reduced the impedance pan pots to 5k and that seems to help as this was the biggest contributor still have to check on fet input opamps there to see what it does to performance. the individual channel is at roughly -100dBu noise (input terminated at 100ohms) or better.

-cheers,

Michael
 
audiomixer said:
Have done the level diagram already,and thanks, I know what that is. what I have not had the chance to test is the interaction (real music) the tracks will have an how much summing will up the level. If you consider something like six sources mixed at similar level, what will happen to the sum? I will try that with me test channels and maybe eight inputs from a mix, but I did digress a little, as so many times before with this project.
theory says that uncorrelated signals combine quadratically, i.e. Vtot² = V1² + V2² + ... +Vn²
2 channels = +3dB, 4 channels = +6dB, 6 channels = +7.78dB
But in music, signals are not 100% uncorrelated. Stereo signals are obviously correlated, but even parts played by different instruments may show some correlation - if they were 100% uncorrelated, that wouldn't be music.
I will have 12 Mono channels, 4 stereo channels go to the mix so this is a limited issue, 20 channels to mix is not a giant board. I am considering using multi level partial summing even it might seem as an overkill,  I have reduced the impedance pan pots to 5k and that seems to help
What type of pan-pot arrangement are you using? I've seen so many mixers where the pan-pot arrangement created an inappropriate loss leading to a loss of 7-10 dB in noise performance...
 
While there is nothing wrong with trying to make every circuit block as low noise as possible, in larger system design we must weigh the cost-benefit of throwing too much time, money, and effort into one area, that may not result in significant overall improvement. 

This is the difference between bottom/up noise floor focus, and top/down dynamic range management.

Just like multiple noise sources in perfect summation combine quadratically, noise floor contributions from multiple circuit blocks in series combine similarly. Since all noise "should" be incoherent, the rule of thumb is that two equal noise paths would sum to +3 dB, but in general for typical design examples we are combining very different noise levels.  If the lower noise path is just 10dB lower noise than the higher noise path, they will combine to be only +0.8dB more noise. More than 10 dB lower makes even less contribution.

As I mentioned earlier, design focus for low noise design should be on the noisiest stages since they will contribute the most noise and dominate the combined path total. Fast forwarded to where we end up...  an even noiseless mic preamp ends up dominated by the Johnson noise of the source resistance of the microphone, boosted by perfect gain. A modest sized sum bus using premium parts, will be well below this mic noise (not to mention room noise). In extreme cases (I've designed consoles with more than 100 stems in L/R mix), bus design requires special attention, but most cases are manageable with modern off the shelf parts and common approaches.

Simple (low noise gain) I/O circuitry is generally so far below dominant noise sources, to be hard to even measure their contribution to total path in use, let alone hear. (of course you can always hear a bad design). 

It is good practice to approach this top down, or managing the worst case noise sources in the entire path, and hold lower gain circuit blocks in proper perspective.

Of course, it is fun to make every single circuit block as good as possible, just prepare yourself for disappointment, if you don't hear a night and day improvement in the total path noise.       

JR

PS: Something I figured out years ago is that high noise gain circuitry deviates from linear in more ways than just noise. Distortion and phase shift also degrade with increasing noise gain (actually lower "loop gain margin" or the difference between open loop and closed loop gain).  Reducing noise, generally reduces distortion and phase shift too.
 
I have made some progress on this, I finally found out how to get reliable measurements..... I was getting all kind of ground and hum noise in my measurements until I put the demo board on top of a shielded (insulated of course) piece of sheet metal tied to ground. seems my wood table has a distinct effect on the circuit at that low level noise measurements or maybe its my lap being just underneath working as a 50hz antenna. I have to switch off all unused power supplies (laptop, soldering iron....) but now I get reliable measurements that are repeatable. before one (identical) channel was noisier by 4db, and variable with the amount of pressure applied - thats all gone now.

ground lead length and a scope type probe are also a good idea. I made one from a piece of coax and an old multimeter lead, shielded the handle with foil and wrapped it in shrink tubing. I have a short ground lead from there to connect to a ground test pin right on the board.

this made it finally possible to characterize the different summing stages and the influence of the noise in the stage before. my test rig sums four ne5532 buffers set to produce a continuous noise voltage comparable to my later input channels. right now the best summing stage I made is from a hybrid transistor / opamp (douglas self again...) running on very low summing resistors (1k, maybe even less). second best is a simple ne5532 as a summing amp with maybe 1 dB more noise when summing the four channels. I know I cant leave the buss resistors that low, but I plan to have a more beefy summing amp that will be able to drive the low impedance. to drive lower buss resistors (i tried 390 ohms) gives a slight edge, maybe 0.7dB for both hybrid and single opamp summing stage, but this will start to load the output buffer of the channels to much so it's more a theoretical experiment.

now to the interesting part - noise sources. the channels I have build so far have a noise floor of around -104dBu. I had to change the pan pot to 5k (from 10k standard value) and the surrounding resistors as this is the main noise source right now in the channel path. reducing that does positively affect the overall performance after summing because the self noise of the summing opamp with no buss resistor connected is around -107.5dBu for the ne5532 (1k sum resistors).
so adding channels quickly reduces the noise performance. with test rig I could set up a less noisy 'channel' running in at -106.5dBu - these 2.5 dB are almost preserved though the summing stage, so the channel noise does matter and sets the total performance right now.

I ran out of time last night to check the best case performance of the unbalanced testboard, but I will see if I can make that less noisy or if the pan pot (identical to the balanced and unbalanced channel....) sets the final noise of the input channel. so this stays a interesting option to optimize the noise performance

tips for young players:
- at that low levels shielding is a must, otherwise you will pick up hum from everywhere
- get reliable and repeatable measurements before making assumptions
- get a useful measurement technique with a insulated probe
 
audiomixer said:
I know I cant leave the buss resistors that low, but I plan to have a more beefy summing amp that will be able to drive the low impedance. to drive lower buss resistors (i tried 390 ohms) gives a slight edge, maybe 0.7dB for both hybrid and single opamp summing stage, but this will start to load the output buffer of the channels to much so it's more a theoretical experiment.
The lower the bus injection resistors, the higher the effect of the reference path resistance will be. It may not be measurable on a 4-post bus, but in larger configurations, you will have to take care of the copper section and contact resistance. With 1k resistors, 0.1 ohm introduces -80dB x-talk, not mentioning noise due to the return currents.
now to the interesting part - noise sources. the channels I have build so far have a noise floor of around -104dBu. I had to change the pan pot to 5k (from 10k standard value) and the surrounding resistors as this is the main noise source right now in the channel path.
Are you saying that you don't have pan buffers?
 
no, I have pan buffers alright, just had to lower the general pan circuit impedance.... I tested 5k pot, resistors of the active panpot halved. offness is still acceptable, but this is one thing I need to check in the final design. I am not sure whether I need absolute offness in the pan stage, I will probably be fine with 70dB or so.

I want to run a staggered - so four channels get (pre) summed and then these go to a final summing stage. again a folly, but hey..... why not (do not reply - this is just a rhetoric question).

right now the pan buffer is a venerable ne5532 - i might test a 5534 or a lm4562 or a even lower noise opa1612 to get that stage quieter. (I ran out of opamps again, this time i killed a few with a mistake on the routeing of the stripboard..... arrghh, beginners mistake!)

- cheers,

michael
 
audiomixer said:
no, I have pan buffers alright, just had to lower the general pan circuit impedance.... I tested 5k pot, resistors of the active panpot halved...
...right now the pan buffer is a venerable ne5532
That's quite strange. With 10k PP, the opamp sees ca. 1.5k, with 5k it sees 750 ohms. 5532's noise current In is 0.7 pA per sqrtHz, which computes at 100pA for 20kHz BW. So noise contribution from In is 150nV, which is >> than the noise contributed by En, which computes at 700nV.  With 5k PP, In contribution is 75nV, but the major contribution is still En. In addition In noise and Vn noise combine quadratically, so that shouldn't make a significant difference. With 10k resulting overall noise is 710nV, with 5k it is 705nV. Can you hear this?
How do you measure noise? Do you monitor the shape with an o'scope? Do you monitor the sound with headphones? Do you have switchable filters that allow separating hiss from hum or ultrasonic?
I have a feeling that what you measure is parasitic coupling of surrounding interference.
 
pan buffer might have been the wrong term.....

for now I have chosen to implement the active panpot as per douglas self small signal audio design. might be that the gain there and the feedback resistor activly contribute to the total noise. I will check later today, what the noise levels in real world. It's true I had lots of problems measuring that low noise before, but now I have become quite confident. I use a Neutrik XL2 with a shielded cable / probe so this has improved much, and adding a grounded bottom 'enclosure' helped in that respect, seems my (wood) table was a source for trouble at that uV level.

-104dBu would be around 4.8uV total, -106dBu 3.8uV  the XL2 is specified for 2uV (-112dBu) - this is what i read when shorting the instrument. would love to have access to a better measurement gear, but hey its what i got at hand.
your calculated 710nV are at -121dBu - definitively far from my real world result. you calculated for a passive pan pot followed by a unity gain buffer? you think trading the active panpot for a simpler loaded potentiometer with buffer would significantly reduce the noise? I might try that, true. by the way, the fader and postfader amp ahead are generating quite some noise too, of course. I have to measure their contribution too.

I am pretty sure nobody can hear the difference - true. but the result in terms of the summed 20 channels should be noticeable, no? I do not mean that you could not work with less of course, but I am keen on pushing the limits - and maybe the final implementation will stay in the real world.

- cheers,

Michael
 
Some more practical sum bus math, when you sum a few tens of channels, the input channel noise sums incoherently (sq rt of sum of squares) while bus amp noise is simple multiple, so 32x bus amp (actually 33x) may exceed sq rt of 32  (<6x) times pan buffer noise.

I have posted in some thread recently about this bottom up noise analysis being somewhat mooted by other larger noise sources in the same path. While there is nothing wrong with making circuits better than they need to be, it's only time and money (not my time or my money).

JR
 
audiomixer said:
pan buffer might have been the wrong term.....

for now I have chosen to implement the active panpot as per douglas self small signal audio design.
Excellent decision.
might be that the gain there and the feedback resistor activly contribute to the total noise.
That is true, and one the few drawbacks of this arrangement. Due to positive feedback, the noise performance in the mid-position is about 6-7 dB worse than the standard arrangement.
  I use a Neutrik XL2 with a shielded cable / probe so this has improved much, and adding a grounded bottom 'enclosure' helped in that respect, seems my (wood) table was a source for trouble at that uV level.
The XL2 is a nice instrument; being portable, you don't have to care about any added interface noise. You may use the 19kHz and the 100Hz or 400 Hz filters to isolate hiss from hum and ultrasonics.
-104dBu would be around 4.8uV total, -106dBu 3.8uV  the XL2 is specified for 2uV (-112dBu) - this is what i read when shorting the instrument. your calculated 710nV are at -121dBu - definitively far from my real world result. you calculated for a passive pan pot followed by a unity gain buffer?
That's correct. You know, real-life measurements are often quite far from theory. Shielding is paramount. You've just lifted a corner of the veil.
you think trading the active panpot for a simpler loaded potentiometer with buffer would significantly reduce the noise?
It may, in theory, but in practice, the noise should be low enough and there are significant advantages that I wouldn't miss.
by the way, the fader and postfader amp ahead are generating quite some noise too, of course. I have to measure their contribution too.
Typically, a postfader amp will have a noise gain of about 12dB, similar to the active pan-pot.
I am pretty sure nobody can hear the difference - true. but the result in terms of the summed 20 channels should be noticeable, no?
Yes and no... Noise combines quadratically, which means 20 channels will increase noise by about 13dB. The signals will combine also quadratically, so the S/N ratio will stay the same. In fact only totally uncorrelated signals combine quadratically. Musical signals are generally somewhat correlated, even when produced by separate instruments and recorded by different channels, so their combination should be a little more efficient.
And mixing is not about having all the channels summed in equal parts; eventually, the S/N performance is governed by a few dominant channels, voice, kick, snare...
 
so, with my new setup I have been able to get some better results (read: more precise readings)
all noise levels dBu, input terminated with 100 ohms, fader at -10dB (worst case)

unbalanced:                                      balanced:

input buffer
-111.7dBu                                        -109.9dBu

post fader buffer (fader gain 0dB)
-110.7dBu                                          -110.6dBu

Pan center with NE5532
-103.6dBu                                          -105.6dBu

Pan center with LM4562
-105.8dBu                                          -107dBu


so: the differences are minimal, not worth the risk of adding noise from unbalanced input and are probably due to the slightly different board layout. the fader and intermediate circuit are the main noise contributors, not the input. there might be a dB lost to the improvised construction here and there, but right now I have to trust my measurements and take these numbers. the active panpot adds most noise right now. reducing pot impedance and choosing optimum opamp reduces noise by some 2dB.
fader offness and panpot offness is more then adequate with 10k fader and 5k panpot.


I have posted in some thread recently about this bottom up noise analysis being somewhat mooted by other larger noise sources in the same path. While there is nothing wrong with making circuits better than they need to be, it's only time and money (not my time or my money).

no money involved, neither at engineering level nor regarding production. no date set for manufacturing (though I had sought in the beginning that I would be finished in four months time.... )

cheers,

Michael

 

Latest posts

Back
Top