Summing mixers: a new idea (?) - cascading attenuation

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

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

leigh

Well-known member
Joined
Jun 4, 2004
Messages
394
Location
Portland, OR
A preamble: I've read a million words, pro and con, about "summing mixers". I've mixed in Pro Tools, I've mixed on boards. The conclusion I've reached with summing mixers is that the summing itself isn't worth it. If it helps your workflow, making it easy to insert outboard compressors, great. If it means that you use a phat toob pre to add magic sonic goo at the makeup gain stage, great. If you pleasantly clip transients with a HEDD during the A-to-D re-conversion step, then that's swell too. But the "summing at unity" itself ain't doing much in the way of mojo.

Now, I don't really care about defending the above points. It's the conclusion I've come to, but I'm not trying to press it on anyone else. Just a preamble to understand where my next thoughts are coming from...

One of the advantages of a traditional mixing console over a unity summing buss is being able to pass audio out of your DAW at full resolution (bitwise). If you're mixing on a console, you can just leave your DAW's channel faders at unity, and a "full-strength" signal gets passed out of your D/A converters. All the attenuation occurs in the analog domain. One the disadvantages of an trad console, of course, is the effort it takes to recall faders manually, or deal with the expense of automated faders.

So... here's the idea: create a summing mixer where the stereo pairs of inputs each have a different attenuation. Start at "0dB", then the next pair is at -3dB, then at -6dB, and so on. (Depending on how many channels the mixer has, of course, those steps could be larger or smaller.) Then, in the box, you could label your various converter outputs to correspond to the summing mixer's inputs. The whole idea is that, instead of using your DAW faders to attenuate signals, you are still attenuating in the analog domain, like on a traditional console. Or, at least, your DAW faders only have to attenuate 3dB at most. And, the setup is totally recallable.

I'm not proposing that running DAW faders at anything but 0dB sounds like shite. That *is* a school of thought, and it may have merit. However, that's not what drives this whole idea. Rather, the idea is to be able to run your D/A converters at near-maximum, to maintain full signal resolution.

And therein, of course, lies the question of whether this would all be worth it. Is the phenomenon of changing resolution (again, number of active bits) when you change gain part of what sucks about digital mixing? Or does it not matter if the delivery format is digital, since a reverb tail at -36dB will still only end up using 10 out of 16 bits on a CD?

Please, let's hear it. I would love some feedback on this.

cheers,
Leigh
 
Bits is bits. Attenuation in the digital domain is a simple multiply, scaling the amplitude down. Actually the multiply increases the resolution with a lot of extra bits below the nominal mix noise floor. A 16bx16b multiply generates a 32 bit result. Further, summing multiple signals in the digital domain results in more information so again more resolution. Adding two 16 bit signals results in 17 bit result. Of course practice varies from theory there is little value in carrying 32 bit data all the way to an output that can't reproduce it. Combining multiple high bit data streams at high sampling rates will take plenty of processing power. I suspect there may be differences between different hardware/software executions, just like there are differences between different analog consoles.

As an old analog designer I'd love to proffer a theoretical basis for the popularity of analog mix down but honestly I can''t think of one. That said, I wouldn't expect all digital systems to be identical, while theory suggests to me that one day they should be much more alike (and better) than analog consoles.

Of course I could be wrong...

JR
 
[quote author="JohnRoberts"]Attenuation in the digital domain is a simple multiply, scaling the amplitude down. Actually the multiply increases the resolution with a lot of extra bits below the nominal mix noise floor. A 16bx16b multiply generates a 32 bit result. Further, summing multiple signals in the digital domain results in more information so again more resolution.[/quote]

Yep, I'm pretty familiar with all that math. It's true, the initial multiplication does increase the resolution, bitwise... but since, at some point, a 32 bit+ number has to be turned into a 24 or 16 bit number, that requires chopping off those extra bits of resolution. The difference between the 32-bit and the 24-bit values is a distortion of the true signal, aka quantization error. Whether or not you apply dither, attenuation in the digital domain always involves some subtle distortion of the signal.

Unless, of course, you are multiplying by one, which, along with multiplying by zero, is a "trivial case" (mathematically speaking). And herein lies the basis of the argument for always leaving your DAW faders at unity. However, as I mentioned above, that's not what was primarily driving this idea.

Thanks for your feedback... please keep it coming!

Leigh
 
[quote author="leigh"]...The conclusion I've reached with summing mixers is that the summing itself isn't worth it. ...
Leigh[/quote]

Leigh, interesting post. I'm surprised you've concluded that analog summing mixers don't sound (insert verb-adjective descriptor here) better than mixing in the box. I haven't tested it yet, so I'm not making any statements yet. But it seems everyone is saying for a few tracks there's not much benefit, but when the track count increases there is a clear (no pun intended) benefit. Can you share more of how you came to your conclusions?

I'm philosophically predisposed to analog mixing, and I'm planning to build a passive summer/mixer. I like the idea of running the tracks straight out of the DAW without attenuation. I like the idea of being able to do some attenuation in the mixer. How do you want to implement it? Let's move ahead on how to build it.
 
I don't recall concluding anything... I have designed analog consoles with over 100 inputs to the L/R bus so I am well aware of the limitations of analog mixing.

I have less practical experience with digital, but what I do know about it suggests to me it doesn't suffer from the same weaknesses that analog does. But like I already said, there can be practical issues with current digital hardware and design decisions or tradeoffs in software between sundry units that may prevent digital from realizing that full theoretical potential.

If I believed there was an inherent advantage to analog I'd love to dust off some old designs, but this strikes me as a fashion, perhaps fueled by some shortcomings in current digital platforms, which I'm sure will continue to improve.
----
For Liegh,, i find it hard to be concerned about all that extra resolution below the noise floor. You haven't lost any resolution in the original signal, just made it smaller (quieter) with the multiply. How you deal with the extra information now below the noise floor seems a minor issue. The equivalent mechanism in analog is the low level detail that gets submerged below the noise floor during analog attenuation.

JR
 
I guess a more simple way to say that is , you get [ retain ] the
resolution on the recording not the playback ,
except for whatever digital device is recording the mix .
it dosen't get any better once it's there
 
[quote author="okgb"]I guess a more simple way to say that is , you get [ retain ] the
resolution on the recording not the playback ,
except for whatever digital device is recording the mix .
it dosen't get any better once it's there[/quote]

Huh? What? Come again? I'm sorry, this makes no sense, I don't get what you're saying. Can you say it again in clear English, please? :)

Please elaborate....
 
[quote author="tommypiper"][quote author="leigh"]...The conclusion I've reached with summing mixers is that the summing itself isn't worth it. ...
Leigh[/quote]

Leigh, interesting post. I'm surprised you've concluded that analog summing mixers don't sound (insert verb-adjective descriptor here) better than mixing in the box. I haven't tested it yet, so I'm not making any statements yet. But it seems everyone is saying for a few tracks there's not much benefit, but when the track count increases there is a clear (no pun intended) benefit. Can you share more of how you came to your conclusions?

I'm philosophically predisposed to analog mixing, and I'm planning to build a passive summer/mixer. I like the idea of running the tracks straight out of the DAW without attenuation. I like the idea of being able to do some attenuation in the mixer. How do you want to implement it? Let's move ahead on how to build it.[/quote]

Thanks for joining in here on this... my conclusions in this case are based on reading of others' experiences, and also reading about the science and theory of analog vs digital mixing.

As for the first part, others' experiences, there are many reports out there of people testing analog summing vs digital summing. Since analog summing requires makeup gain, the way this test is performed is usually something like:

1) send 16 channels out of the DAW into a passive summing buss, and then to a nice preamp or line amp for makeup gain

vs.

2) sum all your channels in your DAW, and send that stereo mix out into one pair of summing buss inputs... the signal also then passes to the makeup gain stage as before.

This way, you are strictly testing analog summing vs digital summing, and both signals pass through the same stages of passive gain reduction, and subsequent makeup gain.

Anyhow... the consensus of this kind of controlled test seems to be that there is not an appreciable difference between the two final mixes.

Now, the second part of what has informed my conclusion: the science and the theory of digital summing vs analog summing. If you're talking about a passive summing buss, I haven't yet heard a convincing argument of how that could be better than a digital summing buss. In a passive summing buss, there's none of the potential non-linearities or pleasing distortions of an analog console - no chance for transformers to saturate, or line amps to amplify with a touch of overdrive. You're just summing signals into what, as far as I can tell, is a totally linear box... and when you're starting in-the-box, with digital signals, I can't see how it makes a difference whether you:

(a) attenuate digitally, convert to analog, sum as analog, and re-convert to digital

or

(b) attenuate digitally, and sum digitally.

...which brings us to why I thought my idea of cascading attenuation might be different: it would be similar to (a), except that it would be:

(c) convert to analog, attenuate the analog, sum as analog, and re-convert to digital.


Now, this approach would make happy those who believe that digital attenuation is evil... but from everything I've read, the quantization error of digital attenuation is at most one half the level of the least-significant-bit. That's pretty damn small, and hard to believe that it would be audible... for a 24-bit signal, something less than -200 dB FS. Maybe there's a secondary effect of this that *does* matter, but I haven't heard a good explanation of it yet.

Since I don't necessarily buy the "digital attenuation is evil" argument, there's something else driving the idea of a summing buss with analog attenuation. At this point, it's an intuition to bring signals back out of "the box" (i.e. the DAW) as loud as they went into the box. And maybe there's an argument to be made too for optimizing the SNR of the DAC channels... output 16 healthy channels, making full use of the DAC's analog circuitry, rather than output 16 pre-attenuated channels.

Whew... I'd best stop here. Ideas?

Leigh
 
Hi Leigh, from the points you listed above, I'm not convinced that you're
correct at points 1 & 2 :shock:

Let me explain:

Forget the fact that 24 inputs or 2 inputs "arrive" at the passive/active summing mixer for a moment and
lets go back one stage.

It's the amount of channels of "digital" information that are "summed together" inside the DAW that
seems to cause problems and as JR said , it's the "quality of the process" within the software and
D to A convertor that can have an effect on this.

It's something that I noticed about 3 years ago and there was an article in I think "Audio Media" which
went into quite some detail technically as to "WHY" this was so.

Lots of tracks with lots of FX and Buss/Master inserts all "cranked" = undefined and "muddy" mixes
with quite a lot of detail missing and quite possibly digital distortion of parts of the signal.

ALSO - and this is very important, most users "mix" running in software, is perhaps 32 bit floating
point - depends on the software - and when "bounced down" becomes either a 24 bit audio file or
a 16 bit audio file ( ready for the CD right ? )
So what's the first thing we do ?

...... put that file right back into the DAW sofware program that we just bounced it with and listen
to it again "inside" the program's 32 bit floating point system, which is dithered down at the output
"again" at the D to A so we can listen to it out of the speakers ..... you see where this is going
right ??

Imagine the process with tape - 24 track 2 inch, down to 1/2 inch 2 track ... oh wait, we have to
put that mix BACK on the 24 track 2 inch before listening to it through all that electronics AGAIN ...

So ......

The recommended adjustment was to run the tracks / Buss / master within the software, with a LOT
more headroom than normal, say around -6 to -8 dBfs ( under digital maximum full scale )
this is VERY important with tracks that have plugins over them, in particular limiters/compressors
which can "hide" extreme levels within the software that cause degredation of the signal.

I tried this immediately on a large mix and was stunned at the difference in detail and stereo
width of the mix ..... good stuff !

I have read comments reflecting your points 1 & 2 and it seems to me that the "sound" of the
analog mixer being introduced, is what is being "enjoyed" by users , who indeed "like it" either
way - across 24 inputs or 2 inputs, this is even more enhanced by expensive things like the Neve
summing box - I'm sure that it's a damm fine "sounding" summer, but do the £'s effect the fact
that you like it ... "It must be fu**in great, it was £2,300 !! "

The fact that it is enjoyed when just listening to the stereo input does not take into account the above
problem that the "digital summing" has taken place - with all it's possible flaws !!

Let your ears decide, I would recommend VERY strongly that you try the above advice on your system
and see what a difference that makes "before" going down the summing route, unless you're using a
cheap card with cheap DA/AD convertors - which would add to the problem.

The system that I used 3 years ago was a Hammerfall 9652 card and Alesis AI3 boxes running on a
Mac with Logic 6.
Now I have Logic pro and an RME fireface 800 - the AI3's are still around on the optical cables but
are un-used at the moment.

Regards,
Marty.
 
[quote author="MartyMart"]Hi Leigh, from the points you listed above, I'm not convinced that you're correct at points 1 & 2 :shock:

Let me explain:

Forget the fact that 24 inputs or 2 inputs "arrive" at the passive/active summing mixer for a moment and lets go back one stage.

It's the amount of channels of "digital" information that are "summed together" inside the DAW that seems to cause problems and as JR said , it's the "quality of the process" within the software and D to A convertor that can have an effect on this.

It's something that I noticed about 3 years ago and there was an article in I think "Audio Media" which went into quite some detail technically as to "WHY" this was so.

Lots of tracks with lots of FX and Buss/Master inserts all "cranked" = undefined and "muddy" mixes with quite a lot of detail missing and quite possibly digital distortion of parts of the signal.[/quote]

Well, I'm not sure what you mean that I'm not "correct" at "points 1 & 2"... I was describing a listening test, and 1 and 2 were describing the two different signals to be compared during that listening test. Are you saying that you don't think that's a fair test for listening? Or that there could be some problems in the signal or 1 or 2 that wouldn't immediately be apparent in a listening test?

I would be interested to read the article you mention from "Audio Media" - does it exist online?

[quote author="MartyMart"]The recommended adjustment was to run the tracks / Buss / master within the software, with a LOT more headroom than normal, say around -6 to -8 dBfs (under digital maximum full scale) this is VERY important with tracks that have plugins over them, in particular limiters/compressors which can "hide" extreme levels within the software that cause degredation of the signal.[/quote]

When mixing digitally into a 32-bit floating point buss (as in Pro Tools LE) it's nearly impossible to clip front side of the mix buss itself. It's certainly possible, however, to clip the next stage after the mix buss, which is the conversion down to a 24 or 16 bit signal, and the subsequent D-to-A stage. However, the easy solution there is just to turn down your Master Fader. (see footnote #1)

In any case, I don't see how this would invalidate the listening test I described. I guess, to be clear, I should have added the caveat that in step 2, your digital summing is performed without clipping the digital summing buss.

cheers,
Leigh

***

Footnote #1: See Peeder's responses on this Charles Dye thread:
http://thewombforums.com/showthread.php?t=6431
 
And to explain my last point another way, here's an explanation from Bob Katz:

"Samplitude, like most native programs, computes in floating point. Therefore it doesn't matter if you attenuate in the master or in the inputs. It doesn't matter where you attenuate, anywhere else in the program EXCEPT where "the rubber meets the road", where you interface with the outside world. That is, if you are feeding an external converter or a digital processor via AES/EBU or SPDIF, then you should at least look at the levels at that point in the circuit and see that they are peaking up to, let's say, -10 dBFS, and no harm as far as I can see, up to -3 dBFS."

This quote is from a thread here, in a Dan Lavry-as-guest-moderator forum:
http://recforums.prosoundweb.com/index.php/t/3078/0/
 
I meant that there's a lot to consider before you even get to those points, as
in - what's coming out of your DAW and how good it is.
Just turning down the master doesn't fix the problems that can occur when
individual tracks/buss are being corrupted before the master output.
I didn't mean obvious "red light" overload, I mean artifacts that occur with
tracks running through plugins with a bad gain structure that can't be "seen"
or even immediately heard, just junk that ends up in your mix.
Times that by 16 or 24 and you can have a muddy and unclear mix.

I don't buy the "rubber meets the road" being the only thing that counts !!
..... who's that Katz guy anyway ?

MM. :wink:
 
[quote author="MartyMart"]I meant that there's a lot to consider before you even get to those points, as in - what's coming out of your DAW and how good it is.[/quote]

Ah, I see... clearer now...


[quote author="MartyMart"]Just turning down the master doesn't fix the problems that can occur when individual tracks/buss are being corrupted before the master output. I didn't mean obvious "red light" overload, I mean artifacts that occur with tracks running through plugins with a bad gain structure that can't be "seen" or even immediately heard, just junk that ends up in your mix.
Times that by 16 or 24 and you can have a muddy and unclear mix.[/quote]

Plug-ins can screw stuff up, yes, if they don't throughput and process in floating-point. Most native plug-ins do, however... and in that case, it's nearly impossible to overload the throughput 32-bit FP stream at any point.

[quote author="MartyMart"]I don't buy the "rubber meets the road" being the only thing that counts !!
..... who's that Katz guy anyway ?[/quote]

Is joke, yes?
 
Last year I was in a mastering session with Don Bartley, who in the last 35 years has mastered tracks for such acts as The Beatles, Metallica, Alanis Morissette and Duran Duran, and I asked him what he is noticing with the tracks that people are bringing to him to master these days, and his response was that he really notices the difference between tracks that were mixed on a console compared to tracks mixed in the box.

Not to say that one way is necessarily superior in itself, but certainly the results that many people get are different.... could be due to work practices rather than technology... or not.
 
Leigh, are you going to build something, or not?

Steve, I've been in mastering sessions with so-called very experienced engineers with similar big names attached to their resume, and these guys haven't had a clue how to listen to or how to treat my music.

In one memorable session, there was a fiddle in the tune, and after an hour of twiddling the knobs the self-satisfied mastering engnineer turns to me and says, "how do you like it?" Well, everything sounded super etched and defined, only one problem. His EQ had totally removed the fiddle from the mix. I mentioned this to him, "sounds fine, but where's the fiddle?" and he was like, "oh, there's a fiddle there?" Bloody Eeejit! I cancelled the session, and I will never go back again.

I don't trust anyone else's ears, I don't care if they win 10 Grammys a year, they are often one or two trick ponies. It's a formula. It's often BS.

We are here for DIY. Let's make our own gear, use our own ears, make OUR music!
 
Bob Katz is a recording guy with one or more books.. He has some well regarded idea's about headroom or nominal zero level below FS wrt different destination media, and mastering monitor SPL levels, IIRC and more.

I don't see a theoretical problem with digital combining, so I can't provide a logical explanation for overload when operated within nominal ranges. That said the actual execution of how all that data is being crunched together could vary between different software/hardware implementations for better or worse.

If running cooler levels opens up the mix that sounds like evidence something may be going on, or you're hearing things.. :wink:

JR
 
[quote author="tommypiper"]Leigh, are you going to build something, or not?[/quote]

Uh, oh... are you getting antsy for me to?

[quote author="tommypiper"]We are here for DIY. Let's make our own gear, use our own ears, make OUR music![/quote]

Done, done, and done. However, I haven't had much luck making my own DAW, or my own science for that matter. The technology of my DAW is one given in this case, and the physics of analog circuitry is another given. I have built gear, and I will build more gear, but I would like to understand all the existing conditions better, so I know how to improve upon my particular setup.

I do plan on building a summing box with varying attenuation. There are two other projects already on the bench, however, so I am using this time to brainstorm and ask for others' input.

Leigh
 
[quote author="leigh"]

[quote author="MartyMart"]I don't buy the "rubber meets the road" being the only thing that counts !!
..... who's that Katz guy anyway ?[/quote]

Is joke, yes?[/quote]

There's a "wink" icon in my post after the Bob Katz Q - I know who he is but
I'm not convinced that the 32 bit floating point system "fixes" a bad gain
structure at all !!
Also note in my big post, the comment about then putting the mastered mix
back through the same software program to listen to it !!

MM.
 
[quote author="MartyMart"]There's a "wink" icon in my post after the Bob Katz Q - I know who he is but I'm not convinced that the 32 bit floating point system "fixes" a bad gain structure at all !!

Also note in my big post, the comment about then putting the mastered mix
back through the same software program to listen to it !![/quote]

Yeah, I figured that's what the "winker" was for, just wanted to be sure...

For what it's worth, I listen back to mixes mostly through iTunes, which feeds the same DAC as the PT rig. I never thought before about its internal resolution - you'd figure it would just play back a 24 bit file straight from disk, but it *does* have its own volume control at the top of the window, and so it does at least one computation (multiply) to the audio. I leave the volume on max, and I hope that means it's either multiplying by 1, or bypassing the computation altogether. Anyways, not going to lose sleep over this...
 
Back
Top