There I was, three years of audio digital engineering to my name and wondering why nothing I bounced ever sounded as good as what I heard in the session itself. Inside my DAW, the mix would feel wide and sound impactful. Then, I’d render the track, and find myself gut-punched by the washed-out sound of the finished product. Lifeless. Harsh. Unacceptable.
What is a summing mixer anyway?
A neophyte still—and banging my head against the digital wall of fixed-point architecture, where harsh digital clipping can be an immutable problem—I called up my trusty audio-engineering store and told them about the problem.
“Have you tried a summing mixer?” the salesman asked.
“What’s that?”
“Your Digital Audio Workstation—speaking in simple terms here—handles the combination of your tracks with nothing more than math. Math is fine, but math is math: cold, analytical, vibeless. A summing mixer gives you some of the mojo of analog consoles. It combines all your tracks through an analog signal path, and the result can be warmer, more delineated, less washed out, and all-around better.”
Well, he sold me—specifically, he sold me a Dangerous Music 2-BUS LT and a Lynx Aurora 16-channel AD/DA. I never turned back, even as I read about vaunted industry pros calling summing mixers “snake oil.” The setup worked for me.
But how far has software come?
Years later, Slate Digital released the Virtual Console Collection. Steven Slate explained in his colorful way how the VCC modeled console summing entirely in software, sidestepping the need for a summing mixer.
Other developers also launched “in-the-box summing solutions”—Sonimus, Waves, and even Airwindows programmed platforms that sought to supplant your DAW’s bland, boring math.
Next came a volley of online articles and videos, much of it centered around tests that missed the point; A/B comparisons don’t really capture what it’s like to mix through a chosen signal path. This, as we’ll examine, is most of the fun.
But leaps and bounds followed in technology as well: DAWs moved from fixed to floating-point architecture (Pro Tools made the switch in 2013), then to a 64-bit mix bus, ensuring that anything you bounced would sound like your session (barring randomly generated effects like algorithmic reverb).
We also saw a small revolution in left-of-center DAWs: first Harrison Mix Buss, with its summing engine modeled on a Harrison console; most recently LUNA, UAD’s platform that lets you swap out analog-style console engines without incurring a hit to your computer’s CPU. They aren’t mainstream, but they’re talked about with increasing fervor.
With such advances in technology, the time has come to reevaluate the concept of analog summing mixers. Are they beneficial? Would you get the same results through an analog preamp at line level? Has software finally outperformed analog in this respect?
We can only answer this question with intelligent testing—not the usual A/B comparisons.
Constructing an Intelligent Test
Some summing mixers are passive, boasting no inherent amplification, and requiring a dedicated preamp to make everything work. You’ll get a lot of color from your preamp if you go this route.
Other summers sport their own active amplification, but within this category, you’ll find further classifications. Some provide panning and level controls, so you can adjust the sound in the hardware. Others have selectable transformers for coloring the tone.
I’ll be using the Dangerous 2-BUS LT for my testing. I’ve used it for years and, as Dangerous writes in its copy, the sound is transparent, yet not invisible: It will change things, but it will be hard to pin down how.
But pin it down we must.
So, I’m starting with the simplest test I can craft: a sine wave test tone, played at 1 kHz, piped through the summing mixer, and bounced digitally[1].
If my summing mixer was doing nothing at all, these two signals should null—I should get silence when the clean, digital test tone is played against the processed version with its polarity inverted.
On their own, tones sound pretty much the same.
Example 1: Test Tone, Digital
Example 2: Test Tone, Dangerous
They don’t just sound similar—they read similarly: On a frequency analyzer, there’s no discernable difference.
Yet they do not null.
Example 3: Test tone null
They cancel each other down to -38 dB, but that’s still audible.
Moving on: Some say summing mixers influence the stereo field. An arrangement piped through a summing amp is said to exhibit better spatial imaging, as well as a deeper soundstage.
Again, to the test tones: two sine waves, panned hard left and right.
Example 4: Stereo Test Tones, Digital_1
Example 5: Stereo Test Tones, Summing_1
You’ll note that again, they sound and read similarly—this time, we’ll show a vectorscope.
The nulls are different, but that we already knew. What’s interesting is the difference in the stereo field.
So, it seems the summing engine is doing something to the stereo response.
After seeing results like this, you might question if there’s something uncalibrated with my hardware setup. I certainly did.
The answer is no. I’ll spare you the rabbit hole, but the left and right channels of each AD/DA path in my rig are matched to within 0.02 dB—which is damned accurate.
Having checked our work, we can postulate our findings: Even though the summing mixer doesn’t add noticeable harmonics or phase distortion, it does alter the sound to a considerable degree.
Now, these tests only tell us how my summing mixer is different from my DAW’s internal summing engine. How does the summing mixer compare to software summing—to Slate VCC, Waves NLS, Sonimus Statson, and others?
I know you probably want to see some musical examples—and I promise, we will get to them—but the test tones tell us something very interesting.
Recall that a test tone played through my Dangerous Summing mixer did not add measurable harmonics to the signal.
The same cannot be said of software summing.
Right off the bat, software summing adds noticeable harmonic coloration to a 1 kHz test tone. Here’s Slate VMS, on the cleaner Brit 4K setting.
I’ve circled the added harmonics. Clearly, we’re getting harmonic coloration from the VCC. It seems harmonic coloration is a fixture here, even in gentler algorithms such as the Brit N.
The same holds true for Wave’s gentler console emulation in the NLS—the Nevo model.
If you’re savvy, you’ve probably noticed that I’ve increased the graph range, because the harmonics generated by these console algorithms are lower than 80 dB.
In the case of the NLS’s Nevo algorithm, we can see Waves has modeled the noise of the power supply—that’s what the big hump below 60 Hz is.
I, too, can show off a noisy power supply if I choose the right (or wrong) outlet in my studio, and if I extend the frequency range of the graph. You’ll see that in the next screenshot. You’ll also see that the summing engine mixer adds no harmonic coloration.
You can clearly see the power supply and the noise floor, but no overtones whatever.
This, to me, is the headshot. Every software summer I tested exhibits predictable harmonic behavior with test tones, no matter the frequency. Even if you pump two intervallic tones together, you’ll get the same response: Overtones generated at perfectly planned intervals.
But my analog summing mixer does not exhibit this behavior, not in the slightest. Already we have a measurable difference.
Again, the more savvy of you might say, “This isn’t a fair test: These emulations model large-format consoles. You’re comparing an SSL emulation to Dangerous hardware; you should be comparing an emulated SSL console to the exact console the company modeled.”
You’re right—I should be doing that. I’d love to be doing that. And, if you want me to be doing that, write in the Comments section to let B&H know.
Right now, however, I can only work with what I have: a summing mixer and a studio full of stereo hardware boxes, each of which I’ve not only used, but studied under the audio microscope. I’ve spent many, many hours looking at the harmonic generation patterns of hardware gear and software plug-ins in frequency analyzers, vectorscopes, and in software such as Plugin Doctor.
I feel confident when I say I can spot math at work. These plug-ins, under visual analysis, generate a purity and consistency of harmonic content, behavior that tracks predictably and seamlessly as the testing-tones are manipulated. Equivalent hardware, in my experience, will not do this.
This exactitude, prevalent in the software, would be compromised by variances among physical products, age and wear of components, quality of power, and other factors endemic to hardware.
Okay! This having been dispensed, we can move on to some music tests.
The Obligatory Shootout
This is the classic summing mixer/summing algorithm shoot-out that you’ve seen in countless articles and videos. I take a static mix, touch nothing on a track level, and route the busses to different summing engines to show off any audible differences.
It goes something like this:
Example 6: Dangerous
Example 7: Slate VCC
Example 8: Waves NLS
Now that you’ve listened to these examples, you’ll note they sound different from each other. Of course they do—they’re different algorithms. We’d have problems if they sounded the same; one company could have grounds for suing the other.
In reality, this test tells us nothing besides personal preferences for a finished product. It’s akin to taking the same mix and hiring different mastering engineers: All processing was made after the fact; no changes were made to the mix itself.
Such shoot-outs do not demonstrate the experience of mixing through the various engines—of working through the product in real time and experiencing the results instantaneously. This, I’d argue, is an important facet of any tool you use.
I can get two straightforward EQs to null pretty closely if I spend enough time on it. I can do the same with two clean compressors. But I’d never do that in a mix. I’d choose the most-suited tool for the job to help me achieve the best results in the shortest time.
I’d mix into the processor because I’d want it to influence my decisions. The reactivity is an essential part of the process—and the reactivity is lost in your typical mixing engine shoot-out.
I have devised a shoot-out that can tell us something of each engine’s reactivity. It’s quite simple: I’m going to automate our little test mix to simulate the process of mixing in real-time.
The Intelligent Shoot-out
I’ll automate this simple static mix in two ways. First, I will conjure a straight level boost of the submixes pushing into a summing engine.
The submixes will be automated by a single VCA move, so they all push as one into their summing mixers. In the analog world, we’ll be juicing the preamp of the summing mixer harder, until it hits the sweet spot, pushes into clipping, and then comes back down. The digital algorithms will give us an emulation of this effect.
Example 9: Level Automation, Dangerous
Example 10: Level Automation - VCC
Example 11: Level Automation - Waves
What do you hear? If you don’t hear much, ask yourself the question: “Which one feels as though it’s breathing more?”
To me, the Dangerous offers the push/pull effect I’d want for a mix like this. The emulations, less so.
As the Dangerous mix grows louder in level, it exhibits a density that I find characteristic of analog processors—a density hard to achieve in the box with the same amount of vivaciousness. The way the analog mix clips, too, feels different; it feels more forgiving.
As summing engines are said to influence width and spatial separation, we’ll move on to our next test: automating both the level and panning of our instrumental tracks.
Again, note any difference among the analog and digital examples.
Example 12: Dangerous Pan
Example 13: VCC Pan
Example 14: Waves NLS Pan
I don’t think the differences here are subtle—the Dangerous example offers more width and transient separation. Go 17 seconds in for the money shot: The baritone guitar and bass feel much more separated in the Dangerous example, and they also seem to land at a wider position. I was able to identify it blindly using HOFA’s 4U+ BlindTest plug-in, and I hope you can hear the difference yourself.
Conclusions
If the question underpinning this article is, “Do digital summing algorithms behave like summing mixers?” the answer is no—at least in my setup. My summing mixer does not add predictable second- and third-order harmonics.
You might say console summing algorithms model consoles, and consoles add coloration. Fine, but the behavior of that coloration isn’t nearly as predictable as what we’ve seen in the box.
For space limitations, I have not shown off results of the dangerous mixer piped into various line-level preamps—a common move in hybrid mixing setups. I can tell you, however, that the results did not read or sound like console emulations: Even in the 2020s, math is math, and electricity is electricity.
If the question is whether in-the-box summing competes with analog solutions, there is no real answer. Preference reigns above all: What do you like more, and how do you want to work? Tell us in the Comments section, below.
[1] You could argue the test would make more sense if I put the test tone through my AD/DA loop. I opted not to, because I judged that the flavor of the AD/DA is part and parcel of using a summing mixer—you wouldn’t be going through a conversion loop if you stayed entirely in the box. However, you can rest assured that digital test tones piped through the AD/DA loop did not display noticeably different results.
