“So how do you model analog gear to create plugins?” is a question we get all the time. Analog modeling is a mysterious process to many—so let’s pull back the curtain and reveal how a physical piece of gear turns into a software plugin.
Some of the most popular plugins in the Waves catalog are those that capture the sound of classic analog gear, thanks to a process called modeling. For example, this is how we managed to capture the sound of the vintage equipment used in the Abbey Road Studios, or the much-loved SSL sound.
Yet to most people who use these plugins every day, modeling remains a mysterious combination of art and science. There’s a natural curiosity about how these plugins become reality. We’d like to share with you our philosophy and technology of modeling, de-mystify the process, and also shine some light on the human element involved in modeling—without which the technology is meaningless.
The first decision that has to made is what gear to model. That may sound simple, but it’s not. Waves strongly believes in collaboration, so decisions are based on multi-way discussions with artists, product managers, our technical team, and industry colleagues. To give credit where credit is due, Waves could not exist as it does today, nor would the company have been able to attain its status as a premier software company, without the feedback and critiques from the people who use our plugins.
As a result, many product ideas come from Waves artists—the producers and engineers we cherish and trust—who discuss their ideas with our Product department. In other instances, we have our own ideas about which models to explore, and we test those ideas with artists for their feedback.
In all cases, everything we do is possible only because of extensive collaboration. For example, we could never have brought Abbey Road’s vintage gear into the virtual world without the expertise of Abbey Road’s engineers and technical staff. They not only have an incredible collection of gear, but it’s well maintained—everything is calibrated and up to spec. That makes the software emulation capture not only the essence of the gear, but the gear as it sounded on the classic recordings of the past.
In fact, one reason why the J37 tape emulation plugin sounds completely authentic is because its biasing calibration was done according to documentation from the ‘60s and ‘70s. Biasing has a huge effect on the sound of tape, and knowing what choices were made half-a-century ago enabled us to make those same choices in the plugins. We were also fortunate that Abbey Road’s engineers were so helpful with feedback and insights about the gear, which made the modeling process accurate. Sound quality can be subjective, and Abbey Road’s personnel knew what the gear was supposed to sound like—not just what it was supposed to look like on an oscilloscope.
Modeling requires serious dedication from our collaborators, because the modeling process takes months—in extreme cases even years—and there’s an immense amount of back-and-forth involved. With the PRS Supermodels guitar amp sims, to pick just one example, it wasn’t just a case of our obtaining an amp, coming up with a plugin model, presenting it to Paul Reed Smith, and having him “bless” it. Paul was involved every step of the way, monitoring our progress, and offering exacting feedback and critique.
WATCH: How the PRS SuperModels were modeled with Paul Reed Smith:
A challenge facing all manufacturers who model analog gear is choosing the actual unit to model. There are many variations with analog products. In addition to component tolerances within the unit itself, products made in one year might have used different components compared to the same product made earlier or later in the product’s life cycle.
Often, software companies search for multiple pieces of the same unit, choose the one they happen to like best, and model it.
Although this can yield good results, Waves takes a slightly different approach.
Because of our relationship with artists, we can model the actual units they’ve used to create their signature sounds and hits. We don’t want just a unit that sounds good: we want the exact unit that sounded best to Andrew Scheps, Chris Lord-Alge or Jack Joseph Puig—the one they personally chose to let into their studios, the one they choose to use again and again, and, importantly, the one they take care to service, calibrate and maintain continuously.
This has been the case when we modeled the Fairchild compressor and PuigTec equalizers at Jack Joseph Puig’s studio; or when we modeled Chris Lord-Alge’s favorite Urei 1176LN, Teletronix LA-2A, and Urei LA-3A compressors. Our goal wasn’t to model just any units we could find, but the specific ones that Chris used on his mixes. We wanted to model units that had stood the test of time in day-in, day-out music production.
Not only that—if you’re an analog hardware user, you know that the same individual unit can feel a little different from one day to the next, for reasons that are sometimes completely mysterious. So, for example, when we modeled the Fairchild with Jack Joseph Puig, Jack insisted that we model and compare our results, not just with the specific Fairchild unit of his choice, but also with that unit on the days when it behaves at its absolute best. In this video Jack explains the process:
After choosing what to model, and deciding which artist, studio and/or manufacturer will collaborate with us during this process, the laborious journey begins of quantifying the performance of the unit being modeled.
This is insanely complicated, which is why it takes so long to model an analog unit. For example, the LA-2A is a popular compressor/limiter that seems like it would be easy to model: it has only two knobs, a trimpot, and a compress/limit switch. However, the controls interact: if you alter the high-frequency trimpot, the compression amount changes for the high frequencies. Also, the response of the electroluminescent gain control technology used in the unit itself is extremely complex: for example, although the release time is not adjustable, it changes dynamically as the signal decays.
In the early days of modeling, analog units were treated like “black boxes,” where you tried to model what came out based on what went in. Modeling had to be simpler, because plugins had to work in real time, and computers could do only so many calculations per second. This is why some older plugins couldn’t work at high sample rates, or added excessive amounts of latency, or had to be rendered before you could hear them. The computers of that time simply couldn’t keep up. Fortunately, computers have become exponentially more powerful, and can perform an ever-increasing number of real-time calculations. This allows for far more detailed modeling, although this kind of precision also increases the time required to model analog gear.
At Waves we model devices on a component basis, which requires analyzing the effect that each and every component in a piece of gear has on the sound, and how that effect can change with different control settings and operational situations.
The first step in this kind of modeling is to open up the hardware we want to model and compare it to the schematics. This is crucial, because often the hardware and schematic don’t correlate—for example, maybe someone forgot to update a component’s value, or to document a part's substitution. Furthermore, this means it’s necessary to find hardware that indeed matches the “classic” circuit, and not a variation that may sound close, but not identical.
After opening up the unit, comparing it to the schematic and to other units, and making sure we have a complete understanding of the signal flow inside the hardware, then the analysis process begins. The goal is to write mathematical equations that quantify how the components perform. Some of these are easy—like how turning up a volume control will make the level louder, or how switching to a different capacitor alters the frequency response in a passive filter. But some are much harder, because something like the phase shift changes that occur when altering multiple analog filter stages, isn’t a “component” per se. So, we need to come up with a way to translate these changes as if they were a component, and then build an equation around that.
Once we’ve translated the entire piece of hardware into solvable equations using programs like MATLAB and Pspice, we can then measure performance of the hardware and compare it to the plugin’s performance. For the hardware, this requires taking a huge matrix of measurements that reflect how each component—e.g., a transistor, tube, transformer, power supply, etc.—affects the sound, and how other components influence it.
For example, a transformer may react differently if there’s more gain in the tube stage feeding it, and the tube may react differently if there’s a transformer prior to it that’s being driven hard. A model has to take all these permutations and combinations into account. This is necessary to give not just the analog sound, but also the analog “feel” as you vary the controls. After taking extensive measurements of the model, we then compare those to the same points in the circuit to make sure they match.
No modeling is complete until there have been extensive listening tests. The final step is to analyze how the hardware compares to the plugin when doing real-time operations—which is the true test of a model. Of course, these tests involve the artist we are collaborating with, the original manufacturer, or in the case of Abbey Road gear, Abbey Road’s own engineers who use the original gear on a day-to-day basis.
Often, this testing process yields some interesting surprises.
For example, we modeled an Aphex Aural Exciter that was one of only a few tube-powered units ever made. Producer/engineer Val Garay, who used the original Aphex unit extensively, confirmed that the beta version of the plugin sounded just like the hardware—with one exception. Due to phase anomalies within the unit, there was a subtly different sound depending on whether the Aphex Exciter was used as an insert effect, or part of an Aux/Send bus. So, we added a mode that wasn’t in the original unit, but allowed obtaining the Aux/Send bus’s distinctive sound when used as an insert effect.
That wasn’t the only surprise. Modeling a unit in software allows removing unintended artifacts of physical hardware, such as noise and hum. Yet, when our testers evaluated our first analog emulation ever, the prototype SSL channel strip plugins, a few said the beta versions didn’t sound “right”—without being sure quite why, because an A/B comparison didn’t reveal any difference in sound quality. It turned out that for some people, the hum and noise was part of the sound. It was only once we added these subtle imperfections that testers performing listening tests could no longer distinguish between the hardware and the plugin. We’ve since included the option to add analog noise floor and hum (50 or 60 Hz) to our analog emulations, including the Jack Joseph Puig Fairchild and Pultec models and many more.
WATCH: Behind the SSL 4000 Collection
These are the kind of characteristics that don’t show up in a schematic. Oscilloscopes and voltmeters don’t understand them, but artists with hyper-sensitive ears do. We can create what appears to be a mathematically perfect model, only to find that sometimes a little of bit of imperfection is what gives the true analog sound. It takes artists—producers and mixing engineers at the top of their game—to hear those differences. Their assistance helps us pin down these tiny—yet ultimately, significant—variations.
In recent years, however, we have made a point to go beyond the modeling of individual pieces of gear, and model entire processes or chains, each involving multiple hardware units set up together in intricate ways. These processes give contemporary producers access to specific production techniques that existed only at a particular time and place and could not be recreated faithfully by any single piece of hardware.
An obvious example is the Abbey Road Chambers plugin, which not only models Abbey Road’s Studio 2 echo chamber—but combines it with a modeling of the intricate STEED effect, which was achieved by splitting the signal between the chamber and the control room, creating a feedback loop from Abbey Road’s REDD console, through a dedicated tape delay, via RS106 and RS127 filters, to the chamber and back. Another example is the Abbey Road Saturator, which combines saturation from overdriven EMI desks with exciter-type processing from the EMI TG12321 Compander.
In all such chain-based plugins, major effort goes towards creating a user interface that packs such a multi-faceted process—which in the real world might take long hours to set up—into an easily operated piece of software.
WATCH: How we modeled Abbey Road’s Studio 2 echo chamber and the S.T.E.E.D. effect:
For Waves, the future of modeling is clear: we will have ever-greater possibilities, and we will be able to tackle ever-more-complex projects. A task like faithfully recreating the sound of Abbey Road Studios would have been impossible only a decade ago—and future developments will take us to places that may seem unimaginable today. Of course, what we’re working on now is proprietary, so we can’t tell you exactly what’s next. But when you find out, we think you’ll be very pleased.