On a lot of guitar amps, the bass, middle and treble knobs share a single network of parts, so turning up the bass changes what the treble knob does. On a vintage-style fuzz, rolling back the guitar's own volume doesn't only make it quieter: the fuzz cleans up, as if you'd turned the pedal down. On an analog drum machine, an accented hit drives the same circuit harder than a plain one. Nobody added these as separate features. So where do they come from?
Start with a drive that wasn't built from a circuit. Before you raise Amount: besides more grit, will anything else about the guitar change?
A curve with one job per knob
Raise Amount. Does anything change besides how gritty and full the guitar gets? Does Mix start doing something different?
Electric Guitar: Andy G. Cohen, CC BY 4.0
The guitar gets thicker and brighter as the curve bends further, and that's it. Drive is a curve someone drew to sound good, and each knob does exactly the job it was given. That's a fine way to build a drive. It's also why it doesn't behave like an old pedal: in a real circuit, nobody decides what each knob does. The parts decide, all at once.
A circuit is a set of rules
You don't need to read schematics to follow this. Three parts, three rules:
- A resistor resists current. The more voltage across it, the more current flows, in proportion. (That's Ohm's law.)
- A capacitor stores charge, like a small bucket. It can't fill or empty instantly, so it resists sudden change.
- A diode lets current through in one direction once the voltage across it passes a threshold, and then lets it through very suddenly.
Put a resistor and a capacitor together and you have a filter. Feed a sudden step of voltage in through the resistor, and the capacitor behind it fills gradually, like a bucket through a narrow pipe.
A resistor and a capacitor smooth a sudden step
A sudden step is full of high frequencies; the bucket can't keep up with them, so they're smoothed away, while slow changes pass. That's a low-pass filter, the same kind of tone control you've turned on synths and amps. With these values it starts cutting around 1.6 kHz. If you like formulas, the corner is 1 / (2π × R × C): a bigger resistor or capacitor makes it darker.
Now replace the capacitor with two diodes, pointing opposite ways, from the same point to ground.
Two diodes make a clipper
Quiet signals pass straight through. Past about half a volt, the diodes start to conduct, and the harder you push, the more current they take, so the output can barely rise. That's a clipper, the heart of many overdrive and distortion pedals, a close cousin of the rounded-off curve beside Drive. The new harmonics it creates are what Distortion and saturation is about.
Here's where it gets interesting. Many pedals put a capacitor and the diodes on that same point, with a pot in front to set the drive. Play quietly and the diodes do nothing, so you hear a gentle low-pass. Play loudly and the diodes take over, shunting current past the capacitor, so the clipping and the filtering change together.
The lab below is that circuit, run part by part by Maru's circuit solver in your browser. It's silent: you watch what the solver computes. The dashed line is a curve copy: the circuit's own transfer curve, measured with each input held still, then applied to the same sine one sample at a time. It's the best a "copy the sound" curve could do, and it has no capacitor.
A clipper with a capacitor and a drive pot
With a quiet input, how far does Drive move the corner? Raise R1 and turn Drive again: does it still move it as far?
Running the circuit…
Start with the quiet sine. The solved circuit comes out well below the curve copy, about 6 dB at the fundamental: the capacitor is filtering, and the copy has none. Raise Input level and the two lines close in, to within 1 dB at 4 volts, because the diodes have taken over. Same circuit, different tone, decided by how hard you play.
Now the knobs. Watch the corner in the quiet-signal response while you turn Drive. The pot isn't only a level control in front of the clipper: its own resistance sits in the path that fills the capacitor, so it's part of the filter. From 80 percent to full, the corner jumps from about 600 Hz to about 6.7 kHz. Now raise R1 to the top, 22 kilohms, and make the same turn: the corner only moves from about 350 to about 700 Hz. Changing one part changed what another knob does. Nobody designed that interaction. The knobs interact because the parts are connected, and every part feels every other part's voltage at the same moment.
Two ways to copy a piece of hardware
Say you love an old pedal and want it in software. There are two classic routes.
Copy the sound. Measure or listen to what the pedal does, then rebuild it from familiar blocks: a filter, a clipping curve like Drive's, another filter. You tune each block until it matches. This is cheap to run and easy to control, and every knob can be given a clean, predictable job. But anything you didn't think to copy isn't there. If the real pedal's tone shifts with your playing, or its tone knob changes how its drive knob feels, the copy only has those behaviours if someone noticed and built them in.
Solve the circuit. Take the schematic, every resistor, capacitor, diode and transistor with its value, and simulate the circuit itself. In the model, nobody decides what a knob does: a pot is a real resistance in the circuit, and the simulation works out the consequences. Interacting knobs, tone that follows your playing, an accent that drives a drum voice harder, the way a fuzz responds to the guitar's volume control (if the guitar's output is in the circuit too): all of it comes with the model, because it was in the circuit from the start.
The price is computation. A designed curve is one small calculation per sample. A circuit model solves a set of equations per sample, more for every diode and transistor. It also needs an accurate schematic and good equations for each part, which is a research job before it's a programming one.
Our take: if you want a tone you can shape freely, a designed drive is often the better instrument, because every control does something clear. If you want an old circuit's character, especially the way it reacts to you, solve the circuit.
There's a third route, too: learn the sound from recordings, with no schematic at all. That's what a neural capture does, and it makes the opposite trade: quick to make from the box alone, but it only knows the settings it was played.
How a solver works
A circuit model starts with a netlist: a plain text list of every part, its value and the two points it connects. Here's the circuit the lab runs, the clipper with a capacitor across the diodes and a drive pot in front, written for Maru's circuit solver:
* Diode clipper with a drive pot and a variable R1 Vin in 0 INPUT=in P1 0 x in 100k NAME=drive VALUE=0.8 TAPER=log R1 x mid 20k NAME=r1 MIN=2.2k VALUE=0 C1 mid 0 10n D1 mid 0 IS=2.52n N=1.752 D2 0 mid IS=2.52n N=1.752 .output out mid 0
Each line is one part. C1 mid 0 10n is a 10 nanofarad capacitor between the point called mid and 0, which is ground. P1 is the drive pot, and NAME=drive makes it a control. R1 is a resistor you can change: MIN=2.2k is its value at the bottom of its travel, and turning it adds up to 20k more, so the lab's R1 runs from 2.2 to 22.2 kilohms. The two diode lines point opposite ways. The format follows SPICE, the circuit simulator engineers have used since the 1970s, so a schematic translates almost line by line.
What the solver does, and when
From the netlist, the solver does three kinds of work.
Once, before any sound. It turns the list into equations, using the rules from the first section, and sorts out which parts are simple. Resistors and capacitors are predictable: their behaviour can be folded into a fixed recipe of multiplications and additions. It also finds the circuit's resting state, so a transistor stage starts out biased the way it would be after the power comes on, instead of charging up from nothing.
Every sample. The fixed recipe gives most of the answer straight away. The diodes and transistors are the hard part, because their current depends on their own voltage in a steeply curved way, and their voltage depends on their current. The solver makes a guess (the answer from the last sample, which is usually close), checks how wrong it is, and uses the slope of the error to make a better guess. That guess-and-refine loop is Newton's method, and on a circuit like this clipper it usually settles in a few rounds. Only the curved parts go through the loop; everything else stays quick arithmetic.
When you move a knob. A pot is a real resistance inside those equations, so turning it means rebuilding them. That costs much more than one sample, so it happens once per control change rather than on every sample. In exchange, the knob is exactly as connected to everything else as it is in the hardware.
The approach is called the nodal DK method, and the solver follows its published mathematics. Maru's version works with resistors, capacitors, inductors, pots and switches, diodes, bipolar transistors and op-amps, many of the building blocks of classic pedals, filters and drum voices.
Where it sits next to physical modeling and neural amps
Three families of technique try to make software behave like a real thing, and each starts from a different description:
| Approach | Starts from | Gets for free | Needs |
|---|---|---|---|
| Physical modeling | How an object vibrates: a string, a bar, a tube | Strike position, damping, size | A good model of the object |
| Circuit modeling | The schematic of an electronic device | Interacting knobs, tone that follows your playing | The schematic and good part equations |
| Captures | Recordings of the device's input and output | Uncanny accuracy at the captured setting | The device, and a capture per setting |
A capture is a photograph of one setting; a circuit model is the blueprint. Physical modeling and circuit modeling are cousins: both write down the rules of a real thing and let the computer play them forward one sample at a time. One simulates air, wood and metal; the other, electrons in parts.
A short history
Engineers have simulated circuits on computers for decades. SPICE, written by Laurence Nagel under Donald Pederson at the University of California, Berkeley, was presented in 1973 and became the standard tool for checking a circuit before building it. But SPICE was built to be accurate, not to keep up with a player in real time.
Audio researchers spent years making circuit simulation fast enough to play, a field often called virtual analog. One influential answer came in 2010, when David Yeh, Jonathan Abel and Julius Smith published the nodal DK method: a way to turn a schematic, automatically, into equations a computer can solve every sample. In 2015, Martin Holters and Udo Zölzer generalized the method so it could describe more kinds of circuits. That's the method behind Maru's solver.
The hardware it's aimed at is everywhere in music. Roland's TR-808, released in 1980, built its drums from analog circuits: the famous bass drum is a ringing circuit struck by a short pulse. Drum synthesis shows how that recipe works with oscillators and envelopes; a circuit model works it out from the parts.
Where it breaks
The schematic isn't the unit. Real parts are made to a tolerance, they age, and two units of the same pedal can sound different. A circuit model is one ideal unit with the printed values. Matching a particular pedal means measuring it.
Every part is an approximation. A diode or transistor equation captures how the part mostly behaves, not every quirk. Maru's solver can give a diode or transistor its junction capacitance and series resistance, but the capacitance is a fixed value, while a real junction's changes with its voltage. Effects that aren't in the equations, such as noise, temperature drift, or what happens when a part is pushed past its ratings, aren't in the model either.
Faithful includes the flaws. A model of a noisy, dark, fizzy circuit is a noisy, dark, fizzy model. That's often the point; sometimes it isn't what you want.
The fizz is still digital. A circuit model clips just like a curve does, so its new harmonics can fold back as aliasing on high notes. Running the circuit at a higher internal rate, oversampling, pushes the fold away. Maru's solver can run a circuit at two, four or eight times the playback rate, which also brings its treble closer to the real circuit's; the cost grows with the factor, and the filters around it add up to about two milliseconds of delay.
Some circuits flip faster than a sample. An oscillator built to snap between two states, like the multivibrators behind many square waves, makes each jump within a single sample. A solver that works one sample at a time from the last answer can't follow that jump: it loses track at every edge, and the pitch comes out depending on the sample rate. The clipper here changes smoothly from sample to sample, so the solver follows it.
Knobs cost more to move. Each knob move rebuilds the equations, so a circuit model takes in control changes at a steady control rate, not on every sample. That suits hand-turned knobs and ordinary automation; for modulation at audio rates, a designed effect is the better tool.
Why amps are hard to copy: a clipper with a capacitor across it gives a different answer depending on what it was doing a moment ago. That's what Signals and systems calls a nonlinear system with memory, the case a single curve can't describe, and the reason an amp is harder to copy than an EQ.
Make it yours
You can tell a curve from a circuit by how it behaves, without opening it. Here's an interaction test to run on any emulation, plugin or pedal you have.
- Knob test. Set one control, sweep another and listen. Then move the first control a long way and sweep the second again. Does the second knob now do something different?
- Level test. Play the same phrase softly and then hard, or turn the input down a few dB, and compare at matched loudness. Does only the amount of grit change, or does the tone shift too?
- Clean-up test. On a drive or fuzz with a guitar, roll the guitar's volume back. Does the sound clean up, or only get quieter?
Try the first two on this article's labs first, as references. On the lab with the Drive module at the top, each knob does its own job. On the circuit lab, R1 changes how far the circuit's Drive pot moves the corner, and Input level changes the tone as well as the grit: that's what a circuit looks like. The clean-up test needs a guitar, and the circuit lab has no guitar in its netlist. Then set the Drive module to a tone you'd actually use, open it from the link on the lab and save it, naming it for its job. Beside it, write a small interaction card for the device you tested: three tests, what changed, and whether it behaves like a curve or like a circuit.
To see what Maru builds with its circuit solver, visit Circuit modeling.
Quick check
- Two drive plugins sound identical at one setting. In one, the tone darkens as you dig in harder; in the other, only the grit changes. Which is more likely to be a circuit model, and why? The first. In a circuit, the parts that clip and the parts that filter are connected, so playing harder changes both. A designed curve does only what it was given to do, unless someone deliberately built in the tone change.
- In the circuit lab, Drive is at full and the input is quiet. You raise R1 from 2.2 to 22 kilohms. Do quiet notes get darker or brighter, and does Drive behave differently afterwards? Darker: more resistance in front of the capacitor fills it more slowly, so the corner moves down, from about 6.7 kHz to about 700 Hz. And Drive changes job: with R1 small, turning Drive down from full swings the corner across several octaves; with R1 large, the same turn moves it about one. The parts share one voltage, so changing one changes what the others do.
- You want the sound of a pedal you own, at the one setting you always use, and you don't have its schematic. Circuit model or capture? A capture. A circuit model needs the schematic and part values. A capture needs only the pedal and recordings, and one setting is exactly what it's good at.
Next steps
- Compare: Neural amps and captured gear. A solver needs the schematic. What if all you have is the box? See what a model learns from recordings, and what it misses.
- Sideways: Physical modeling. The same idea for acoustic instruments: write down how an object vibrates, then strike it anywhere.
- Use it: A guitar drive pedal. Build a three-knob drive (tone, grit and echo) from designed stages, and give each knob one clear job.





















