Distortion is all about how a nonlinear transfer function reshapes a waveform around some chosen bias point. That's what decides which harmonics — even, odd, or both — appear and how they evolve through stages. This document gives you the conceptual framework for designing distortion, not just using it.
Think of an amp stage as a function y = f(x) applied sample by sample: the input waveform x(t) goes into a nonlinear curve, and the output y(t) is that same waveform but reshaped.
If f(x) is smooth, you can approximate it with a power series around the region where the signal lives:
Feed in a pure sine x(t) = sin(ωt), and each power of x generates specific harmonics via trig identities:
| Term | Expansion | Produces |
|---|---|---|
| x² | sin²(ωt) = ½ − ½cos(2ωt) | DC + 2nd harmonic |
| x³ | sin³(ωt) = ¾sin(ωt) − ¼sin(3ωt) | fundamental + 3rd harmonic |
| x⁴ | mix of DC + cos(2ωt) + cos(4ωt) | DC + 2nd + 4th |
| x⁵ | mix of sin(ωt) + sin(3ωt) + sin(5ωt) | fundamental + 3rd + 5th |
Even powers produce even harmonics (2nd, 4th, …) plus DC.
Odd powers produce odd harmonics (1st, 3rd, 5th, …).
The FFT/spectrum analyzer only decomposes the already-reshaped waveform into these
components. It doesn't create them.
Draw the horizontal axis and call it the zero line — zero volts, zero pressure. The waveform oscillates above and below this line.
Break that mirror symmetry and you inevitably get even harmonics. As soon as the positive and negative halves no longer match, representing that lopsided shape requires even-order components (2nd, 4th, …).
Bias is the DC offset you apply so that, with no audio signal present, the device sits at a chosen operating point on its transfer curve.
Changing B moves the oscillation of x(t) into different parts of the curve:
| Bias Position | What Happens | Harmonics |
|---|---|---|
| Centered | Signal swings symmetrically around the linear-ish region. Both halves hit the same part of the curve. | Mostly odd-order when pushed hard |
| Shifted | Positive and negative swings live in different regions of the curve. One side compresses or clips earlier than the other. | Both even and odd as soon as you drive into nonlinearity |
From the power series perspective, adding bias effectively injects even-power behavior into the local expansion, even if the underlying device is mathematically odd-symmetric around zero.
You can take a perfectly symmetric nonlinear element and make it produce even harmonics just by shifting the bias. The same circuit, different bias, gives a fundamentally different harmonic recipe.
This isn't about tubes magically "loving music" and transistors "hating it." It's about how typical stages are biased and what their transfer curves look like.
| Solid-State (Typical) | Tube (Typical) | |
|---|---|---|
| Curve | Tight, symmetric clipping when overdriven. Hard rails, feedback around op-amps, carefully biased transistor pairs. | Biased and coupled so the signal rides asymmetrically on the tube's S-shaped curve. Transformers and coupling caps further tilt the operating point. |
| On a scope | Clipping is abrupt and fairly symmetric about zero. The envelope stays similar until you really slam it. | One half-cycle rounds/compresses before the other, or the entire waveform sits slightly above or below the nominal zero line. |
| Harmonics | Strong odd-order content, especially higher odd harmonics → "buzzy," "grainy" | Even-order components alongside odd; the mixture plus softer onset → "warm," "thick," "singing" |
You can demonstrate this directly: same input sine, three cases — clean (linear), symmetric clip, asymmetric clip — and tie each to the harmonic story: "fundamental only," "odd only," "even + odd."
Try it in the Transfer Function Lab →
If you're planning multiple stages, think of each stage as a different combination of curve shape + bias point. Here's a four-stage mental model:
Each stage chooses two things: what curve we use, and where we bias the input on that curve. That's the entire design space for distortion. Everything else — "tube vs. solid-state," "warm vs. harsh," "vintage vs. modern" — is a consequence of those two choices repeated through a chain of stages.
Six building blocks you can mix and match in any order to design a signal chain:
| Stage Type | Curve Shape | Bias / Symmetry | Harmonics Focus | Sonic / Visual Note |
|---|---|---|---|---|
| Odd‑soft pre‑stage | Gently rounded | Centered, symmetric | Mostly low‑order odd | Adds mild edge; sine becomes "plumper" but still centered on the scope. |
| Odd‑hard clipper | Abrupt flat rails | Centered, symmetric | Strong high‑order odd | Square‑ish tops and bottoms; buzzy, aggressive, scope looks very rectangular. |
| Even+odd soft "tube‑ish" | S‑curve (tanh‑like) | Slightly off‑center bias | Mix of even + odd, softer onset | One side rounds earlier; waveform rides above/below the zero line, sounds warm/chewy. |
| Even‑rich rectifier | Half / full rectifier | Strongly asymmetric (biased) | Strong even (2nd, 4th) + some odd | One polarity suppressed; envelope "pulses" at twice the original frequency. |
| Saturating compressor | Gradual ceiling | Often slightly asymmetric | Mostly low‑order, both even/odd | Peaks gently squash; envelope flattens while oscillation still visible. |
| Post‑shaper EQ/filter | Linear (filters only) | N/A (no new nonlinearity) | No new harmonics; selective emphasis | Carves the existing spectrum; makes certain harmonics or bands read as "voice." |
You can describe a chain by combining these archetypes. For example:
Odd‑soft pre‑stage to wake up harmonics → even+odd soft tube‑ish for warmth → odd‑hard clipper for aggression → post‑shaper EQ to sculpt what's been generated.
The Dallas-Arbiter Fuzz Face is one of the simplest distortion circuits ever made — just two transistors and a handful of passive components. But it maps cleanly onto the archetype framework.
The original germanium transistors (AC128, NKT275) have a lower, softer turn-on voltage (~0.2V) compared to silicon (~0.6V). This means the transfer curve bends more gradually — the transition from linear to clipping is smoother. In archetype terms, germanium makes Q1 more "odd‑soft" and Q2's clipping onset more gradual. Silicon replacements (BC108, 2N3904) have a sharper knee, pushing the character toward "odd‑hard" with a more abrupt, buzzy clip. Same topology, different curve shape → different harmonic recipe.
The Digital Octavia lab demonstrates a more complex staging approach. Roger Mayer's Octavia chains five stages:
Each stage has a purpose in the harmonic story. The fuzz gives edge, the transformer gives body, the rectifier gives the octave, and the filter makes it playable. Remove any stage and the character changes fundamentally.
With this framework in mind:
Each stage chooses two things: what curve we use and where we bias the input on that curve. That's the entire design space for distortion. Everything else — "tube vs. solid-state," "warm vs. harsh," "vintage vs. modern" — is a consequence of those two choices repeated through a chain of stages.