Analog vs digital

One of the spoke's two founding through-lines (the other is modular-routing): the argument over continuous-voltage circuitry versus sampled, computed sound, which runs across both the instruments that make sound and the effects that process it.

The two poles

• Analog — the audio is a continuous electrical signal, shaped directly by circuit components (oscillators, filters, resistor/capacitor/transistor stages). The claimed payoff is a "warmer" / more organic character, immediate (low-latency) response, and simplicity; the cost is fixed behaviour, no recall, and (often) one job per box.
• Digital — the audio is sampled (ADC), manipulated as numbers by an algorithm (DSP), and reconstructed (DAC). The payoff is programmability, versatility, many functions in one unit, and exact preset recall; the argument against is a colder or "less organic" character and (historically) latency.
• Hybrid — an analog signal path or front end paired with digital control/effects, pitched as the best of both; increasingly the default rather than a compromise.

Where it shows up in this spoke

• Instruments. All-analog synths trade on this exact character claim — the filtopia filterbank, polyuanalog, the test-equipment purists music-from-test-equipment — against digital methods like granular-synthesis and fm-synthesis. The synthesizer page carries the analog↔digital axis as one of its defining dimensions.
• Effects. The guitar-pedal is the effects-side stage: analog for overdrive/fuzz (circuit nonlinearity is the tone), digital for delay/reverb/modulation (memory + math) analog-vs-digital-pedals. The software audio-effect-plugin is digital by construction, so it's the far-digital end of the same line.

Measured, at last — and only the sampling half

Added 2026-08-08 via the research pass. Two AES papers put numbers where this page had only the argument.

meyer-moran-adda-loop inserted a CD-standard 16-bit/44.1 kHz A/D/A round trip into high-resolution playback, double-blind, over a year, across professional monitors and electrostatic systems: undetectable by any subject on any system. digital-filter-audibility isolated the filters inside that conversion and found listeners are sensitive to them. Both peer-reviewed, opposite verdicts on adjacent questions, and both kept.

What survives the disagreement is enough to sharpen this page. The character difference does not come from sampling. Whatever people hear between an analog box and a digital one is coming from the circuit — filter topology, nonlinearity, the specific implementation — not from the fact that one of them counted the signal. The section below reached that conclusion by reasoning; it now has measurement behind it, and the "colder / less organic" line in the digital pole above should be read as a claim about designs, not about digitizing.

Added 2026-08-20. moog-ladder-nonlinear-digital locates the character precisely, in the one circuit this argument is usually conducted about. Huovilainen derives the Moog ladder from the 1965 schematic and finds the nonlinearity is a tanh at each transistor differential pair, which makes the filter's behaviour depend on how hard it is driven: the warmth is a function of input amplitude, not a coefficient anybody tuned by ear. So the "specific nonlinear character" this page credits to analog is a describable circuit property, and a digital model that includes the tanh terms is reproducing it rather than approximating a mystery.

Untested here: whether an analog filter and its DSP model are discriminable. Huovilainen publishes no bench measurement, no hardware and no listening test, so the mechanism is now sourced and the comparison still is not. That is the claim instrument makers actually make, and no source in this wiki measures it.

Added 2026-09-09. firmware-emulation is a case this page's two poles do not describe, and it arrives via VES. The question above is always some version of how closely can computation imitate a circuit — the ADC/DAC round trip, the ladder filter's tanh, the model against the hardware. Firmware emulation asks nothing of the kind. When the original instrument was already digital, running its own ROM on a simulated CPU is not an imitation of the sound but a re-execution of the process that produced it, and the fidelity question dissolves instead of being answered. The pole labels still apply to the instrument (a DX100 is digital), and the analog half of the argument is simply out of reach: emulating a CPU says nothing about a Prophet-5's voice boards, which is why the machines VES can offer are almost all digital ones.

Added 2026-08-30. sample-rate splits out of this page as its own node, because the sampling question turned out to have two halves that this page kept treating as one. Playback: nobody has demonstrated an audible gain above 44.1 kHz, and the corpus's own primary (meyer-moran-adda-loop) is a null. Production: oversampling headroom is genuinely useful, for pitch shifting, time stretching, extreme EQ and any nonlinear processor that generates harmonics above Nyquist — which is most of what this spoke's effect plugins do. So a digital process can need a high rate to be correct while the file you ship needs nothing of the sort, and the trade this page describes is between a studio decision and a distribution format rather than between two sounds. headphonesty-192khz-vs-441khz is the source, T3 and consumer-facing; the primaries above it still carry the argument.

The through-line

The recurring lesson is that the divide is less about "which sounds better" than about what each buys you — analog buys immediacy and a specific nonlinear character; digital buys flexibility, recall, and density — and that hybrids increasingly dissolve the binary. It pairs with modular-routing: together they're the two axes the spoke reads almost every instrument and effect against (how it's routed, and whether it's continuous or sampled).

The trade shows up in the supply chain too (2026-07-28)

hermetic-alchemy-lab runs on an Electrosmith Daisy — and so, per the same source, do Noise Engineering's Versio/Alia/Legio and Bastl's Kastle 2. Separate boutique brands, one commodity DSP board.

That's this axis at the business layer. In analog the circuit is the product: filtopia's five-filter parallel bank isn't something a competitor can drop into their own panel, and polyuanalog publishing its board files is a genuine disclosure. Digital makes the compute fungible, so identity moves up into the panel, the jack layout and the DSP code. Same trade the page describes for sound — analog's character is fixed in the circuit, digital's is a file — read one level out, where it decides what a company actually owns.

A third position: audible but unmeasurable (2026-09-04)

The two poles above argue about character. wadax-digital-audio-interview adds a claim of a different kind, and this page should keep it separate rather than fold it into "digital sounds colder."

The claim is that bit-identical data sounds different depending on the cable and the shape of the digital waveform delivering it, that these effects reach the analog output through buffering and regeneration, and that they are "largely invisible to traditional measurement methods." wadax sells adjustable rise time, droop and receiver sensitivity as the remedy.

Note what this does to the argument. The analog-purism claim on this page is at least the kind of thing an experiment can address, and on the sampling half it was addressed — meyer-moran-adda-loop and digital-filter-audibility both ran double-blind and reported numbers. The unmeasurable claim exits that game: it cannot be confirmed, refuted, or ranked, and the only witness is the party with a product to sell. Recorded here as a position taken in the field, not as evidence about digital audio.

One half of the same source does not have this problem. Feed-forward pre-correction — model the output stage's nonlinearity, apply the inverse before conversion — is ordinary engineering, measurable at the output, and belongs to the "digital buys you correction" side of this page rather than to the mystical one.