Photo: Matheus Bertelli / Pexels
- The ceiling
- No filter, and more parameters than a panel could expose
- What imposed it
- FM architecture priced down into membrane buttons and a two-digit display
- What it produced
- A decade of factory presets, and a timbre nothing else could make
Every synthesiser that preceded it had one idea at its centre: start with a harmonically rich waveform — a sawtooth, a square, something buzzing with overtones — and subtract. A filter removed frequencies until what remained matched a shape in your head. The voltage-controlled filter was so central to synthesis that entire schools of sound had grown up around how individual implementations behaved. Moog's ladder, Roland's IR3109, the Oberheim SEM — each filter had a character, a warmth or a bite or a peculiar bloom at high resonance, and that character was part of the instrument's identity. The filter was the instrument, in any practical sense.
The Yamaha DX7, released in 1983, removed the filter entirely.
There was no sawtooth to sculpt. No resonance knob. No cutoff frequency climbing through a sequence, no slow-attack sweep that let harmonics arrive over time. In its place was a system built on FM synthesis, a method worked out by the Stanford composer and computer music researcher John Chowning in the late 1960s and formalised through the 1970s. Chowning's insight was that if you modulate the frequency of one sine wave oscillator with the output of another oscillator — rather than modulating amplitude, which audio electronics had been doing for decades in tremolo effects — the result was a waveform whose harmonic content changed dramatically as the modulation depth increased. A sine wave, fed with enough FM, could produce something bright and complex and spectacularly unlike a sine wave. You could generate clangorous inharmonic spectra or glassy bell tones or reedy woodwind textures without a single filter in the signal path. The harmonics were not revealed by subtraction; they were generated by mathematics in real time.
Chowning licensed the technology to Yamaha in 1973. The company spent a decade learning how to implement it cheaply enough in digital hardware to sell to working musicians, releasing a series of expensive, complex instruments before the DX7 brought the architecture to a price point the market would actually touch. At roughly two thousand dollars in 1983, it was affordable — not cheap, but affordable. It sold somewhere in the region of two hundred thousand units in its first three years, a number that made it one of the most commercially successful synthesisers ever manufactured.
- Subtractive synthesisbegin with a complex waveform, remove frequencies with a filter until the desired timbre remains
- FM synthesismodulate the frequency of a sine-wave oscillator with another oscillator; modulation depth controls harmonic complexity without filtering
- Operatorone sine-wave oscillator plus its amplitude envelope; the DX7 has six
- Carrierthe operator whose output is heard directly
- Modulatorthe operator that alters the carrier's frequency; never heard directly
- Algorithmthe fixed routing diagram that determines which operators modulate which; the DX7 has thirty-two
- Sidebanda frequency component generated by FM modulation, appearing at sums and differences of carrier and modulator frequencies
- Inharmonic partiala sideband that falls outside the harmonic series; responsible for bell-like and metallic timbres
What Six Operators Actually Means
The DX7's FM engine organised its synthesis around six operators. Each operator was a digitally generated sine wave oscillator paired with an amplitude envelope. The way operators connected to each other — whether operator A modulated operator B, or B and C both modulated D, or an operator fed back into itself — was called an algorithm, and the DX7 offered thirty-two fixed algorithms. Some configurations placed several operators in series, each modulating the next, building complexity in stages. Others branched, with multiple modulators feeding a single carrier whose output you actually heard. The feedback parameter on a single operator could take a clean sine all the way to bright noise.
This was not an intuitive interface. Subtractive synthesis, for all its technicality, had a physical grammar — turn this knob and the sound opens, turn that one and it darkens. FM synthesis had no such grammar. Changing the ratio between an operator's frequency and the carrier's frequency shifted the position of sidebands according to Bessel functions, a fact precisely as approachable as it sounds. Raising modulation depth increased the amplitude of those sidebands, spreading the spectrum, but the relationship between parameter and result was non-linear and could produce dramatic, unpredictable changes. There was no sweep of a single control that predictably brightened or darkened a sound the way filter cutoff did.
The front panel, a single data slider and two increment/decrement buttons serving a nested menu of parameters behind an LED readout, has become a canonical example of what happens when an interface fails to represent the underlying complexity of its instrument. To programme an FM sound from scratch was to work at a level of abstraction that most professional musicians found actively hostile. The manual was thorough. It did not help much.
What Everyone Did Instead
The factory presets saved the instrument. Or more precisely: the factory presets made the instrument ubiquitous, and ubiquity produced a decade's worth of sonic signature.
The DX7 shipped with thirty-two ROM presets that Yamaha's programmers had spent considerable time crafting. They were good. Electric Piano 1 — an immediate, slightly glassy interpretation of a Rhodes electric piano with a particular percussive attack — became one of the defining sounds of mid-1980s popular music. The bass presets were tight and punchy in a way analogue could not quite match at similar cost. The vibraphone, the marimba, the harpsichord: FM excelled at tonal percussion because the partial structures of struck metal and wood are naturally inharmonic in ways Bessel-function sidebands approximate rather well. The DX7 was better at sounding like a vibraphone than it was at sounding like a synthesiser, in the traditional sense, and that was exactly what large numbers of record producers wanted.
- 1973 — Yamaha licenses FM synthesis from Stanford / John Chowning
- 1983 — DX7 released at approximately $2,000 USD
- ~200,000 — units sold in the first three years
- 6 — operators per voice
- 32 — available algorithms
- 16 — voice polyphony
- ~49 kHz — internal sample rate
- 12-bit — DAC resolution
- 48 — envelope parameters across all six operators (eight per operator)
RAM cartridges carrying additional presets created a cottage industry. Third-party sound designers — many of whom had put in the hours to understand FM programming at a technical level most users would not — sold libraries of patches for instruments across many genres. The practical result was that a huge percentage of DX7 recordings used sounds that the performer had not programmed and, in many cases, did not fully understand mechanically. This was not laziness. It was a rational response to an interface that offered no gentle entry point. The presets sounded good. The mathematics beneath them was impenetrable to intuition.
This is worth sitting with for a moment. The DX7 was the dominant commercial synthesiser of its era and most of the people playing it were using it as a preset player. Its actual synthesis architecture was, for most users, effectively invisible — a black box that produced particular recognisable timbres on demand. The instrument's identity became its factory sounds rather than its synthesis method, which is a peculiar inversion: the technology that made those sounds possible was exactly what most players never engaged with.
Why It Sounded Like It Did
FM synthesis produces sidebands at frequencies that are sums and differences of the carrier and modulator frequencies. If those frequencies are in simple integer ratios — 1:1, 1:2, 2:3 — the sidebands fall on harmonic partials and the result is a tonally coherent timbre. Shift the ratio slightly off integer values — 1:1.41, 1:3.14 — and the sidebands become inharmonic, landing between the harmonic series, producing the metallic or bell-like quality that is instantly recognisable in DX7 electric piano and gong sounds.
The DX7 used 12-bit digital-to-analogue conversion running at approximately 49 kHz, and all six operators ran synchronously at full polyphony across sixteen simultaneous voices. This was a substantial computational task for 1983 hardware — Yamaha engineered custom chips to make it work — and the consequences of that computation were audible. The DX7 had a precision and a cleanliness that analogue synthesis did not. The tuning drift that gave analogue polysynths their warm movement was absent; every voice tracked every other voice with digital exactness. That cleanliness was new. To ears trained on analogue warmth, the DX7 sounded slightly cold, slightly hard-edged, particularly in the high-frequency shimmer of its brighter presets. It was a digital sound, and at the time that distinction was sharp.
The envelopes on each operator were also more sophisticated than most subtractive synthesisers offered. Each operator had an independent four-stage envelope with separate rate and level controls — eight parameters per operator, forty-eight envelope parameters across the whole instrument before you touched anything else. This gave FM synthesis its capacity for the extraordinarily complex timbral evolution over time that characterises good FM bell tones and the way the DX7's electric piano changes character through a sustained note. Subtractive synthesis typically applied one or two envelopes to the whole voice; FM applied independent temporal shaping to every component of the spectrum.
That is the engineering reason the DX7 electric piano sounds the way it does. The attack transient is a momentary burst of high-modulation-depth content — a splash of inharmonic brightness — that decays rapidly as envelope rates pull the modulation back down, leaving the cleaner fundamental. You cannot do that with a VCA and a filter. You need independent envelopes on the spectral components themselves.
The DX7's commercial dominance lasted roughly until the late 1980s, when sample-based synthesis — instruments that used actual recordings rather than generated waveforms — offered a more literal approach to instrumental imitation. Its architecture has persisted, revised and extended in subsequent Yamaha hardware and, later, in software, but the original instrument's particular version of the sound remains unmistakable: that hard attack, that slight metallic shimmer, that absolutely consistent tuning across all sixteen voices. It is the sound of mathematics applied to timbre, at the moment mathematics became cheap enough to sell to everyone.