Choosing a waveform is choosing a harmonic recipe — and the recipe decides what the filter has left to work with. Sawtooth, square, triangle and sine, plus pulse width, detuning and sync.
The oscillator is the first box in the chain, and the single decision it asks of you — which waveform — is more consequential than it looks. It is not a choice of tone colour to be adjusted later. It is a choice of what harmonics exist, and a filter can only remove what is already there.
An oscillator produces a waveform that repeats, continuously, at a controllable rate. That rate is the pitch — press A3 on a keyboard and the oscillator repeats its cycle 220 times a second, exactly as a piano string at that pitch would.
The shape of the cycle has nothing to do with the pitch and everything to do with the timbre. Every waveform played at 220 repetitions per second is the same note; they sound completely different from each other because of what is inside the cycle.
And what is inside the cycle can always be described the same way. Any repeating waveform is a sum of sine waves at whole-number multiples of its repetition rate — the fundamental, then the second harmonic at twice that, the third at three times, and so on. Choosing a waveform is choosing how much of each of those is present.
The fundamental and nothing else. A single frequency, no harmonics at all.
Which makes it nearly useless as a subtractive starting point — there is nothing for a filter to remove, so filtering can only make it quieter. It earns its place elsewhere: as a sub-bass, as a modulator in FM synthesis, and as the sound a resonant filter makes when it is pushed into self-oscillation.
Every harmonic, with each one's level falling in proportion to its number. The second harmonic is half the amplitude of the fundamental, the third a third, the fourth a quarter.
For a 100 Hz sawtooth that means content at 200, 300, 400, 500 Hz and onward, with nothing missing. It is the richest of the standard waveforms and therefore the default starting point for subtractive synthesis — it gives the filter the most raw material. If you are unsure which waveform to start on, it is this one.
Odd harmonics only — third, fifth, seventh — again falling in proportion to harmonic number. For a 100 Hz square: 300, 500, 700 Hz, with 200, 400 and 600 simply absent.
That gap is what you are hearing when a square sounds hollow and woody next to a sawtooth's brightness. It is the same absence that gives a clarinet its character, for the same physical reason.
Odd harmonics, like a square — but falling far more steeply, as the square of the harmonic number rather than in proportion to it.
The difference that makes is larger than it sounds. Compare the third harmonic in each: on a square wave it is one third of the fundamental, about 9.5 dB down. On a triangle it is one ninth, about 19 dB down. By the fifth harmonic a square is 14 dB down and a triangle is 28. So a triangle is very nearly a sine wave with a little edge on it, and it is the right choice when you want something soft that is not completely plain.
A square wave is a special case of a more general shape. A pulse wave spends part of its cycle high and the rest low, and the proportion is its duty cycle or pulse width. At 50 percent — equal time high and low — it is a square, and that symmetry is precisely what cancels the even harmonics.
Move the width away from 50 percent and the symmetry breaks, so the even harmonics return, and the balance tilts towards the upper end. At 25 percent the tone is brighter and reedier; at 10 percent it is thin and nasal, with very little fundamental left.
Now modulate that width slowly with an LFO and you get pulse width modulation, which is worth understanding as a distinct thing rather than as another kind of wobble. A filter sweep changes how much of each harmonic survives. PWM changes which harmonics exist, continuously, at the source. That is why PWM has a shimmering, internally moving quality that no filter movement reproduces, and why one oscillator with PWM can sound like a small ensemble.
Every oscillator has tuning controls, and there are usually three of them at different resolutions:
The reason detuning is measured in cents rather than hertz is that pitch is logarithmic: the same musical interval is a different number of hertz at every octave. A semitone at the bottom of a piano is under 2 Hz; the same semitone at the top is over 200. Cents describe the ratio, so 5 cents means the same musical amount everywhere — which is exactly why its beating does not.
Between the oscillators and the filter is a mixer, setting how much of each source goes forward. It looks like a housekeeping stage and it is not, for two reasons.
The first is arithmetic. Two oscillators at full level sum to twice the amplitude, and three sum to three times. Turn everything up and the signal reaching the filter is far hotter than one oscillator alone, which on a digital instrument may simply clip.
The second is that on many analog and analog-modelled designs, how hard you drive the filter changes its character. The same filter fed a modest signal sounds clean and fed a hot one sounds thick and slightly saturated, and that saturation is a substantial part of what people mean by analog warmth. Which makes the oscillator mixer a tone control in disguise: turning two oscillators down and the output up is not the same as leaving them up, even though the final level matches.
One small control worth knowing about: whether the oscillator resets its phase at the start of each note or free-runs continuously.
Reset means every note begins at the same point in the cycle, so the attack is identical every time — tight, consistent, and what you want for a bass or a percussive patch where the first few milliseconds carry the punch. Free-running means each note catches the waveform wherever it happens to be, so the attack varies slightly from note to note, which sounds more alive on a pad and less controlled on a kick.
With two detuned oscillators the effect is larger than it sounds, because their phase relationship at the moment of the attack determines whether they reinforce or partially cancel. Free-running, that relationship is different on every note, and the note-to-note level variation people describe as analog instability is often exactly this.
Most synths give you at least two oscillators, and the commonest thing to do with the second is to set it to the same note and detune it very slightly.
The reason it thickens the sound is beating. Two tones at slightly different frequencies drift in and out of phase with each other, and the rate at which they do is simply the difference between them.
Work an example, because it produces a practical rule most people arrive at by trial and error. Take A3 at 220 Hz, and detune the second oscillator 5 cents sharp. Cents are ratios, so that is 220 × 2^(5/1200), which is about 220.64 Hz — a difference of 0.64 Hz, so the two beat together a little more than once every one and a half seconds. Slow, rich, and exactly what you want on a bass.
Now play the same patch two octaves up, at A5. The fundamental is 880 Hz, the detuned oscillator is at about 882.5, and the difference is 2.55 Hz — four times faster.
So a fixed detune in cents produces beating that doubles in rate with every octave. A detune tuned to sound lush at the bottom of the keyboard will sound unsteady and seasick at the top. Nobody tells you this and everybody eventually notices it; the fix is to use less detune on patches that live high, and to check the top of the range before committing.
The other things a second oscillator is for:
The logic of two detuned oscillators extends, and one particular extension became so widespread it is worth naming.
Stack seven sawtooth oscillators on the same note, spread across a small range of detuning with the outer ones furthest from centre, and the result is enormously wide and thick — far more so than two. Multiple simultaneous beat rates, all different, produce a shimmer rather than a single pulsing. Instruments label this a supersaw, a unison oscillator, or simply a voice-count control on the oscillator itself.
Two controls usually accompany it. Detune sets how far the outer oscillators are spread, and everything from the previous section about cents and octaves applies. Stereo spread pans them across the field, which is what turns a wide sound into a surrounding one — and is also the first thing to check when such a patch collapses in mono, since the spread is doing a great deal of the work.
The cost is processing and headroom. Seven oscillators are seven times the signal into the filter, so the mixer point from earlier matters more here than anywhere, and a patch built this way will clip if it is treated like a single-oscillator one.
A trick that produces a sound you would never get by mixing, and it takes one sentence to explain: oscillator two is forced to restart its cycle every time oscillator one completes one.
Because the restart happens at oscillator one's rate, oscillator two now repeats at that rate too — so its pitch is locked to the master no matter what its own tuning knob says. What its own frequency now controls is how much of its waveform fits into each forced cycle, which is to say the shape of the cycle, which is to say the harmonic content.
Sweep the synced oscillator's frequency and the pitch stays rock solid while the timbre tears through a series of hard, metallic, formant-like changes. It is one of the most recognisable sounds in synthesis, and it is entirely a consequence of that one restart rule.
Not every source has to be periodic. A noise generator produces a signal with no repeating cycle and therefore no pitch, and it is a legitimate oscillator in its own right — essential for percussion, breath, wind, surf, and the attack transient of anything struck or plucked.
Two flavours are standard, and the distinction matters:
In a patch, noise is usually mixed in underneath a pitched oscillator and then shaped hard by an envelope, so it contributes an attack transient rather than a continuous hiss.
Worth knowing because it explains a difference people hear and misattribute.
A mathematically perfect sawtooth has harmonics continuing upward forever. In a digital system, any harmonic above half the sample rate cannot be represented, and instead folds back into the audible range as a frequency that has no harmonic relationship to the note being played. Worse, as you play up the keyboard those folded components move down, which is audibly wrong in a way that is hard to name but easy to notice.
Good digital oscillators go to considerable lengths to avoid this — band-limiting their output, or synthesising waveforms from harmonics chosen to stay under the limit. Cheap ones do not, which is a substantial part of why some software synths sound harsh in the top octave while others do not. It is not analog warmth versus digital coldness; it is one specific and solvable engineering problem.
A filter can only remove what the oscillator provided. Pick the waveform for what you want to be left with, not for how it sounds on its own.
Students pay for getting unstuck: ask a concept question, routing issue, DAW confusion, or mix decision tied to this lesson.
Free readers can learn from the public Q&A archive. Paid students can ask their own lesson-specific questions and get mentor replies.