An analog oscillator's pitch tracks temperature, and without correction, a five-octave keyboard would land roughly 464 cents out of tune. That is nearly four semitones of error from heat alone.
The fix is a temperature-compensating resistor sitting inside the exponential converter, and the ideal value works out at about 3356 parts per million per degree. Nobody can reliably buy that part.
Tempco resistors ship with tolerances near ten percent, and they shift between manufacturing batches. So two identical synths coming off the same production line never quite behave the same way.
Drop to 3000 ppm, and the error stretches to 49 cents. That is close to a quarter tone, and it is the difference between a chord sounding rich and a chord sounding plain wrong.
This is why old polysynths want a warm-up period before a session, and why every voice wanders by a slightly different amount once they are running. Each voice carries its own components, its own tolerances, and its own idea of what an A is.
The payoff is immediate. Every voice stays in tune with every other voice, because they all count down from the same source, even when the global tuning of the instrument shifts.
The loss is quieter and harder to name. Voices tied to a single clock stop beating against each other the way fully independent circuits do, which is a big part of why some players still call those machines thin.
The filter chips carry their own signature on top of that. The Rev1 and Rev2 Prophet-5 ran SSM parts, rich in character and famously temperamental, while the Rev3 switched to Curtis chips that behaved far more predictably. Rev4 owners get a switch between the two.
Software took the same route, far more cheaply. Plenty of small instruments now run an independent drift generator on every voice, nudging pitch and filter cutoff by tiny amounts that never land the same way twice, including a scrappy budget synth with drift wired into every voice.
The important word there is per-voice. Global drift moves the entire patch together, which just reads as a detuned instrument rather than a living one, and it explains why digital polysynths sound static next to old hardware.
Envelopes deserve the same treatment and rarely get it. Analog envelope timings vary between voices because the capacitors and resistors setting them are never perfectly matched, so a fast attack on one voice is not quite the fast attack on the next.
Smoothing matters as much as depth. Raw random values jumping at audio rate read as noise, so the modulation has to be filtered into a slow wander before it touches pitch.
Depth needs restraint too. A few cents of movement per voice, plus a little cutoff variation, sits at the edge of conscious hearing, and that is precisely where it does its work. Push it past roughly ten cents, and listeners stop hearing character and start hearing a fault.
The fix is a temperature-compensating resistor sitting inside the exponential converter, and the ideal value works out at about 3356 parts per million per degree. Nobody can reliably buy that part.
Tempco resistors ship with tolerances near ten percent, and they shift between manufacturing batches. So two identical synths coming off the same production line never quite behave the same way.
Temperature drift can throw a VCO badly out of tune
The numbers get specific fast. Across a 25 degree swing and five octaves, a 3300 ppm part lands about 7.7 cents flat, a 3400 ppm part about 6.1 cents sharp, and a 3500 ppm part almost 20 cents sharp.Drop to 3000 ppm, and the error stretches to 49 cents. That is close to a quarter tone, and it is the difference between a chord sounding rich and a chord sounding plain wrong.
This is why old polysynths want a warm-up period before a session, and why every voice wanders by a slightly different amount once they are running. Each voice carries its own components, its own tolerances, and its own idea of what an A is.
DCOs traded drift for rock-solid tuning stability
Roland's answer showed up in the Juno line. A digitally controlled oscillator pins pitch to a master clock, with the Juno-106 dividing an 8 MHz crystal from 1984 onward, while the earlier Juno-6 and Juno-60 ran an analog clock at 1.902 MHz.The payoff is immediate. Every voice stays in tune with every other voice, because they all count down from the same source, even when the global tuning of the instrument shifts.
The loss is quieter and harder to name. Voices tied to a single clock stop beating against each other the way fully independent circuits do, which is a big part of why some players still call those machines thin.
The filter chips carry their own signature on top of that. The Rev1 and Rev2 Prophet-5 ran SSM parts, rich in character and famously temperamental, while the Rev3 switched to Curtis chips that behaved far more predictably. Rev4 owners get a switch between the two.
Per-voice drift is what modern soft synths chase
Hardware makers now sell the instability back to you. The Vintage control on Sequential's Prophet-5 Rev4 deliberately varies the behaviour of oscillators, envelopes, amplifiers, and other components, sliding from the stable Rev4 setting down towards the temperamental Rev1 end of the range.Software took the same route, far more cheaply. Plenty of small instruments now run an independent drift generator on every voice, nudging pitch and filter cutoff by tiny amounts that never land the same way twice, including a scrappy budget synth with drift wired into every voice.
The important word there is per-voice. Global drift moves the entire patch together, which just reads as a detuned instrument rather than a living one, and it explains why digital polysynths sound static next to old hardware.
Envelopes deserve the same treatment and rarely get it. Analog envelope timings vary between voices because the capacitors and resistors setting them are never perfectly matched, so a fast attack on one voice is not quite the fast attack on the next.
Smoothing matters as much as depth. Raw random values jumping at audio rate read as noise, so the modulation has to be filtered into a slow wander before it touches pitch.
Depth needs restraint too. A few cents of movement per voice, plus a little cutoff variation, sits at the edge of conscious hearing, and that is precisely where it does its work. Push it past roughly ten cents, and listeners stop hearing character and start hearing a fault.
