Every digital synth has a hard frequency ceiling, and any harmonic that climbs past it comes back down as something ugly. That ceiling sits at exactly half your sample rate. Run a session at 48 kHz, and nothing above 24 kHz can exist inside it.
The energy does not vanish, though. It folds. A 30 kHz component in a 48 kHz session is 6 kHz over the line, so it reappears at 18 kHz, parked in the brightest part of your mix.
Sawtooth and pulse waves are the usual culprits, because both carry a theoretically endless stack of harmonics. Play one low, and you barely notice anything wrong. Play the same patch two octaves up, and the folded content piles in fast.
The real tell is movement. Play a rising melodic line, and listen for tones that slide downward while your melody climbs. Those descending ghost tones are the giveaway, and once you have heard them, you cannot unhear them.
Synths are not the only source, either. Saturation, hard clipping, aggressive limiting, and pitch shifting all manufacture fresh high harmonics, and bit crushing and decimation artefacts are practically made of the stuff.
BLEP and BLIT designs do this by repairing the sharp corners of a waveform, since those sudden jumps are where the wild harmonics come from. Round the edges correctly, and most of the aliasing never happens.
Wavetable engines handle it another way, by storing several versions of each table with progressively fewer harmonics in each one, then switching between them as you play higher. Skip that step, and a wavetable synth will alias just as hard as a naive sawtooth.
Differentiated polynomial waveform oscillators take another route entirely. They generate a smooth polynomial shape and then differentiate it, which suppresses the alias content while staying cheap on processing power. That efficiency is why the approach keeps showing up in small, lightweight instruments, including a twelve-dollar synth that gambles a fresh patch out of nowhere.
Push the internal rate high enough, and the folded content lands so far above human hearing that it stops mattering musically. It reduces aliasing. It does not delete it.
The cost is processor load, and it scales badly. Heavy oversampling across a dozen plugin instances will find the limit of most machines, which is exactly why so many developers leave the option switched off by default.
There is a quality question buried in there too. A weak downsampling filter can smear the top end or add ringing of its own, so a poorly built oversampler sometimes trades one artefact for another.
Your DAW settings matter less than people assume. Recording at 96 kHz raises the ceiling, but it does nothing about a plugin generating garbage internally at its own rate, and it doubles your storage for the privilege.
Price is a poor guide to any of this. Some inexpensive instruments band-limit properly across every waveform and every mode, while some costly ones alias badly in one overlooked corner of the engine, and only a slow chromatic run into the top octave will tell you which one you own.
The energy does not vanish, though. It folds. A 30 kHz component in a 48 kHz session is 6 kHz over the line, so it reappears at 18 kHz, parked in the brightest part of your mix.
Sawtooth and pulse waves are the usual culprits, because both carry a theoretically endless stack of harmonics. Play one low, and you barely notice anything wrong. Play the same patch two octaves up, and the folded content piles in fast.
Harmonics above Nyquist fold back as audible grit
Aliased tones are inharmonic, so they have no musical relationship to the note you actually played. They read as fizz, glassiness, or a faint metallic ring rather than as brightness. That is why bright synth leads turn brittle up high.The real tell is movement. Play a rising melodic line, and listen for tones that slide downward while your melody climbs. Those descending ghost tones are the giveaway, and once you have heard them, you cannot unhear them.
Synths are not the only source, either. Saturation, hard clipping, aggressive limiting, and pitch shifting all manufacture fresh high harmonics, and bit crushing and decimation artefacts are practically made of the stuff.
Band-limited oscillators kill the fizz at the source
The clean fix happens inside the oscillator, long before anything reaches your speakers. Band-limited methods build a saw or a pulse that simply refuses to generate content above the ceiling in the first place.BLEP and BLIT designs do this by repairing the sharp corners of a waveform, since those sudden jumps are where the wild harmonics come from. Round the edges correctly, and most of the aliasing never happens.
Wavetable engines handle it another way, by storing several versions of each table with progressively fewer harmonics in each one, then switching between them as you play higher. Skip that step, and a wavetable synth will alias just as hard as a naive sawtooth.
Differentiated polynomial waveform oscillators take another route entirely. They generate a smooth polynomial shape and then differentiate it, which suppresses the alias content while staying cheap on processing power. That efficiency is why the approach keeps showing up in small, lightweight instruments, including a twelve-dollar synth that gambles a fresh patch out of nowhere.
Oversampling buys clean highs at a real CPU cost
Oversampling works differently. It runs the maths at two, four, or eight times the session rate, then applies a steep low-pass filter before dropping the signal back down again.Push the internal rate high enough, and the folded content lands so far above human hearing that it stops mattering musically. It reduces aliasing. It does not delete it.
The cost is processor load, and it scales badly. Heavy oversampling across a dozen plugin instances will find the limit of most machines, which is exactly why so many developers leave the option switched off by default.
There is a quality question buried in there too. A weak downsampling filter can smear the top end or add ringing of its own, so a poorly built oversampler sometimes trades one artefact for another.
Your DAW settings matter less than people assume. Recording at 96 kHz raises the ceiling, but it does nothing about a plugin generating garbage internally at its own rate, and it doubles your storage for the privilege.
Price is a poor guide to any of this. Some inexpensive instruments band-limit properly across every waveform and every mode, while some costly ones alias badly in one overlooked corner of the engine, and only a slow chromatic run into the top octave will tell you which one you own.
