A digital sawtooth contains harmonics extending far beyond the highest frequency a 44.1 or 48 kHz audio system can represent. The problem becomes easier to hear as you play higher notes because fewer legitimate harmonics fit below Nyquist.
A low saw note can carry dozens of harmonics before the ceiling matters. Move several octaves upward, and the available space collapses, so a naive oscillator folds more of its intended spectrum back into the audible range. Understanding why frequencies fold into audio aliasing explains the destination, but oscillator design explains why those unwanted frequencies appeared.
A sine oscillator avoids this specific problem because a sine contains only its fundamental. Triangle waves are friendlier because their harmonic amplitudes fall much faster than those of a saw or square, although their sharp changes in slope still need care.
The crude way to generate a saw in software is simple. Increase a phase value every sample, turn the phase into a rising ramp, then wrap it back to the start once per cycle. The waveform looks right on a screen, yet its instantaneous reset sprays energy above Nyquist.
Those ultrasonic harmonics do not vanish. They fold into valid digital frequencies and become inharmonic components whose pitch may move differently from the note you played. Filtering afterward cannot cleanly recover the original signal because the aliases have already landed among wanted harmonics.
A well-designed synth responds by removing harmonics before they cross the boundary. High notes can consequently sound smoother or thinner than the same waveform played lower, which is not automatically evidence of a weak oscillator. Sometimes the brighter version is simply carrying aliases.
Band-limited wavetables handle this by storing several versions of a waveform with different harmonic limits. The oscillator selects a cleaner table as pitch rises instead of playing one full-bandwidth table across the keyboard. Crude switching between versions can create timbre jumps during bends or glide, so practical engines often blend between neighboring tables.
This pitch-dependent behavior is easy to miss when judging a synth from one comfortable octave. Play a bright patch across four or five octaves with effects disabled and listen for the top register changing unevenly. Clean harmonic thinning should follow pitch predictably, while aliasing tends to add unrelated whistles, brittle motion, or components that appear to travel backward.
No single antialiasing method solves every oscillator trick. Pulse-width modulation moves waveform edges continuously, hard sync forces extra resets, and audio-rate modulation can change the spectrum faster than a static wavetable scheme expects. Each operation can create fresh high-frequency content even when the starting oscillator was clean.
Hard sync is especially revealing because the slave oscillator is interrupted whenever the master completes a cycle. Every forced reset can introduce another discontinuity, so antialiasing must account for the reset itself rather than merely generate a clean free-running saw. A synth may sound excellent on plain notes and become dirtier once sync is engaged.
Wavetable warping and waveshaping create a related problem. A table can be perfectly band-limited when stored, then a later transformation can reshape it and generate new harmonics above Nyquist. Antialiasing has to follow the point where new bandwidth is created, not merely exist earlier in the chain.
Oversampling can provide more room for those difficult operations, but it spends CPU to raise the temporary Nyquist limit and still needs filtering before returning to the session rate. Efficient synths often mix techniques, using band-limited tables or BLEP corrections for ordinary waveforms and heavier processing for sync, nonlinear shaping, or modes that genuinely need it.
When you compare two synths, test the oscillator before the filter, distortion, chorus, reverb, and limiter get involved. Hold a bright saw high on the keyboard, sweep the pitch slowly, then try pulse width or sync. A clean engine should lose impossible harmonics without replacing them with new tones that wander through the spectrum.
A low saw note can carry dozens of harmonics before the ceiling matters. Move several octaves upward, and the available space collapses, so a naive oscillator folds more of its intended spectrum back into the audible range. Understanding why frequencies fold into audio aliasing explains the destination, but oscillator design explains why those unwanted frequencies appeared.
Sharp waveform edges create the trouble
A mathematically ideal sawtooth resets instantly from one level to another at the end of every cycle. An ideal square or pulse wave also jumps between levels with no transition time. Those abrupt edges require an unlimited harmonic series, which a sampled system cannot preserve.A sine oscillator avoids this specific problem because a sine contains only its fundamental. Triangle waves are friendlier because their harmonic amplitudes fall much faster than those of a saw or square, although their sharp changes in slope still need care.
The crude way to generate a saw in software is simple. Increase a phase value every sample, turn the phase into a rising ramp, then wrap it back to the start once per cycle. The waveform looks right on a screen, yet its instantaneous reset sprays energy above Nyquist.
Those ultrasonic harmonics do not vanish. They fold into valid digital frequencies and become inharmonic components whose pitch may move differently from the note you played. Filtering afterward cannot cleanly recover the original signal because the aliases have already landed among wanted harmonics.
High notes leave fewer clean harmonics
The harmonic limit changes with every note. At 48 kHz, Nyquist sits at 24 kHz, so a 100 Hz saw can theoretically keep 240 harmonics below the boundary while a 4 kHz saw can keep only six. Raising the pitch changes how much of the ideal waveform a digital oscillator can reproduce.A well-designed synth responds by removing harmonics before they cross the boundary. High notes can consequently sound smoother or thinner than the same waveform played lower, which is not automatically evidence of a weak oscillator. Sometimes the brighter version is simply carrying aliases.
Band-limited wavetables handle this by storing several versions of a waveform with different harmonic limits. The oscillator selects a cleaner table as pitch rises instead of playing one full-bandwidth table across the keyboard. Crude switching between versions can create timbre jumps during bends or glide, so practical engines often blend between neighboring tables.
This pitch-dependent behavior is easy to miss when judging a synth from one comfortable octave. Play a bright patch across four or five octaves with effects disabled and listen for the top register changing unevenly. Clean harmonic thinning should follow pitch predictably, while aliasing tends to add unrelated whistles, brittle motion, or components that appear to travel backward.
Good oscillators limit bandwidth before it folds
BLEP-style oscillators attack the discontinuity directly instead of storing many complete waveforms. A correction is applied around each saw reset or pulse edge so the transition behaves more like a band-limited event. A bandlimited oscillator method based on polynomial correction achieved perceptually alias-free sawtooth emulation up to a 7.8 kHz fundamental at 44.1 kHz in its reported evaluation.No single antialiasing method solves every oscillator trick. Pulse-width modulation moves waveform edges continuously, hard sync forces extra resets, and audio-rate modulation can change the spectrum faster than a static wavetable scheme expects. Each operation can create fresh high-frequency content even when the starting oscillator was clean.
Hard sync is especially revealing because the slave oscillator is interrupted whenever the master completes a cycle. Every forced reset can introduce another discontinuity, so antialiasing must account for the reset itself rather than merely generate a clean free-running saw. A synth may sound excellent on plain notes and become dirtier once sync is engaged.
Wavetable warping and waveshaping create a related problem. A table can be perfectly band-limited when stored, then a later transformation can reshape it and generate new harmonics above Nyquist. Antialiasing has to follow the point where new bandwidth is created, not merely exist earlier in the chain.
Oversampling can provide more room for those difficult operations, but it spends CPU to raise the temporary Nyquist limit and still needs filtering before returning to the session rate. Efficient synths often mix techniques, using band-limited tables or BLEP corrections for ordinary waveforms and heavier processing for sync, nonlinear shaping, or modes that genuinely need it.
When you compare two synths, test the oscillator before the filter, distortion, chorus, reverb, and limiter get involved. Hold a bright saw high on the keyboard, sweep the pitch slowly, then try pulse width or sync. A clean engine should lose impossible harmonics without replacing them with new tones that wander through the spectrum.
