The sampling theorem promises perfect reconstruction, but on one condition: the signal must not contain frequencies above half the sample rate. When that condition is broken, aliasing appears, a flaw that does not just lose information but creates false information, generating tones that never existed in the original sound.
The folding mechanism
Imagine sampling at 48 kHz. The Nyquist frequency is 24 kHz. If the signal contains a component at 30 kHz, the samples taken cannot tell it apart from a lower component: the system interprets it as an 18 kHz tone, that is 48 − 30.
In general, a frequency above Nyquist is "mirrored" downward around that limit. That is why it is called folding. A 40 kHz frequency, still sampling at 48 kHz, would end up at 8 kHz: right in the audible band.
The same phenomenon shows up in films when a car's wheels seem to spin backwards: the camera samples the motion too slowly and the brain reconstructs the wrong rotation.
Why aliasing sounds bad
The harmonics of a musical sound are whole-number multiples of the fundamental and sound natural to the ear. Folded frequencies, on the other hand, land in positions that bear no harmonic relationship to the note being played. The typical result is:
- a metallic harshness in the highs;
- "ghost" tones that move in the opposite direction as the note rises (a rising note can produce falling aliases);
- a grainy noise on harmonically rich sounds, such as square or sawtooth synths.
Where aliasing comes from
During analog-to-digital conversion
Every modern converter includes an anti-aliasing filter that removes frequencies above Nyquist before sampling. Today's converters also run internally at much higher rates and then reduce the rate digitally, which makes the filter very effective. In a microphone recording, conversion aliasing is therefore rarely a problem.
Inside the computer
The real risk today lies in digital processing. Some operations generate new frequencies, and if those exceed Nyquist they fold back:
- Saturation and distortion: they add harmonics; an 8 kHz signal that gets distorted produces harmonics at 16, 24, 32 kHz and beyond.
- Very fast compressors and limiters: with extremely short attack times they modulate the signal so quickly that they generate high components.
- Software synthesizers that generate ideal waveforms without band-limiting them.
- Pitch and sample rate changes performed with poor algorithms.
How to prevent it
Oversampling
The most common solution in plug-ins is oversampling: the plug-in temporarily raises the signal to a rate 2, 4 or 8 times higher, performs the non-linear processing, then filters and returns to the original rate. Harmonics that would have folded at 48 kHz stay below the new limit at 192 kHz and are removed by the filter before the return.
The price is higher CPU load and, in some cases, a little more latency. A sensible strategy is to keep oversampling low while you work and raise it at export time, if the plug-in allows it.
Good sample rate conversion
When converting from 48 to 44.1 kHz, or the other way round, use the highest-quality algorithm your DAW or a dedicated tool offers. "Fast" settings sometimes sacrifice exactly the filtering.
Band-limited synthesis
In synthesizers, good-quality oscillators generate waveforms that already lack harmonics above Nyquist. If a synth sounds harsh on high notes, check whether it has a higher-quality mode.
How to spot it with a test
A simple test lets you see a plug-in's aliasing:
- Generate a sine wave at a high frequency, for example 5 kHz, in a 44.1 or 48 kHz project.
- Insert the saturation plug-in with a strong setting.
- Look at the result with a spectrum analyzer.
The correct harmonics appear at 10, 15 and 20 kHz. If you also see peaks at positions that are not multiples of 5 kHz, those are aliases. Turning on oversampling should reduce them visibly. In my work I use this test to decide which plug-ins to keep on the brightest tracks.
How much to worry about it
For a podcast or a voice recording with light processing, aliasing is almost always negligible. It becomes relevant when you chain many saturators, work on very bright material, or produce aggressive synthetic sounds. There is no need to chase it everywhere: it is enough to know where it comes from and act at the critical points.
In short
- Aliasing occurs when frequencies above Nyquist are sampled and fold back into the audible band.
- It produces non-harmonic tones, harshness and ghost notes that move unnaturally.
- Modern converters prevent it at the input; the main risk lies in non-linear plug-ins.
- Oversampling and high-quality conversion algorithms are the main defenses.
- A spectrum analyzer and a test sine wave are enough to check for it.
