When Should You Use Oversampling?

We explain why aliasing in plug‑ins isn’t always a problem — and why oversampling isn’t always the solution.

Aliasing and oversampling are popular topics in forums, product marketing, and video tutorials. The one is an unpleasant artefact of sampling audio; the other, a process that increases the sample rate of audio and can help eliminate the former. Hot takes abound, often telling you to max out oversampling whenever your computer can handle it. As it turns out, not all oversampling is the same, you don’t always need it, and sometimes it causes problems in the mix. This article aims to help you understand when oversampling is a good idea and the different issues it may introduce.

What Is Aliasing?

Figure 1: When sampling (or downsampling), frequencies in the input signal that are above the sampling rate’s Nyquist frequency alias back towards zero, as illustrated by the red line.Aliasing is a type of distortion that is characteristic of digital audio. Aliasing occurs when you try to add or record frequencies that are too high for your working sample rate to represent. Those frequencies ‘bounce off’ or ‘fold back’ over the Nyquist frequency — the highest frequency your sample rate can represent — and head back towards zero, then bounce back up, and so on. It’s similar to watching propellers on a plane: as they get too fast for your eyes, they appear to slow down, then turn backwards. That’s aliasing!

In audio plug‑ins, aliasing occurs when the maths the plug‑in performs is non‑linear, meaning it changes the shape of the waveform and creates sonic energy higher up in the spectrum than the original sound. We call this sonic energy ‘harmonic distortion’. Harmonic distortion below Nyquist is often desirable, and may even be the purpose of the plug‑in (such as with a saturator). But, when harmonic distortion passes Nyquist, the sample rate cannot accurately represent it, and aliasing is stamped onto the audio signal permanently. When audible, aliasing sounds noisy, because it is not harmonically related to the original material.

In a single plug‑in, aliasing is not always a problem worth tackling, because it’s often too quiet to hear. Most sonic energy in audio is in the low end and midrange anyway, and usually, the higher the order of harmonics that are created, the quieter they are. So, the low end and midrange have to be distorted a lot for the harmonics to reach Nyquist, getting very quiet along the way, and while the high‑end energy is closer to Nyquist to start with, it also starts out at a very low level in the first place. To be audible, aliasing must be very bad, or your audio must contain an unusual amount of high‑end energy.

However, as you stack multiple plug‑ins in series, aliasing distortion from one plug‑in is fed into the next plug‑in, and the next. It can build up throughout the production, mixing and mastering process, like a digital version of line noise.

Oversampling is a method plug‑in developers use to alleviate aliasing.

How Does Oversampling Help?

Figure 2: Non‑linear DSP introduces harmonic distortion that can quickly climb above the Nyquist frequency. This chart shows the first two even harmonics (blue dashed lines) and the aliasing they create (red dashed lines).Oversampling is a method plug‑in developers use to alleviate aliasing. Let’s say you are working at 48kHz. At that rate, audio will begin to alias at 24kHz, the Nyquist frequency. The plug‑in applies a series of filters and techniques to accurately represent your signal at a much higher rate, say 192kHz. Nyquist is now at 96kHz, providing more frequency ‘headroom’, and making it much more likely that any sonic energy there is too quiet to hear, even in aggregate. After performing the non‑linear maths, the plug‑in uses a low‑pass filter to remove all the frequencies that are above your original sample rate’s Nyquist frequency (24kHz), and converts your signal back to 48kHz. Voila! Much less aliasing distortion.

It’s important to note that, in addition to adding harmonic distortion to a signal, non‑linear maths also adds intermodulation distortion. This type of distortion occurs at the sum and difference of each frequency present in the signal, at different amounts depending on the non‑linear maths. Given that most audio signals are complex and contain many frequencies, that’s a lot to consider. Unfortunately, oversampling does not reduce this type of...