A nonlinear processor fed with two sine tones can create new components at their sums, differences, and mixed integer combinations. Aliasing can happen to some of those components later, but it is not the same distortion mechanism.
The distinction matters when you judge saturators, clippers, amp models, compressors with nonlinear stages, and virtual analog processors. A plugin can have low aliasing yet still generate plenty of intermodulation distortion, while another can produce both at once. Knowing how nonlinear harmonics become audio aliasing keeps the two problems from being lumped together.
Intermodulation needs more than one frequency at the input. Feed the same processor 1 kHz and 1.3 kHz together, and the nonlinearity can generate components at 300 Hz, 2.3 kHz, 700 Hz, 1.6 kHz, and other combinations. None needs to cross the Nyquist frequency to exist.
A useful two-tone test therefore exposes something a single sine cannot. The processor is being asked to handle frequencies simultaneously, so the nonlinear transfer function mixes them as well as generating ordinary harmonics. Two-tone nonlinear measurements are built around this property because intermodulation products reveal behavior hidden by a one-tone test.
The same distortion stage can produce harmonic distortion and intermodulation distortion. They are not separate switches inside the plugin. Change the input from one tone to several and the very same nonlinear curve now has more frequency combinations available to generate.
Music makes the distinction harder because almost every note already contains multiple partials. Run a chord, drum bus, vocal, or full mix through saturation and the processor sees a crowded spectrum rather than one laboratory tone. New products can land between legitimate harmonics, which is one reason complex material can sound rough even when a single-note test seems clean.
In-band intermodulation products are different. If 1 kHz and 1.3 kHz generate a 300 Hz difference component inside the nonlinear stage, raising the internal sample rate does not make 300 Hz disappear. It is already a valid frequency well below Nyquist.
This is where casual plugin comparisons go sideways. Switching from 1x to 8x oversampling may remove descending aliases and clean the upper spectrum while leaving low and midrange intermodulation products almost where they were. A cleaner oversampled render does not prove the underlying nonlinearity stopped creating IMD.
Some intermodulation products can themselves exceed Nyquist and alias, especially with bright material, high input frequencies, or aggressive nonlinear processing. Oversampling can reduce those folded copies too. What it cannot do is selectively remove every legitimate in-band sum and difference product without changing the processor's nonlinear behavior.
Then use two fixed tones placed comfortably below Nyquist and inspect the output around their expected sum and difference combinations. Keep the drive setting unchanged. Stable products at frequencies such as f2 minus f1 or 2f1 minus f2 point to intermodulation generated by the nonlinearity itself.
The most revealing comparison changes sample rate without changing the test frequencies. Aliases depend on the Nyquist boundary, so their landing positions can shift when the sample rate changes. Genuine in-band IMD products are set by the input frequencies and nonlinear relationship, so a 300 Hz difference product stays at 300 Hz.
Level matters in both tests. Drive a nonlinear stage harder, and its higher-order terms contribute more strongly, so harmonic and intermodulation products can both rise. Matching output gain prevents a louder render from masquerading as a more detailed one.
A plugin can therefore pass the aliasing test and still make a dense chord sound dirtier than expected. Two-tone testing shows whether the extra mess comes from frequency interaction inside the nonlinear stage, while the sweep shows whether Nyquist folding adds another layer on top.
The distinction matters when you judge saturators, clippers, amp models, compressors with nonlinear stages, and virtual analog processors. A plugin can have low aliasing yet still generate plenty of intermodulation distortion, while another can produce both at once. Knowing how nonlinear harmonics become audio aliasing keeps the two problems from being lumped together.
A single tone cannot reveal intermodulation distortion
Feed a nonlinear processor one clean sine wave and the new components are usually harmonics at whole-number multiples of the input frequency. A 1 kHz tone might produce energy at 2, 3, 4, and 5 kHz depending on the shape and symmetry of the nonlinearity. Push those harmonics above Nyquist, and some can fold back as aliases.Intermodulation needs more than one frequency at the input. Feed the same processor 1 kHz and 1.3 kHz together, and the nonlinearity can generate components at 300 Hz, 2.3 kHz, 700 Hz, 1.6 kHz, and other combinations. None needs to cross the Nyquist frequency to exist.
A useful two-tone test therefore exposes something a single sine cannot. The processor is being asked to handle frequencies simultaneously, so the nonlinear transfer function mixes them as well as generating ordinary harmonics. Two-tone nonlinear measurements are built around this property because intermodulation products reveal behavior hidden by a one-tone test.
The same distortion stage can produce harmonic distortion and intermodulation distortion. They are not separate switches inside the plugin. Change the input from one tone to several and the very same nonlinear curve now has more frequency combinations available to generate.
Music makes the distinction harder because almost every note already contains multiple partials. Run a chord, drum bus, vocal, or full mix through saturation and the processor sees a crowded spectrum rather than one laboratory tone. New products can land between legitimate harmonics, which is one reason complex material can sound rough even when a single-note test seems clean.
Oversampling fixes aliasing without erasing in-band IMD
Oversampling gives a nonlinear process a higher temporary sample rate. Harmonics and intermodulation products can then extend further upward before reaching the internal Nyquist limit, after which filtering removes content that cannot safely return to the session rate. Fewer high-frequency products fold into the audible band.In-band intermodulation products are different. If 1 kHz and 1.3 kHz generate a 300 Hz difference component inside the nonlinear stage, raising the internal sample rate does not make 300 Hz disappear. It is already a valid frequency well below Nyquist.
This is where casual plugin comparisons go sideways. Switching from 1x to 8x oversampling may remove descending aliases and clean the upper spectrum while leaving low and midrange intermodulation products almost where they were. A cleaner oversampled render does not prove the underlying nonlinearity stopped creating IMD.
Some intermodulation products can themselves exceed Nyquist and alias, especially with bright material, high input frequencies, or aggressive nonlinear processing. Oversampling can reduce those folded copies too. What it cannot do is selectively remove every legitimate in-band sum and difference product without changing the processor's nonlinear behavior.
Two tests separate the problems quickly
Start with a high sine sweep through the processor and watch for new components that rise, hit the Nyquist boundary, then reverse direction. Repeat at a higher session rate or with oversampling enabled. Components that move, vanish, or get pushed upward are strong evidence of alias-dependent behavior.Then use two fixed tones placed comfortably below Nyquist and inspect the output around their expected sum and difference combinations. Keep the drive setting unchanged. Stable products at frequencies such as f2 minus f1 or 2f1 minus f2 point to intermodulation generated by the nonlinearity itself.
The most revealing comparison changes sample rate without changing the test frequencies. Aliases depend on the Nyquist boundary, so their landing positions can shift when the sample rate changes. Genuine in-band IMD products are set by the input frequencies and nonlinear relationship, so a 300 Hz difference product stays at 300 Hz.
Level matters in both tests. Drive a nonlinear stage harder, and its higher-order terms contribute more strongly, so harmonic and intermodulation products can both rise. Matching output gain prevents a louder render from masquerading as a more detailed one.
A plugin can therefore pass the aliasing test and still make a dense chord sound dirtier than expected. Two-tone testing shows whether the extra mess comes from frequency interaction inside the nonlinear stage, while the sweep shows whether Nyquist folding adds another layer on top.
