--- title: "Intermodulation Distortion in Audio: Two-Tone Interactions | Gawergy Audio" description: "Explore how simultaneous tones interact in nonlinear audio systems, why sum-and-difference products arise, and what two-tone tests can and cannot predict for music." canonical_url: "https://gawergy.com/learn/intermodulation-distortion" md_url: "https://gawergy.com/learn/intermodulation-distortion.md" last_updated: "2026-09-23" date_published: "2026-09-23" --- # What Is Intermodulation Distortion in Audio? Intermodulation distortion, or IMD, is the creation of new frequencies from the interaction of two or more input components in a nonlinear system. A tone can influence how another tone is transformed, producing sum, difference, and higher-order combinations. Two-tone measurements make this visible, but real music contains many changing components. Understanding the mechanism helps explain why a processor's one-tone harmonic spectrum cannot fully predict its behavior on a mix. ## Key takeaways - IMD requires interacting input components and a nonlinear or otherwise mixing mechanism. - Products can appear at sums, differences, and higher-order combinations of source frequencies. - Some products are inharmonic relative to either individual source tone. - Two-tone tests isolate behavior, but their result depends on test levels, frequencies, bandwidth, and the device. ## Why two tones create frequencies neither tone contained Take two clean sinusoids and add them. A purely linear gain stage scales the sum without creating new spectral lines. A nonlinear stage does something different because its response to the combined instantaneous amplitude is not the sum of its separate responses. A simple squared term illustrates the mechanism: squaring the sum of two inputs includes a cross term proportional to their product. Trigonometric relationships express that product as components at the sum and difference of their frequencies. Higher-order nonlinear terms create additional combinations. This is the foundation of intermodulation distortion. A concrete example uses 1 kHz and 1.2 kHz test tones. A second-order interaction can produce components at 200 Hz and 2.2 kHz. Third-order terms can create lines such as 800 Hz and 1.4 kHz. These figures are arithmetic examples, not recommended test settings or predictions of a particular plug-in's output levels. The transfer curve determines which terms are strong; filters and sampling limits can further change what appears. Audio Precision's public amplifier-testing note describes sum-and-difference sidebands in two-tone nonlinear tests. The [harmonic versus intermodulation comparison](/learn/harmonic-vs-intermodulation-distortion) distinguishes these combination products from integer multiples generated by a single test tone. This page goes deeper into why multiple frequencies interact and why interpreting IMD requires more than spotting extra lines on a graph. ## Harmonic and inharmonic outcomes depend on the inputs If two input tones have a simple harmonic relationship, some combination products may coincide with familiar musical frequencies. If they are unrelated, the new products may sit between pitches or below both inputs. The mathematical IMD mechanism is the same; the perceived result can differ. Describing all IMD as automatically dissonant misses the role of source frequencies and the rest of the mix. Describing it as automatically musical because it comes from analog saturation is equally weak. The processor and input together determine the spectrum. A single bass fundamental plus upper partials already supplies multiple frequencies to a nonlinear curve. That means a supposedly simple note can produce intermodulation among its own partials. Add a cymbal, vocal, or chord and the number of possible interactions grows rapidly. Not every theoretical product is strong enough to matter. Some are filtered, masked, or outside the audible band. Yet the combinatorial growth explains why a dense mix can react differently from an isolated sine test even when the same processor and apparent drive are used. The source's time behavior matters too. A drum transient may briefly change the total amplitude driving a curve, causing other simultaneous content to be modulated for a short period. A compressor or dynamic saturator adds time-dependent behavior beyond a static transfer curve. The simple polynomial example remains a useful starting point, but it is not a full model of every real processor. JUCE's waveshaping tutorial shows one type of nonlinear mapping; FabFilter's Saturn documentation illustrates how distortion designs can also include bands and dynamics. ## What a two-tone measurement reveals A controlled IMD test feeds two known tones through a device and looks for output components not present at the input. Audio Precision describes several two-tone approaches, including high-frequency pairs whose difference products appear inside the measurement band. Such tests can expose nonlinear behavior that a single high-frequency THD test might miss because its harmonics fall above the analyzer's band. That is a practical measurement advantage, not proof that one IMD standard is universally best. Interpreting the graph requires the test conditions. Input tone frequencies, relative levels, total level, sample rate, analysis bandwidth, and any filtering can change the apparent ratio. A processor may be nearly linear at one level and strongly nonlinear at another. A two-tone test at one operating point cannot certify every musical use. Comparing two devices without matching the stimulus and measurement method risks attributing a difference to design quality when it came from different test conditions. Two-tone testing is especially good at isolating causality. If the input contains only two clean spectral lines and the output contains new sum-and-difference lines, the nonlinear stage generated them. A full song is much harder to interpret because it already contains many frequencies. A controlled test gives a fingerprint of one operating state, while listening to actual music establishes whether the behavior matters in context. Both views are useful and neither substitutes entirely for the other. ## Why THD does not answer the IMD question alone THD summarizes harmonics generated from a single test tone under a specified method. It says nothing directly about interactions between separate simultaneous tones because the stimulus did not include them. In high-frequency testing, harmonics can also fall beyond the available measurement band. Audio Precision explicitly describes two-tone IMD testing as a way to investigate upper-band distortion when conventional harmonic measurements become less informative. This is why a product's low THD specification is not a complete prediction of its behavior on complex audio. That does not make THD useless. It can quantify one aspect of nonlinearity, track changes with level, and reveal individual harmonic patterns. IMD and harmonic tests are complementary. The mistake is ranking all processors by one percentage without checking conditions or application. A deliberately colored plug-in can have obvious distortion by design. A transparent interface is judged against another purpose. A numeric comparison only becomes meaningful when it is tied to the question being asked. A device can also create modulation or aliasing through mechanisms that a simple static IMD calculation does not fully capture. Digital processors with nonlinear stages may generate components above Nyquist that fold back. JUCE's oversampling documentation names that problem. A spectrum at the output then reflects both the nonlinear products and the digital rate boundary. Without that context, one might label every new line as a native IMD product even when some are aliases of higher components. ## Why dense mixes can sound rough after nonlinear processing A dense arrangement supplies many simultaneous partials. A clipper or saturator driven by their combined waveform can generate cross-products among them. Some may cluster in frequency bands where there was previously space, changing the apparent texture or masking. Transients can briefly increase contact with the nonlinear region, making interaction time-dependent even when the transfer curve itself is memoryless. This gives a plausible technical explanation for a processor sounding smooth on a solo track but crowded on a mix bus. It is not a guarantee that IMD is the sole cause of that impression. Other explanations include simple harmonic enrichment, envelope change, frequency-response changes, aliasing, or level bias. A louder processed comparison often sounds preferable or more exciting even if its distortion is greater. To isolate IMD, compare matched levels and, where practical, use controlled two-tone or multitone stimuli in addition to music. The goal is understanding, not turning the article into a prescribed chain or fixed clipping amount. The [can digital clipping sound good page](/learn/can-digital-clipping-sound-good) addresses artistic choice separately. The exact role of IMD in perception is source-dependent. Combination frequencies that align with existing musical content may blend; others may stand out. Masking changes with playback level and arrangement. A spectrum can show objective products yet not tell you how a listener will rank two versions. That gap is not a reason to ignore measurement; it is a reason to state what the measurement establishes and what still needs a listening judgment. ## Measurement and language limitations There is no single universal IMD number detached from a test. SMPTE-style, DIN-style, and high-frequency difference-tone methods use different signals and calculations. Audio Precision's documents discuss these families because each emphasizes different system behavior. If two published IMD figures do not state the method, levels, and bandwidth, comparing them as if they were one scale is unreliable. In a plug-in, project sample rate and oversampling mode can also alter the test result. The conditions belong beside the number. In audio discussion, the term 'intermodulation' sometimes gets used loosely for any harsh interaction. A disciplined use identifies at least two input components, a nonlinear system, and new combination components or a credible mechanism. A compressed or EQ'd mix may sound crowded for reasons unrelated to IMD. Conversely, a nonlinear curve can produce IMD without obvious audible roughness in a specific piece of music. Distinguishing mechanism from impression keeps the discussion useful. A further subtlety is that real processors may have memory. Hysteresis, filtering in a feedback path, envelope followers, and modulation can make output depend on history as well as instantaneous input. The simple sum-and-difference derivation still teaches the basic two-tone phenomenon, but observed products may be broadened or altered. Characterizing such a device may require more than a static curve or one pair of tones. ## Use IMD to understand interaction, not to assign taste Intermodulation distortion names new frequency components created when simultaneous inputs interact through a nonlinear system. The products can be sums, differences, or higher-order combinations, and they may be harmonically related to the source or not. Two-tone tests isolate them, while music reveals whether they matter in a realistic arrangement. A one-tone THD figure cannot fully replace that view. When evaluating a processor, identify the source components, operating level, sample rate, and test method. Then separate the objective products from the subjective question of whether their presence helps or harms the track. That approach is more precise than claiming every nonlinear stage is either automatically warm or automatically dirty. ## About G-Clipper Pro On a full mix, clipping can create combination products among simultaneous sources. A one-tone curve or THD figure alone will not predict every audible interaction. ## Sources & References - [Automotive Audio Testing - Amplifiers](https://www.audioprecision.com/fileadmin-ap/technical-library/Audio-Precision-AppNote-Automotive-Audio-Amplifier-Testing.pdf) - [Linear Time-Invariant Digital Filters — Introduction to Digital Filters](https://www.dsprelated.com/freebooks/filters/Linear_Time_Invariant_Digital_Filters.html) - [Amplifiers — Analog Devices University](https://wiki.analog.com/university/tools/pluto/users/amp) - [Add distortion through waveshaping and convolution](https://juce.com/tutorials/tutorial_dsp_convolution/) - [juce::dsp::Oversampling Class Template Reference](https://docs.juce.com/master/classjuce_1_1dsp_1_1Oversampling.html) - [FabFilter Saturn 2 Help — Overview](https://www.fabfilter.com/help/saturn/using/overview) - [Live Audio Effect Reference — Ableton Reference Manual Version 12](https://www.ableton.com/en/manual/live-audio-effect-reference/) ## Continue Reading - [Harmonic Distortion vs Intermodulation Distortion](https://gawergy.com/learn/harmonic-vs-intermodulation-distortion) - [Linear vs Nonlinear Audio Processing](https://gawergy.com/learn/linear-vs-nonlinear-audio-processing) - [Why Clipping Creates Harmonics](https://gawergy.com/learn/why-clipping-creates-harmonics) ## Sitemap See the full [Gawergy.com sitemap](https://gawergy.com/sitemap.md).