--- title: "Nyquist Frequency in Audio: Sampling, Aliasing, and DSP | Gawergy Audio" description: "Understand why Nyquist is half the sample rate, how aliasing occurs, what real filters do near the boundary, and why nonlinear plug-ins can generate new aliases." canonical_url: "https://gawergy.com/learn/nyquist-frequency" md_url: "https://gawergy.com/learn/nyquist-frequency.md" last_updated: "2026-09-23" date_published: "2026-09-23" --- # What Is the Nyquist Frequency? The Nyquist frequency is half a system's sample rate. It marks the upper boundary of the frequency band that a properly sampled, band-limited signal can represent uniquely under the sampling theorem. Frequencies above that boundary can reappear as false lower frequencies called aliases. The simple formula is only the start: real filters need transition space, tones exactly at the boundary are a special case, and nonlinear processing can create new out-of-band energy after the original recording was captured. ## Key takeaways - Nyquist frequency equals sample rate divided by two. - A band-limited signal below that boundary can be reconstructed under the theorem's assumptions. - Frequencies above the boundary can fold into the represented band as aliases. - Nonlinear DSP can create new high harmonics that require their own anti-alias strategy. ## Half the sample rate is a frequency boundary A system sampling at 48,000 values per second has a Nyquist frequency of 24,000 Hz. The arithmetic is simple, but its interpretation matters. MathWorks' public explanation of the sampling theorem describes the need for a sample rate more than twice the highest signal frequency to avoid ambiguity. Adobe's digital-audio guide gives common sample-rate and half-rate examples. Nyquist is a property of the sampling grid, not a loudness limit, bit-depth value, or guarantee that a real converter perfectly handles every tone below it. The theorem assumes a signal whose spectrum is restricted to a suitable band. If content above that band enters sampling untreated, multiple continuous signals can produce the same sample sequence. The resulting sampled data cannot tell which was the original. A high frequency can appear at an incorrect lower frequency, and that false component can survive later playback. This is aliasing. The [sample-rate article](/learn/sample-rate-audio) places the boundary in the broader capture and conversion system. A useful mental model is not that a waveform needs a certain number of dots to look smooth on a screen. The relevant issue is whether the signal's spectrum fits the available band and whether the reconstruction assumptions hold. A sampled sinusoid may look sparse when plotted, yet a properly band-limited sequence can still specify its continuous waveform. Visual interpolation between displayed points is not the same as a designed reconstruction filter. ## Why above-boundary tones can appear below it Sampling repeats spectral information at intervals related to the sample rate. If original content occupies frequencies beyond the permitted band, repeated spectra overlap the baseband. The overlap can create a lower-frequency component that was not present in the original analog signal. MathWorks illustrates this folding in its Nyquist explanation. A tone just above the half-rate boundary does not simply vanish; without filtering, it can be misrepresented below the boundary. Once stored that way, it may be indistinguishable from genuine low-frequency content in the file. This is different from a transient having a high sample peak or from quantization noise. Aliasing changes apparent frequency relationships. If a nonlinear processor generates harmonics that cross the boundary, the folded products may move in a way that does not follow the source's natural harmonic series. In an extreme test, a rising input tone may produce an alias that descends. That pitch-dependent behavior is one reason aliases can sound synthetic or rough. Real music is more complex than a single tone, so listening evidence can be ambiguous. Other harmonics, masking, codec artifacts, and level differences can conceal or imitate aliasing. A spectrum analyzer and a controlled sweep can help identify the mechanism, but a clean graph alone does not establish audibility in a mix. The correct technical claim is about frequency folding; perceptual significance depends on level, material, and context. ## Exactly at Nyquist is not a comfortable passband Descriptions sometimes say that a frequency at half the sample rate is 'captured' because it appears in the formula. The exact boundary is delicate. Depending on sampling phase, a sinusoid exactly at Nyquist can yield alternating samples of differing amplitude, or even zero samples if every measurement falls at a zero crossing. That sequence cannot robustly describe arbitrary phase and amplitude in the same way a comfortably in-band sinusoid can. Practical statements of the theorem therefore use a strict bandwidth relationship and filter transition space rather than treating the endpoint as an ideal working frequency. A real anti-alias filter cannot jump instantaneously from full transmission to infinite rejection at one precise frequency. Designers leave a transition region near the upper edge so that out-of-band components are sufficiently attenuated before sampling or downsampling. Reconstruction filters also must suppress images outside the intended band. The fact that a 48 kHz format has a 24 kHz mathematical boundary does not mean its usable passband is a perfect rectangle ending there. This nuance matters when comparing rates. A higher sample rate can move the boundary and give filters more transition room, but the actual result depends on filter design and the entire converter or software path. It is inaccurate to say that any content nominally below Nyquist must be reproduced identically by every device. It is equally inaccurate to treat the mathematical boundary as proof that a conventional audio rate is inherently inadequate for its intended band. ## Filtering protects capture, playback, and conversion An input anti-alias filter reduces high-frequency energy before a continuous signal is sampled. Without enough attenuation, out-of-band energy can fold into the target band. When converting an existing digital file to a lower rate, a low-pass filter must remove content that exceeds the new Nyquist boundary before samples are discarded. MathWorks' `resample` reference documents this anti-alias filtering explicitly. Simply deleting every other sample from a file is generally not an adequate rate conversion because it ignores what happens to high frequencies. On playback, reconstruction addresses a different manifestation of the sampling grid: spectral images above the baseband. The converter and its filters produce a continuous output that follows the intended band-limited waveform instead of leaving a staircase of held values as the final analog result. Input and output filters may be implemented with oversampled internal stages, but the concept is not a brand feature. It is part of making sampled audio work as predicted by the theorem. Filter design involves tradeoffs such as transition width, attenuation, delay, and possible passband effects. Two systems using the same nominal sample rate can therefore measure or sound different because their filters differ. A sample-rate setting identifies the boundary; it does not specify the complete response of a real filter. When reading a converter or resampler's performance, inspect the implementation or measurements rather than assuming the formula tells the entire story. ## Why the issue returns inside a plug-in A cleanly captured in-band signal can develop out-of-band components later. A nonlinear transfer curve, including clipping or saturation, creates harmonics from its input. Some generated harmonics may fall above the project's Nyquist frequency, even though the input itself was perfectly band-limited. If the processor produces those harmonics at the project sample rate without adequate treatment, they can fold back as aliases. This is why anti-alias design is a live concern in nonlinear DSP, not only in analog-to-digital conversion. JUCE's oversampling documentation specifically frames its oversampling class as a way to reduce aliasing from nonlinear processes. Oversampling temporarily raises the processing rate, applies the nonlinear operation in that wider numerical band, filters, and returns to the project rate. This gives generated high-frequency energy room to be removed before it folds into the final band. It does not mean a higher multiplier is automatically superior in every context. Filter response, latency, CPU load, and the processor's particular curve matter. The [oversampling comparison](/learn/upsampling-vs-oversampling) explains the difference between this internal method and ordinary file resampling. The risk depends partly on the input spectrum. If the source is already bright, newly generated harmonics may cross the boundary sooner than those from a low fundamental. That observation motivates the [high-frequency aliasing article](/learn/high-frequencies-aliasing). It does not imply that every bright sound will reveal audible aliasing or that bass can never produce it. The amount of nonlinearity, harmonic strengths, filters, and listening context all matter. ## What Nyquist does and does not tell you Nyquist does not state that audible music suddenly loses all waveform shape at a particular digital rate. It states a relationship between sampling rate and a properly band-limited signal. When those assumptions are met, the sample sequence can describe continuous behavior within its band. It also does not say that a higher rate always sounds better. If the source and destination are already appropriately band-limited and the system's filters work well, the extra representable band may offer no practical audible benefit for that material. Nor does Nyquist make bit depth irrelevant. Bit depth concerns amplitude coding and quantization error, while Nyquist concerns time sampling and spectral ambiguity. A high-bit-depth file at a low rate still has the lower frequency boundary. A high-rate file at low bit depth still has coarse amplitude coding. The two properties solve different problems. The [bit-depth explanation](/learn/bit-depth-audio) makes that distinction explicit. A single formula also cannot predict distortion audibility. An alias may exist at a measured level yet be masked in music, or it may become obvious on an exposed tone. The engineering goal is to control unwanted artifacts to an appropriate degree for the intended use. Listening and measurement both have roles, but neither should be replaced by slogans about a sample-rate number. ## Use the boundary as a diagnostic tool The Nyquist frequency is half the sample rate, and frequencies beyond it cannot be uniquely represented without suitable filtering. That simple relationship explains why capture needs an anti-alias filter, why resampling to a lower rate needs a low-pass stage, and why nonlinear plug-ins may oversample. The edge itself is not a practical passband with unlimited precision. Real systems leave room for filters and have measurable behavior that the formula alone does not describe. When diagnosing an artifact, identify the sample rate at the stage where new frequency content was created or a rate change occurred. Then ask what filtering was present before the next lower-rate boundary. That reasoning is much more useful than assuming every harsh sound is aliasing or every higher sample rate is an automatic improvement. ## About G-Clipper Pro A nonlinear clipper can generate harmonics beyond a project's Nyquist boundary. The audible result depends on its DSP and the source, not on the sample-rate label alone. ## Sources & References - [What Is the Nyquist Theorem?](https://www.mathworks.com/discovery/nyquist-theorem.html) - [resample — Resample uniform or nonuniform data to new fixed rate](https://www.mathworks.com/help/signal/ref/resample.html) - [Digitizing audio in Audition](https://helpx.adobe.com/audition/desktop/digital-audio-fundamentals/digitizing-audio.html) - [juce::dsp::Oversampling Class Template Reference](https://docs.juce.com/master/classjuce_1_1dsp_1_1Oversampling.html) - [Audio Fact Sheet — Ableton Reference Manual Version 12](https://www.ableton.com/en/manual/audio-fact-sheet/) ## Continue Reading - [What Is Sample Rate in Digital Audio?](https://gawergy.com/learn/sample-rate-audio) - [Why High Frequencies Reveal Aliasing More Easily](https://gawergy.com/learn/high-frequencies-aliasing) - [Upsampling vs Oversampling](https://gawergy.com/learn/upsampling-vs-oversampling) ## Sitemap See the full [Gawergy.com sitemap](https://gawergy.com/sitemap.md).