The theorem includes a condition
For an ideal baseband signal with no content above a maximum frequency fmax, sampling at a rate greater than twice fmax permits exact reconstruction from ideal samples. The bandwidth restriction is essential. Taking a few points from an arbitrary curve does not uniquely determine what happened between them. The theorem is about a carefully specified class of signals.
Evidence: 1
Nyquist frequency and Nyquist rate differ
For a chosen sample rate fₛ, the Nyquist frequency is fₛ/2. The minimum-rate idea associated with a chosen signal bandwidth is often called its Nyquist rate. At 48 kHz sampling, half the rate is 24 kHz. That is not a guarantee that every real converter passes all frequencies up to exactly 24 kHz without attenuation or distortion.
Evidence: 2
The boundary itself needs care
A sine wave sampled exactly twice per cycle can be sampled at all its zero crossings for a particular phase. Its nonzero oscillation would then produce all-zero samples. Using a strict below-Nyquist bandwidth and leaving a practical transition region avoids treating the exact boundary as a robust operating point. Real anti-alias filters cannot switch instantaneously from full pass to complete rejection.
Evidence: 1
Apply it to every component
A bright harmonic note may have a modest fundamental but high upper components. Checking only the fundamental is insufficient. If a synthetic note is 1,000 Hz and includes its thirtieth harmonic, that component is 30 kHz. A synthesis system must omit, filter or otherwise handle components above its allowed range. The aliasing lab makes that issue visible through a single-component example first.
Evidence: 3
State the inequality
For the ideal low-pass case used here, fₛ > 2fmax, equivalently fmax < fₛ/2. Real systems need a filter transition region and other engineering allowances.
Faithful ideal reconstruction needs a band limit and a sample rate above twice that limit.
Make the connection
When samples tell another story
Move the input toward and beyond half the selected sample rate and inspect the computed alias.
Open the experiment ↗Useful words
Nyquist frequency
Half a sample rate. Frequency content must be appropriately limited below this boundary for the standard unambiguous reconstruction argument.
See in glossary ↗Sources & further reading
These sources support the explanations and model assumptions. Recording credits are listed in the listening panels and the source register.
- The Nyquist-Shannon sampling theorem ↗Chris Donahue / Carnegie Mellon University · 2026
- Sampling in practice ↗Chris Donahue / Carnegie Mellon University · 2026
- Basic waveform shapes ↗Chris Donahue / Carnegie Mellon University · 2026