In signal processing, the Nyquist–Shannon sampling theorem states that a continuous-time signal bandlimited to maximum frequency f_max (no spectral energy above f_max) can be exactly reconstructed from its uniform samples taken at sampling frequency f_s provided f_s > 2 f_max (often stated as f_s ≥ 2 f_max with additional technical conditions). Reconstruction requires ideal low-pass interpolation (sinc kernel) and assumes exact bandlimitedness and noiseless, infinite-duration sampling.

Principle

Sampling a bandlimited signal at or above twice its highest frequency (the Nyquist rate) preserves all information required for perfect reconstruction via an ideal interpolation filter; failure to meet the rate causes spectral overlap (aliasing) that irreversibly corrupts the sampled representation.

Demonstration

Illustrative scenario — Single tone and aliasing: Situation: Continuous sine wave x(t)=sin(2π f0 t) with f0 < f_s/2. Recognition: Sample x(t) uniformly at rate f_s>2 f0. Action: Reconstruct using ideal sinc interpolation or low-pass filter with cutoff between f0 and f_s−f0. Consequence: Recovered signal equals original. Contrast: If f_s<2 f0, sampled sequence is indistinguishable from a lower-frequency sine (alias), so reconstruction without additional information is impossible.

Misapplication

Assuming real-world signals are strictly bandlimited and therefore that sampling at the Nyquist rate alone ensures perfect reconstruction ignores practical issues: signals are seldom perfectly bandlimited, analog anti-aliasing filters are nonideal, sampling is noisy and finite in duration, and quantization introduces error—these are category errors conflating the mathematical theorem with engineering implementation.

Consequence

Provides a fundamental lower bound on uniform sampling density needed to avoid aliasing in the ideal mathematical model; in practice it guides the design of sampling systems together with anti-aliasing filters, oversampling, and reconstruction filters but does not by itself guarantee perfect practical recovery.

Reversal

Nonuniform sampling, compressed sensing, and other structured acquisition strategies permit sub-Nyquist reconstruction when the signal obeys additional constraints (sparsity, finite innovation rate); conversely, if the signal is not bandlimited, no uniform sampling rate can ensure perfect reconstruction without extra prior information.

Boundary

Clearly within: Ideal mathematical bandlimited signals sampled uniformly with noiseless infinite precision and infinite-duration data; Boundary case: Signals effectively bandlimited after analog anti-alias filtering and with finite-duration samples—approximate reconstruction with errors; Clearly outside: Signals with nonnegligible high-frequency content (impulses, white noise) where theorem's conditions do not hold and aliasing is unavoidable without preprocessing.

Semantic Tension

Tension between the theorem's mathematical exactness (requiring strict bandlimitedness, ideal filters, and infinite precision) and engineering reality (nonideal filters, noise, finite data, quantization), which forces practical compromises like oversampling and filter design.

Synthesis

The Nyquist–Shannon theorem identifies a precise information-theoretic sampling density for ideal bandlimited signals: it is a mathematical foundation that informs practical sampling system design but must be complemented by filtering, noise management, and modeling assumptions to achieve reliable real-world reconstruction.