◄ WORLD V · SONNY 5DART 302 · a helldive at the net

NYQUIST–SHANNON sample fast enough and the continuous signal is perfectly recoverable from its dots

The sampling theorem is the bridge between the analog and the digital: a signal containing no frequency above B is completely determined by samples taken at rate > 2B (the Nyquist rate), and can be reconstructed exactly by summing shifted sinc pulses. Sample too slowly and different frequencies become indistinguishable — aliasing — and a high tone masquerades as a low one (the wagon-wheel effect, the moiré on a screen). The whole of digital audio, imaging, and communication rests on it. Recovery is caused by the band-limit: it is that structure — no energy above B — that makes the dots enough to rebuild the curve.

THE TECHNIQUE band-limited to B ⇒ samples above 2B reconstruct exactly; below ⇒ aliasing

A 7 Hz tone. The demo shows it reconstructs exactly above the Nyquist rate (14 Hz) but aliases to a lower frequency when undersampled: live demo


HISTORY & CREDIT WKS theorem · Nyquist = the rate

“Sampling always loses the in-between.” — not if the signal is band-limited and you sample above twice its bandwidth: then the samples are the whole story, and the continuous curve is recovered exactly. The band-limit is the guarantee. cited

1915–49 · the reconstruction theorem is Whittaker–Kotel'nikov–Shannon (WKS)E.T. Whittaker (1915, the sinc interpolation), Kotel'nikov (1933), Shannon (1949). Harry Nyquist (1928) established the rate (2B pulses through bandwidth B, hence the “Nyquist rate”) but did not state the sampling/reconstruction theorem.
the fact · band-limited to B, sample above 2B, reconstruct exactly by sinc interpolation.
now · all digital audio, imaging, radio, and data acquisition.

Above the Nyquist rate the dots are the signal; below it, frequencies collide and a tone lies about its pitch. Recovery guaranteed by the band-limit — the structure the samples ride on. Nyquist-Shannon

RECOMMEND FOR I-13 reconstruction vs aliasing at the Nyquist rate, computed

On the canonical compiler, a 7 Hz tone has Nyquist rate 14; sampled at 10 (below) it aliases to |7−10|=3 Hz; sampled above 14 it reconstructs as 7:

$ i13 run rec_nyquist.i13 # 7 Hz tone, various sample rates nyquist_rate = 14 -- must sample above 2*7 alias_under = 3 -- at fs=10 (< 14): 7 Hz masquerades as 3 Hz recovered_ok = 7 -- above the Nyquist rate: exact reconstruction
Recommend: Nyquist–Shannon is reconstruction gated by a structural precondition, computed by i13 as a threshold. It reports the Nyquist rate 14, the aliased frequency 3 when undersampled at 10, and exact recovery 7 above the rate. The recovery is caused by the band-limit: only because the signal has no energy above B are the samples sufficient to rebuild it — violate that and the reconstruction returns the wrong signal (the alias), not a degraded one. It is the continuous cousin of the batch's discrete reconstructions: the right structure makes the parts (samples) enough for the whole (the curve).