The History of AI

Kotelnikov's sampling theorem

Era 1 · The foundations (1763–1942) › Information and circuits

Lineage
Machines and computation
Kind
Publication

Who

What happened

Vladimir Kotelnikov, a telecommunications engineer in Moscow, publishes among the proceedings of a Soviet congress on communications the proof that a signal can be reconstructed from samples taken at regular intervals.

Why it mattered

The sampling theorem: sampling at twice the highest frequency in a signal is enough to recover all of it. It is the basis of every conversion between continuous signals and digital data.

The full story

In the early 1930s, telephone lines and radio carried continuous signals, such as the voice. Vladimir Kotelnikov, a 24-year-old engineer trained in Moscow, asked how much information those signals really contain and how many communications can be packed into a single channel.

In a paper prepared in 1932 for a congress on the technical reconstruction of communications, and published in 1933 with the congress proceedings, Kotelnikov gave a rigorous proof of the result that bears Kotelnikov's name in Russia today: a signal that contains no frequencies above some value F is completely determined by its values at instants 1/(2F) seconds apart. In other words, if samples are taken at twice the highest frequency, nothing is lost.

The paper went almost unnoticed outside the Soviet Union. The mathematician Edmund Whittaker had reached a similar result in 1915 without radio in mind, and Claude Shannon published a version in 1949 as something known in communications practice. That is why the theorem is often cited with all three names.

International recognition came late: in 1999 Kotelnikov received the Eduard Rhein Prize for this proof. Today the sampling theorem is in every microphone, camera, and digital audio file, and it is what makes it possible to turn the continuous world into the discrete data computers work with.

Under the hood

If a signal contains no frequencies above F hertz, knowing its values every 1/(2F) seconds is enough to reconstruct it completely. The reconstruction adds up, for each sample, a function sin(x)/x centered on it and scaled by its value.

That is why digital audio is sampled 44,100 times a second: human hearing reaches about 20,000 hertz, and twice that leaves a margin. Sample more slowly and high frequencies get mistaken for lower ones, an effect called aliasing.

Leads to

  • A Mathematical Theory of Communication (July 1948)

    Independent convergence

    The sampling theorem appears in Shannon's work with no reference to Kotelnikov, whose paper was unknown outside the USSR

Sources

How to cite

“Kotelnikov's sampling theorem.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/kotelnikov-sampling-theorem

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline