What makes an error-correcting code perfect?

Perfect is a technical term rather than praise. It means the code packs its codewords so tightly that every possible received word sits within one bit-flip of exactly one codeword, with none left over and no ambiguity. The smallest useful member of the family is the Hamming code that protects four data bits with three extra ones, giving a seven-bit word.

Error-correcting codes exist because channels corrupt what passes through them, and the interesting question is how little redundancy you can get away with. A binary Hamming code is a linear code with a parity-check matrix of a particular kind. Its columns consist of all the nonzero binary vectors of a given length, each used exactly once.

That construction sounds abstract and is really a filing system. Each column of the matrix corresponds to one bit position, and each row to one parity check. Because every nonzero pattern appears exactly once, the pattern of which checks fail identifies which single bit went wrong, with no two positions sharing a signature.

MathWorld’s summary of the consequence is the heart of it: Hamming codes are perfect single error-correcting codes. Perfect is a precise term. The codewords are packed so that every possible received word lies within one bit-flip of exactly one codeword, leaving nothing ambiguous and nothing unused, which is why the redundancy cannot be trimmed.

The familiar member of the family is the Hamming code that protects four information bits inside a seven-bit word, which MathWorld gives as the case where the construction parameter equals three. Two cautions about this card. It is written from a mathematical definition rather than from Hamming’s 1950 paper, which the browser could not reach, so it describes the object and not its history. And the property is single error correction: two errors in one word are outside what these codes promise.