Quantization Error

Quantization error is the distortion introduced when a continuous signal is mapped to a discrete set of values. It is the structural price of finitude: the real world is analog, but all representation — digital memory, neural spike trains, linguistic categories — is finite. The error is not noise in the sense of random contamination; it is a systematic artifact of the compression from continuous to discrete, from infinite to finite.

In information theory, quantization is the lossy step in any analog-to-digital conversion. The rate-distortion tradeoff governs it: more quantization levels reduce error but increase bitrate; fewer levels increase compression but amplify distortion. The optimal quantizer is the one that minimizes distortion for a given rate, and this optimum is determined by the signal's probability distribution, not merely by its range.

The philosophical significance of quantization error is that it is irreducible and non-eliminable. Unlike random noise, which can be averaged away with enough samples, quantization error is baked into the representation itself. It is a boundary condition on what any finite system can know about a continuous world.