Quantization

Quantization is the process of mapping a continuous range of values to a finite set of discrete levels — the operation that converts a sampled analog signal into a digital representation suitable for transmission, storage, or processing. Where sampling discretizes time, quantization discretizes amplitude. The two operations together constitute the analog-to-digital boundary that makes Digital Communication possible.

The error introduced by quantization — the difference between the original continuous value and its discrete approximation — is bounded by half the quantization step size. In uniform quantization, all steps are equal; in non-uniform quantization (as used in telephony), step sizes vary with signal level to exploit the non-uniform sensitivity of human perception. The Lloyd-Max algorithm and its information-theoretic generalizations find optimal quantizers for given source distributions.

Quantization appears far beyond signal processing. In Quantum Mechanics, quantization refers to the discretization of physical quantities like energy and angular momentum — a different concept with a shared formal structure. In machine learning, quantization-aware training reduces model precision to decrease memory and computation costs, trading a small accuracy loss for dramatic efficiency gains.

Quantization is always lossy, and the loss is irreversible. This is why it is philosophically distinct from sampling: sampling is an isomorphism under the right conditions, while quantization is a projection onto a lower-dimensional space. The information theorist who forgets this difference treats a lossy operation as lossless, and the engineer who forgets it builds systems that accumulate irrecoverable distortion.