Process that maps a stochastic information source to a sequence of code symbols to reduce the average number of bits required to represent the source by identifying and removing statistical or structural redundancy at the source; includes both lossless and lossy mappings and is specified with respect to a source model, an allowable distortion criterion (if any), and implementation constraints.

Principle

Representational cost can be reduced to the extent that a source exhibits predictable structure; the optimal average description length is bounded below by the source entropy (for lossless coding) or by rate–distortion limits (for lossy coding).

Demonstration

Situation: A sensor produces mostly repeated readings of the same value. Recognition: the output distribution is highly skewed. Action: design a variable-length or transform-plus-quantize code that assigns short codewords to frequent values (or sparsifies then quantizes dominant components). Consequence: fewer bits per reading on average; if lossy, some fidelity is traded for rate according to the chosen distortion criterion.

Misapplication

Treating source coding as a method for correcting channel errors (confusing compression with error correction). The semantic error is conflating representational reduction (removing redundancy) with adding redundancy for reliability; they serve opposite operational goals.

Consequence

When correctly applied, source coding reduces storage and transmission resource use but may increase computational cost, latency, and sensitivity to source-model mismatch; when lossy, it also alters signal fidelity in ways determined by the distortion metric.

Reversal

If source statistics are unknown, nonstationary, or available only one-shot, universal or online schemes may perform substantially worse than model-based coding; similarly, when transmission occurs over a noisy channel without adequate channel protection, compression gains can increase effective error rates and require joint source–channel design.

Boundary

Clearly within: lossless Huffman or Lempel–Ziv compression applied to a modeled discrete source. Boundary case: transform coding (e.g., block transform + quantization) where transform choice and quantization interact with subsequent entropy coding. Clearly outside: channel coding and modulation, whose purpose is to provide reliability and physical-medium mapping rather than representational minimization.

Semantic Tension

Compression efficiency ↔ computational complexity and latency; for lossy coding: bitrate (rate) ↔ fidelity (distortion).

Synthesis

Source coding converts predictable structure of the information source into fewer symbols; its practical value depends on how well the source model, distortion criteria, and system constraints align with the chosen coding method.