Interleavers

Interleavers are permutations applied to data sequences before transmission across noisy channels. Their purpose is to transform burst errors — clusters of consecutive corrupted bits caused by fading, scratches, or interference — into pseudo-random errors that the decoder expects. By spreading a burst across multiple codewords, the interleaver ensures that no single codeword bears the full weight of the burst, allowing the error-correction code to operate within its design assumptions.

The simplest interleaver is a block interleaver: data is written into a matrix row by row and read out column by column. A burst of length equal to the matrix width is broken into single errors, each separated by the matrix height. Convolutional interleavers achieve similar spreading with less memory and delay. Turbo codes use random interleavers — pseudo-random permutations that maximize the minimum distance of the concatenated code by ensuring that low-weight input patterns are mapped to widely separated positions.

The interleaver is a structural response to a mismatch between model and reality. Codes designed for the binary symmetric channel assume independent errors. Real channels violate this assumption. Rather than redesign the code for burst channels, the interleaver reshapes the error pattern to fit the code. It is a pre-processing layer that makes the channel look more like the model.

The cost of interleaving is latency. To spread a burst of length B, the interleaver must buffer at least B symbols, introducing a delay proportional to the burst protection level. In real-time systems — voice calls, video conferencing, control systems — this delay budget is a hard constraint that limits the interleaving depth. The design of an interleaver is always a tradeoff between burst protection and latency.

Interleaving is the lie that makes the truth work. It tells the decoder that the errors are independent when they are not, and by telling this lie consistently enough, it makes the lie indistinguishable from the truth. This is not deception in the moral sense; it is the engineering principle that systems work when their assumptions are met, and that when reality violates the assumptions, the engineer's job is to change reality, not the system.