The encyclopedia · Engineering & Operations · Technical decision · 1993
Berrou's turbo code reached near the Shannon limit by letting two decoders argue
Turbo coding interleaves two weak codes and passes them each other's soft confidence, recovering data a stronger single code could not.
France Telecom / Telcom Bretagne
the move
Convolutional codes were good but sat far above the Shannon limit. Berrou's team at Telecom Bretagne asked whether a block of two codes, applied to the same data via an interleaver, could do better than either alone.
The decoder is iterative: two soft decoders each produce a likelihood, exchange them, and re-run. The interleaver is the key — it ensures the two sees'' of the message fail on different bits, so each decoder's confidence patches the other's doubt.
The result, presented in 1993, reached within a fraction of a decibel of the theoretical channel capacity — a margin all earlier codes treated as unreachable.
why it works
- The interleaver makes the two component codes' error patterns independent.
- Passing soft (probabilistic) values, not hard bits, keeps the available information through each round.
- The iteration lets the two opinions converge, so even a weak code pair collectively reaches the limit.
what transfers
Two weak estimators that make independent mistakes beat one strong estimator. The trick is to make the errors independent, then give the two a way to tell each other what they are sure about.
what came after
Turbo codes were immediately taken as the breakthrough in channel coding. They were adopted in the 3G cellular standards, satellite communications and Deep Space One, and they triggered a wave of research into iterative decoding that changed modern communication.
references
- Near Shannon limit error-correcting coding and decoding: Turbo-codes (1)
- Near Shannon limit error-correcting coding and decoding: Turbo-codes. 1
spotted an error? The archive wants to know.