Introduction
In the field of error-correction coding, a persistent challenge is bridging the gap between practical decoding performance and theoretical limits. While industry standards like Turbo and Low-Density Parity-Check (LDPC) codes are highly effective, they often suffer from a gap between their Belief Propagation (BP) threshold—where practical decoders operate—and their Maximum A Posteriori (MAP) threshold, which represents the code's structural limit. This paper explores Spatially Coupled Serially Concatenated Codes (SC-SCCs), demonstrating that "coupling" simple turbo codes into a chain can cause the BP threshold to saturate to the optimal MAP threshold, effectively approaching the Shannon limit.
Methodology
The authors propose a framework for SC-SCCs that links a sequence of $L$ individual serially concatenated codes. Instead of treating data blocks in isolation, the output of an outer encoder at time $t$ is split and distributed to inner encoders at both time $t$ and $t+1$ (for a coupling memory of $m=1$). This "braiding" creates a strong dependency across the chain, which is fundamental to the performance gains observed.
Key methodological components include:
- Density Evolution (DE): Exact DE equations were derived for the Binary Erasure Channel (BEC) to track how erasure probabilities evolve through the iterative decoder.
- Simple Components: The design utilizes simple 4-state recursive systematic convolutional encoders to prove that the power of the code stems from the coupling architecture rather than component complexity.
- Sliding Window Decoding: To address latency, the authors demonstrate that a sliding window decoder can achieve most of the coupling gain without waiting for the entire chain to be received.
- Rate-Compatibility: A specialized puncturing scheme is introduced to maintain "interleaver gain" even at high code rates, making the codes flexible for various applications.
Key Results
The most significant finding is the confirmation of threshold saturation. As coupling memory increases, the BP threshold of the coupled ensemble moves toward the MAP threshold of the original uncoupled code. For a rate $R=1/3$ SC-SCC, the BP threshold reaches $0.6654$, coming within $0.0012$ of the Shannon limit ($0.6667$).
Finite-length simulations (using $K=1024$ bit blocks) show a massive improvement in Bit Error Rate (BER) for spatially coupled versions compared to their uncoupled counterparts. This improvement was observed consistently across various rates, from $1/3$ up to $9/10$. The results indicate that while uncoupled parallel codes (PCCs) might have better initial BP thresholds, coupled SCCs eventually provide superior performance at higher rates due to better distance properties.
Conclusion and Future Work
Spatially coupled turbo codes offer a robust path toward near-capacity performance using low-complexity implementation. By leveraging threshold saturation, simple 4-state encoders can achieve results previously reserved for much more complex systems. While this study focuses on the Binary Erasure Channel, the principles established suggest broad applicability. Future work may involve extending this analysis to Additive White Gaussian Noise (AWGN) channels and further optimizing puncturing patterns for specific hardware constraints.
Summary updated