Super-FEC Codes for 40/100 Gbps Networking

Zhongfeng Wang

Introduction

As optical transport networks (OTN) transition to 40 Gbps and 100 Gbps speeds, maintaining signal integrity becomes increasingly difficult. To ensure carrier-grade reliability, these high-speed links require an output bit-error rate (BER) of less than $10^{-15}$. Traditional Forward Error Correction (FEC) schemes, such as the Reed-Solomon (RS) codes defined in the ITU-T G.709 standard, are no longer sufficient. These older codes provide a Net Coding Gain (NCG) of approximately 6.2 dB, which falls short of the requirements for ultra-long-haul high-speed transmission.

This paper proposes a "Super-FEC" solution based on a Super-Product BCH (SP-BCH) code. By utilizing a true product code structure and a novel decoding strategy, the proposed system achieves near-Shannon limit performance and effectively eliminates the "error floor" issues common in modern pseudo-product code constructions.

The Challenge of Low Bit-Error Rates

Modern FEC research often relies on software simulations to estimate performance. However, standard simulations are computationally limited and typically cannot reach the $10^{-15}$ BER threshold required by the industry. Most researchers stop at $10^{-12}$ and extrapolate the results.

This extrapolation is dangerous because many high-performance codes suffer from an "error floor"—a region where the BER curve flattens out due to specific "dead patterns." These are small groups of errors that the code's structure cannot correct. If an error floor exists at $10^{-13}$, a simulation stopping at $10^{-12}$ will completely miss it, leading to over-optimistic performance claims.

Proposed Architecture: Super-Product BCH (SP-BCH)

The proposed SP-BCH code is a true product code. Unlike "pseudo-product" codes where rows and columns only partially overlap, a true product code ensures that every bit in the block is part of both a complete row codeword and a complete column codeword.

Structure of the proposed Super-Product BCH (SP-BCH) code.

The architecture consists of a massive block with 960 row codes and 987 column codes. Both the rows and columns utilize BCH codes with a correction capability of $t=3$. Because $t$ is small, error locations can be determined using direct algebraic equation solvers rather than complex iterative algorithms like the Berlekamp-Massey algorithm. This significantly reduces the computational complexity per bit, making the code suitable for high-throughput VLSI implementation.

Dynamic Reverting

To decode this large structure efficiently, the paper introduces "dynamic reverting." In standard iterative decoding, a false correction in one stage can propagate and cause more errors in the next. Dynamic reverting allows the decoder to detect when a column correction is likely erroneous and "revert" the bits to their previous state. This technique allows the code to achieve better performance in just 7 iterations than conventional decoding achieves in 10.

Analytical Performance Bounding

To address the simulation limit problem, the author presents a mathematical framework to derive a lower bound for the BER. By identifying the most likely "dead patterns"—configurations of errors that exceed the $t=3$ capacity of both rows and columns—the total BER can be estimated by summing the probabilities of these specific patterns. This "simple summation" principle is highly accurate at extremely low BERs because the probability of multiple major error patterns occurring simultaneously is negligible. This method allows the author to mathematically prove the absence of an error floor down to $10^{-21}$.

Results and Discussion

The SP-BCH code was evaluated against the standard RS(255, 239) code and several "Example Codes" designed to mimic current industry proposals (like G.975.1.I4).

Performance comparison and analytical BER bounds for the SP-BCH code.

As shown in the performance curves, the SP-BCH code achieves a massive Net Coding Gain of 9.4 dB at a BER of $10^{-15}$. This is a significant improvement over the 6.2 dB provided by standard RS codes. Furthermore, the analytical bounds (dashed lines in the figure) confirm that the error floor for the SP-BCH code remains below $10^{-21}$, far exceeding the requirements for 100 Gbps optical networking.

Conclusion

The Super-Product BCH code offers a compelling balance of high coding gain, hardware efficiency, and ultra-low error floors. While the large block size introduces some decoding latency, the performance benefits make it a prime candidate for the next generation of 40/100 Gbps optical transport equipment. The analytical bounding method also provides a vital tool for verifying the reliability of FEC codes at BER levels where traditional simulation is impossible.

Summary updated

BibTeX
@article{wang2012super,
  author = {Zhongfeng Wang},
  title = {{Super-FEC Codes for 40/100 Gbps Networking}},
  journal = {IEEE Communications Letters},
  year = {2012},
  doi = {10.1109/lcomm.2012.112012.122083},
}
Figure