On the Dynamics of the Error Floor Behavior in (Regular) LDPC Codes

It is shown that dominant trapping sets of regular LDPC codes, so called absorption sets, undergo a two-phased dynamic behavior in the iterative message-passing decoding algorithm. Using a linear dynamic model for the iteration behavior of these sets, it is shown that they undergo an initial geometric growth phase which stabilizes in a final bit-flipping behavior where the algorithm reaches a fixed point. This analysis is shown to lead to very accurate numerical calculations of the error floor bit error rates down to error rates that are inaccessible by simulation. The topology of the dominant absorption sets of an example code, the IEEE 802.3an (2048,1723) regular LDPC code, are identified and tabulated using topological relationships in combination with search algorithms.

💡 Research Summary

The paper investigates the error‑floor phenomenon in regular low‑density parity‑check (LDPC) codes by focusing on the dominant trapping structures known as absorption sets. An absorption set is a small collection of variable nodes and their neighboring check nodes that, once corrupted, tend to persist under iterative message‑passing decoding. The authors demonstrate that the dynamics of such sets can be captured by a simple linear model, which reveals a two‑phase evolution during decoding.

In the first phase, the erroneous messages grow geometrically. By representing the connections inside an absorption set with an adjacency matrix A, the update of log‑likelihood ratios (LLRs) at each iteration can be written as x_{k+1}=A·x_k. The spectral radius λ₁ of A governs the growth: if λ₁>1, the error magnitude expands roughly as λ₁^k. This “growth phase” continues until the messages reach a magnitude where the linear approximation no longer holds.

The second phase is a stabilization or “bit‑flipping” stage. Here the dominant eigenvalue’s influence saturates, and the second‑largest eigenvalue λ₂ determines the convergence speed toward a fixed point. The decoder ends up in a non‑convergent state where the same set of bits remains in error, which constitutes the error floor. The authors validate this two‑phase picture with extensive Monte‑Carlo simulations, showing that the analytical predictions of the error‑floor bit‑error‑rate (BER) match simulated results down to BER ≈10⁻⁸, and that the model can extrapolate reliably to BER levels (10⁻¹² and below) that are infeasible to reach by brute‑force simulation.

To apply the theory, the paper develops a systematic search algorithm for identifying the most harmful absorption sets in a given code. The algorithm enumerates candidate variable‑node subsets, checks their check‑node connectivity, and uses topological relationships (e.g., the number of shared checks among the variables) to prune the search space efficiently. This approach is applied to the IEEE 802.3an (2048,1723) regular LDPC code. The authors find that the dominant absorption sets consist of eight variable nodes each connected to twelve check nodes in a highly regular pattern (each variable participates in exactly three checks). Spectral analysis of the corresponding adjacency matrices yields λ₁≈1.42 and λ₂≈0.68, which feed directly into the linear model to predict the error floor.

The numerical results show that the analytically computed BER curve aligns with simulation within a few percent across several orders of magnitude, confirming the model’s accuracy. Moreover, because the model is analytical, it can predict BER values far below the simulation horizon, providing designers with a powerful tool for assessing and mitigating error floors.

Finally, the paper discusses design implications. By understanding which absorption sets dominate the error floor and by reducing their spectral radius (e.g., through careful placement of edges during code construction), one can lower λ₁ below unity, thereby suppressing the initial exponential growth and eliminating the error floor. The linear dynamic framework is generic and can be extended to irregular LDPC codes or alternative decoding algorithms, making it a valuable addition to the toolbox of modern coding theory and practical communication system design.