5G LDPC Codes as Root LDPC Codes via Diversity Alignment

This paper studies the diversity of protographbased quasi-cyclic low-density parity-check (QC-LDPC) codes over nonergodic block-fading channels under iterative beliefpropagation decoding. We introduce diversity evolution (DivE), a Boolean-function-based analysis method that tracks how the fading dependence of belief-propagation messages evolves across decoding iterations. Under a Boolean approximation of block fading, DivE derives a Boolean fading function for each variable node (VN) output (i.e., the a-posteriori reliability after iterative decoding), from which the VN diversity order can be directly determined. Building on this insight, we develop a greedy blockmapping search that assigns protograph VNs to fading blocks so that all information VNs achieve full diversity, while including the minimum additional parity VNs when full diversity is infeasible at the nominal rate. Numerical results on the 5G New Radio LDPC codes show that the proposed search finds block mappings that guarantee full diversity for all information bits without modifying the base-graph structure, yielding a markedly steeper high-SNR slope and lower BLER than random mappings.

💡 Research Summary

The paper addresses the challenge of achieving full diversity on non‑ergodic block‑fading channels (BFCs) using the protograph‑based quasi‑cyclic LDPC (QC‑LDPC) codes that are already standardized for 5G‑NR. While these codes are highly optimized for additive white Gaussian noise (AWGN) channels, their performance on BFCs is limited by the diversity order, which is bounded by a Singleton‑like limit. Existing root‑LDPC constructions guarantee full diversity by imposing rigid “root‑check” constraints, but this structural rigidity degrades AWGN performance.

To bridge this gap, the authors introduce a novel analytical framework called Diversity Evolution (DivE). The key idea is to approximate each fading block’s channel gain with a binary indicator (A_m) (1 if the instantaneous SNR exceeds a threshold, 0 otherwise). During belief‑propagation (BP) decoding, messages are no longer scalar LLRs but Boolean expressions. The min‑sum update rules are recast as Boolean AND (·) for check‑node processing and OR (+) for variable‑node processing. By iterating these Boolean updates over the protograph, DivE yields a Boolean fading function for every variable node (VN). If a VN’s final function equals the sum of all block indicators (A_0+\dots+A_{M-1}), the VN enjoys full diversity.

Building on DivE, the paper defines a “generalized root‑check”. Unlike conventional root‑checks, which rely solely on the block assignment of neighboring VNs, a generalized root‑check can emerge during decoding when all incoming messages to a check node are either already full‑diversity or share the same single‑block indicator. This dynamic structure allows the decoder to create diversity‑providing connections without altering the underlying graph.

The authors then devise a greedy block‑mapping search algorithm for the common case (M=2). The search proceeds in three pre‑assignment steps: (1) preventing stopping sets by assigning VNs that share the same neighbor set to different blocks; (2) assigning all parity VNs to a common block to leave the opposite block free for information VNs; (3) handling punctured information VNs (as in 5G‑NR base graphs) so that they receive both block indicators through their neighboring checks. After these reductions, the algorithm incrementally assigns the remaining VNs. At each step it groups check nodes by the number of unassigned neighbors, uses DivE to test whether a feasible assignment can create a generalized root‑check, and, if so, randomly selects one such assignment. If full diversity cannot be achieved at the nominal code rate, the algorithm progressively includes additional parity VNs, effectively lowering the rate, and repeats the search.

Experimental validation uses the 5G‑NR base graph 1 (BG1) with code length (N=46Z) and nominal rate (R\approx22/46). For a two‑block fading scenario, the greedy search finds a mapping (shown in Table I) where every information VN attains the full‑diversity Boolean function. Monte‑Carlo simulations demonstrate that this mapping yields a markedly steeper BLER slope at high SNR compared with random mappings, achieving roughly a 1 dB gain and reaching the theoretical diversity order (d_c=M=2). When full diversity is infeasible at the original rate, the algorithm successfully recovers it by adding a few parity VNs, confirming the robustness of the approach.

In conclusion, the paper shows that 5G‑NR LDPC codes can be turned into effective root‑LDPC codes solely by optimizing the block‑to‑VN assignment, without any modification of the base‑graph structure. The DivE framework provides a tractable Boolean‑algebraic tool for analyzing and designing such assignments, and the concept of generalized root‑checks extends the applicability of diversity‑aligned coding to a wide range of protographs and higher numbers of fading blocks. This work opens a practical pathway for achieving both AWGN optimality and full diversity in future wireless systems operating over severe fading environments.