Structured Masked Diffusion for Shared-Codebook Unsourced Random Access Decoding
Abstract
In shared-codebook unsourced random access (URA), a receiver must recover an unordered set of messages from simultaneous transmissions. Its outer decoder assembles ambiguous slot-wise evidence into distinct codewords while enforcing parity constraints. We propose CIDER, a learned outer decoder that combines masked-diffusion refinement with cross-row demixing and parity-aware propagation on the code's Tanner graph. At higher per-bin loads, quality-guided remasking selectively re-decodes low-confidence rows. On the evaluated GF(64) LDPC codes at K=2, CIDER has lower measured symbol and codeword error rates than SIC-BP and FFT-BP at all four code lengths. In same-GPU decoder-only measurements, the deployed mixed-precision implementation is 8.2× to 167× faster than our optimized FFT-BP implementation across those lengths. We also evaluate higher loads, estimated user counts, and detector mismatch. Code is available at https://github.com/jiyunyoung/CIDER/tree/neurips-2026.
Related
- Route Based Map Matching via a Structured Codebook and Token Sequence Decoding
- Exact Joint MAP Decoding for Unknown Load and Message Counts in Orthogonal Unsourced Random Access
- Representation-based Masked Diffusion Model
- Query-Efficient Rejection Sampling for Parallel Masked Diffusion
- Serial Depth of Adaptive Parallel Sampling: Lower Bounds and Schedules for Masked Diffusion Models
BibTeX
@misc{lee2026structured,
title = {{Structured Masked Diffusion for Shared-Codebook Unsourced Random Access Decoding}},
author = {Taekyun Lee and Jiyoung Yun and Jeffrey G. Andrews and Hyeji Kim},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2605.26580},
eprint = {2605.26580},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.26580},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Structured Masked Diffusion for Shared-Codebook Unsourced Random Access Decoding
AU - Taekyun Lee
AU - Jiyoung Yun
AU - Jeffrey G. Andrews
AU - Hyeji Kim
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2605.26580
UR - https://arxiv.org/abs/2605.26580
AB - In shared-codebook unsourced random access (URA), a receiver must recover an unordered set of messages from simultaneous transmissions. Its outer decoder assembles ambiguous slot-wise evidence into distinct codewords while enforcing parity constraints. We propose CIDER, a learned outer decoder that combines masked-diffusion refinement with cross-row demixing and parity-aware propagation on the code's Tanner graph. At higher per-bin loads, quality-guided remasking selectively re-decodes low-confidence rows. On the evaluated GF(64) LDPC codes at K=2, CIDER has lower measured symbol and codeword error rates than SIC-BP and FFT-BP at all four code lengths. In same-GPU decoder-only measurements, the deployed mixed-precision implementation is 8.2× to 167× faster than our optimized FFT-BP implementation across those lengths. We also evaluate higher loads, estimated user counts, and detector mismatch. Code is available at https://github.com/jiyunyoung/CIDER/tree/neurips-2026.
ER -