Work

Prime-Ulam LDPC: Erasure Codes Generated from Modular Spirals

Year: 2026
Type: article
Abstract: Prime-Ulam LDPC: Erasure Codes Generated from Modular Spirals Prime-Ulam LDPC is a family of binary erasure codes built entirely from a formula. A prime number *p* and a small set of integers define a complete code, so no parity-check matrix is stored. The description consists of 34 integers besides *p*, plus a 32-row incidence table, which is the one ingredient found by a seeded search rather than by formula. The construction combines three parts derived from the same modular arithmetic. The sparse part places data symbols on families of modular spirals. Because *p* is prime, two symbols never share two equations, so the sparse part has no length-4 cycles. This is a theorem, not a search result. The parity symbols form a closed accumulator chain. Sixteen dense equations, defined by norms of Gaussian integers modulo *p*, are added to remove a residual error floor. At rate 0.80 and 1,480 transmitted symbols, the code recovers all data in 150,000 random erasure patterns at 17.5% loss with no failures. At 19% loss, its failure rate is 1.3×10⁻⁴, compared with 1.9×10⁻³ for the 5G NR LDPC code under the same conditions. Decoding uses only XOR operations on packets and runs at 2.5 to 3.7 GB/s for loss rates up to 15% in a plain C implementation. The limits are stated plainly. Only rate 0.80 has been built and measured. The results cover the erasure channel only, not noisy channels. RaptorQ and Reed–Solomon recover every pattern up to the information-theoretic limit, which Prime-Ulam does not reach. Against a well-tuned PEG code with an accumulator, the advantage is not established at 17.5% loss with the sample sizes used. The contribution lies mainly in the compact, theorem-backed construction and its geometric derivation. The code, data, and reproduction scripts are released under the Apache License 2.0.
Language: English
Open Access status: green
OpenAlex ID: W7236157005
FWCI:
Cites:
Cited by:
Related to: 10