Artwork by Sampson Wilcox
Peihong Yuan, Ken R. Duffy, Evan P. Gabhart, Muriel Médard
DOI: 10.1109/TCOMM.2026.3714295
Abstract:
We investigate guessing random additive noise decoding (GRAND) in the context of front-end constraints, particularly focusing on scenarios where analog-to-digital conversion and signal processing are limited to low-bit resolutions. To address the performance bottlenecks imposed by coarse quantization at the receiver front-end, we propose discretized soft GRAND (DSGRAND), a decoding framework that effectively exploits quantized soft information while respecting front-end hardware limitations. Our results show that DSGRAND can closely approximate maximum-likelihood decoding with only a few bits of quantized input, making it suitable for energy- and bandwidth-constrained systems. For example, with a (128, 106) CRC-polar code, even a basic implementation of ORBGRAND using 3‑bit quantization achieves performance on par with or better than CRC-aided successive cancellation list (CA-SCL) decoding with list size 64. Moreover, DSGRAND surpasses CA-SCL with list size 128 in both error performance and efficiency, achieving nearly an order of magnitude lower average complexity and significantly reduced memory requirement. We also demonstrate the applicability of DSGRAND in iterative decoding of component codes within product code structures. These findings highlight DSGRAND’s potential as a soft-decision decoding method well-suited to systems constrained by front-end quantization, offering a favorable balance between I/O cost, memory usage, and decoding performance.