Guessing random additive noise decoding (GRAND) in concatenated schemes: An investigation of complexity and BER behavior

Giovani Chrestani (2023)

Sammanfattning

Recently, a newly developed family of decoders named as Guessing Random Additive Noise Decoder (GRAND) has been gaining attention. The key idea of the decoder is to look for the most probable error vectors that interacted with the transmitted message, producing the received block. Among the different variations of the algorithm, the soft-output soft-input (SISO) decoder, known as SO-GRAND, emerges as a good candidate for using with iterative schemes by em- plying list decoding. In this work, SO-GRAND’s complexity and performance are analyzed against a trellis-based soft output decoder. These decoders are used as concatenated schemes of a generalized LDPC code. Results show that, while the soft-input hard-output version of GRAND (known as ORBGRAND) demonstrates good complexity qualities, the same is not true for the SO-GRAND that, in general, does not match the amount of computations of the trellis algorithm for typical scenarios of list size. Nevertheless, SO-GRAND has the nice property of trading off complexity at the expense of performance. This approach may look interesting when performance is good enough. Better performance of SO-GRAND algorithms is expected for higher-order approximations in the ORBGRAND algorithm. The study also features several simulations with different types of codes (CRC, Hamming and random), demonstrating the universality quality of both trellis-based and SO-GRAND algorithms.