US 11,955,987 B2
Neural self-corrected min-sum decoder and an electronic device comprising the decoder
Taehyun Kim, Suwon-si (KR); and Joosung Park, Suwon-si (KR)
Assigned to Samsung Electronics Co., Ltd., Suwon-si (KR)
Filed by Samsung Electronics Co., Ltd., Suwon-si (KR)
Filed on Jul. 25, 2022, as Appl. No. 17/872,692.
Application 17/872,692 is a continuation of application No. PCT/KR2022/006018, filed on Apr. 27, 2022.
Claims priority of application No. 10-2021-0066935 (KR), filed on May 25, 2021.
Prior Publication US 2022/0385305 A1, Dec. 1, 2022
Int. Cl. H03M 13/11 (2006.01); G06N 3/04 (2023.01); G06N 3/082 (2023.01); H03M 13/09 (2006.01)
CPC H03M 13/1111 (2013.01) [G06N 3/04 (2013.01); G06N 3/082 (2013.01); H03M 13/09 (2013.01); H03M 13/1128 (2013.01); H03M 13/1177 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A method for decoding a low-density parity-check (LDPC) code in a wireless communication system, the method comprising:
receiving, by an electronic device, a signal, a codeword and information about an LPDC code used for encoding the codeword;
configuring, by the electronic device, a neural network self-correction condition for adjusting a neural network node deletion and dropout rate;
performing, by the electronic device, iterative decoding on the received signal and the codeword based on the information about the LPDC code used for encoding the codeword, wherein the iterative decoding uses decoding factors and a neural network self-correction technique;
determining, by the electronic device, whether the iterative decoding of the received signal and the codeword succeeds or fails;
storing, by the electronic device, the received signal and the codeword which are successfully decoded, based on a result of the determination; and
updating, by the electronic device, the decoding factors, based on the stored received signal and the codeword,
wherein the updating of the decoding factors comprises:
reading the received signal and the codeword;
erasing at least one node of the neural network through hysteresis-based neural network self-correction;
learning a normalization factor, an offset factor, a channel normalization factor, and a channel offset factor of a corresponding iteration from a structure of the neural network having the read received signal as an input; and
generating learning data for the neural network for optimizing a factor of a next iteration using the learned factors.