Logistic Regression for LDPC Decoding Failure Prediction
Taehyun Kim, Joosung Park · 2021
In this paper, we propose a regression analysis for the early stopping of low-density parity-check (LDPC) decoder to reduce computational complexity and decoding latency. The proposed scheme predicts the decoding failure of the LDPC decoder using logistic regression. The data sets for the decodability classification consist of the index of iteration and the number of unsatisfied parity checks. Simulation results show that with the proposed scheme, the average number of iterations of the LDPC decoder can be significantly reduced without performance loss. In addition, since the proposed scheme determines the decoding failure based on the instantaneous value, it has advantages in terms of memory and computational complexity compared to the early stopping schemes based on the changes in parameters.