Machine Learning Based Early Termination for Turbo and LDPC Decoders

Abstract

Turbo and low-density parity-check (LDPC) codes have been chosen in wireless communications because of their near channel capacity performance. However, they consume huge power and also induce delay because of the iterative nature of the decoders. Various types of early termination (ET) techniques have been introduced within these decoders to decrease the power consumption and delay. Recently, machine learning (ML) algorithms are being explored to replace or improve the complex receiver algorithms in wireless communications, as a part of 6G research. In this paper, we propose a novel algorithm to use ML within the turbo and LDPC decoders, to identify the iteration for ET. We show that the proposed algorithm outperforms the improved hard decision aided ET method by 25% $\sim 57$%, in reducing the average number of iterations (ANI) of turbo decoder at 10% block error rate (BLER), for multiple modulation schemes. We also show that the proposed ET method outperforms parity check equation ET method of LDPC decoder by 30%$;\sim 36$%, in reducing the ANI at 10% BLER, for multiple modulation schemes. The proposed method has negligible loss in BLER performance compared to typical implementation of fixed iteration decoders.