t-SNE-Assisted Permuted Sparse Graphs Selection for Polar Code BPL Decoding

doi:10.1109/jiot.2026.3720345

t-SNE-Assisted Permuted Sparse Graphs Selection for Polar Code BPL Decoding

2026-08-04

0

PRE

AI

DOI：10.1109/jiot.2026.3720345

Original

Original request for help

Save

Abstract

En 中文

Many researchers regard polar codes as low-density parity check-like (LDPC-like) codes and adopt parallel belief propagation (BP) decoders for polar codes, namely LDPC-like BP list (BPL) decoding. The sparse graphs adopted in LDPC-like BPL decoding are constructed via graph pruning, which destroys the inherent fixed butterfly structure of polar codes. This leads to great difficulties in structural analysis and graph selection. Since sparse graphs are generally represented by high-dimensional adjacency matrices, this article proposes a sparse graph selection scheme based on t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction. The proposed algorithm first reduces the dimensionality of complex sparse graph data, and then applies K-means clustering to select cluster centers for constructing the decoding list. Simulation results demonstrate that the proposed decoder achieves a prominent performance gain of over 0.2 dB over random and cyclic-shift schemes with no extra computational complexity. It also outperforms the Kernel principal component analysis (KPCA)-based scheme in decoding performance.

Keywords:

Dimensionality reduction

low-density parity check-like (LDPC-like) belief propagation (BP)

permuted sparse graph

polar code

t-SNE-Assisted Permuted Sparse Graphs Selection for Polar Code BPL Decoding

Abstract

Journal

Organization

Cited Papers