Learning to Construct Nested Polar Codes: An Attention-Based Set-to-Element Model

doi:10.1109/LCOMM.2021.3114118

Learning to Construct Nested Polar Codes: An Attention-Based Set-to-Element Model

2021-12-01

14

OA

AI

DOI：10.1109/LCOMM.2021.3114118

Original

Original request for help

Save

Abstract

En 中文

As capacity-achieving codes under successive cancellation (SC) decoding, nested polar codes have been adopted in 5G enhanced mobile broadband. To optimize the performance of the code construction under practical decoding, e.g. SC list (SCL) decoding, artificial intelligence based methods have been explored in the literature. However, the structure of nested polar codes has not been fully exploited for code construction. To address this issue, this letter transforms the original combinatorial optimization problem for the construction of nested polar codes into a policy optimization problem for sequential decision, and proposes an attention-based set-to-element model, which incorporates the nested structure into the policy design. Based on the proposed architecture for the policy, a gradient based algorithm for code construction and a divide-and-conquer strategy for parallel implementation are further developed. Simulation results demonstrate that the proposed construction outperforms the state-of-the-art nested polar codes for SCL decoding.

Keywords:

Polar codes

Codes

Decoding

Neural networks

Optimization

Reliability

Transforms

Nested polar codes

learning to optimize

neural networks

attention mechanism

Learning to Construct Nested Polar Codes: An Attention-Based Set-to-Element Model

Abstract

Journal

Organization