Neural Network-Based Adaptive Polar Coding
Vera Miloslavskaya, Yonghui Li, Branka Vucetic
Abstract
In this paper, we propose a novel artificial intelligence (AI) based adaptive polar coding scheme that adapts to various channel conditions and quality of service requirements. To ensure tight adaptation, we develop a new AI-based performance prediction framework for the precoded polar codes under the successive cancellation list (SCL) decoder. This AI-based framework relies on a neural network and recent advancements in the analysis of precoded polar codes, SCL and SC decoders. Then we apply the proposed framework to optimise precoded polar codes for various target frame error rates (FER), signal-to-noise ratios (SNR) and decoding list sizes <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L$ </tex-math></inline-formula> , where the code length is fixed to a power of two, but the code rate may vary. We predict the throughput and maximise it over the code rates with bit-level granularity. The proposed approach paves the way towards online adaptive polar coding with high error-correction capability. The constructed codes can be compactly specified using the reliability sequence from the 5G New Radio standard and a single parameter whose value is specific to each code. The simulation results show that the proposed codes outperform 5G polar codes with CRC11 under SCL decoding with various <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L$ </tex-math></inline-formula> .