Reinforcement Learning for Bit-Flipping Decoding of Polar Codes.

Area of Science:

• Coding Theory
• Machine Learning
• Digital Communications

Background:

• Traditional successive cancellation (SC) decoding for polar codes suffers from error propagation, limiting performance.
• Existing methods to mitigate error propagation often introduce decoding delays.

Purpose of the Study:

• To develop a novel decoding algorithm for polar codes that overcomes error propagation and reduces decoding delay.
• To integrate reinforcement learning with SC flip (SCF) decoding to enhance efficiency.

Main Methods:

• A Q-learning-assisted SCF (QLSCF) decoding algorithm is proposed, combining reinforcement learning with SCF decoding.
• A reinforcement learning model is established for selecting candidate bits during the decoding process.
• The QLSCF algorithm removes decoding delay associated with metric ordering.

Main Results:

• The proposed QLSCF algorithm effectively selects candidate bits for SC flipping decoding.
• Simulation results show a reduced decoding delay compared to the standard SCF decoding algorithm.
• The QLSCF algorithm achieves this reduction without compromising decoding performance.

Conclusions:

• The QLSCF decoding algorithm offers an efficient solution for polar code decoding by leveraging reinforcement learning.
• This approach successfully addresses error propagation and minimizes decoding latency.
• QLSCF presents a promising advancement in decoding techniques for polar codes.