DLSTM-Based Successive Cancellation Flipping Decoder for Short Polar Codes

Area of Science:

• Telecommunications Engineering
• Information Theory
• Machine Learning for Signal Processing

Background:

• Polar codes are the 5G control channel coding standard.
• Existing successive cancellation flipping (SC flipping) algorithms show poor performance with short polar codes.
• There is a need for improved decoding methods for short polar codes.

Purpose of the Study:

• To propose a novel decoding algorithm for short polar codes.
• To enhance the performance of polar code decoding, particularly for short block lengths.
• To address the limitations of the SC flipping algorithm in 5G control channels.

Main Methods:

• A Double Long Short-Term Memory (DLSTM) neural network is proposed to identify the first error bit.
• Frozen bits are clipped in the DLSTM output layer to improve prediction accuracy.
• Gaussian approximation and multi-bit flipping strategies are employed to enhance channel reliability assessment and error correction.
• Padding and masking techniques ensure compatibility with various block lengths.

Main Results:

• The proposed DLSTM-based algorithm demonstrates competitive error-correction performance compared to the CA-SCL algorithm.
• The DLSTM decoder significantly outperforms the machine learning-based multi-bit flipping SC (ML-MSCF) decoder.
• Superior performance is achieved over the dynamic SC flipping (DSCF) decoder for short polar codes.

Conclusions:

• The DLSTM neural network offers a robust and effective solution for decoding short polar codes.
• This approach significantly improves error-correction capabilities in 5G control channels.
• The proposed method represents a substantial advancement over current machine learning-based and dynamic flipping decoders.