gracebaoxp / cnn-aided-bit-flipping-for-belief-propagation

CNN-Aided Bit-Flipping for Belief Propagation Polar Decoder

CNN-Aided Bit-Flipping for Belief Propagation Polar Decoder

We propose a convolutional neural network-aided bit-flipping (CNN-BF) mechanism to further enhance BP decoding of polar codes. It can achieve much higher prediction accuracy and better error correction capability than prior work of critical-set bit-flipping (CS-BF) but with only half latency. Hope this code is useful for peer researchers. If you use this code or parts of it in your research, please kindly cite our paper:

• Related publication 1: Chieh-Fang Teng, Andrew Kuan-Shiuan Ho, Chen-Hsi (Derek) Wu, Sin-Sheng Wong, and An-Yeu (Andy) Wu, "Convolutional Neural Network-aided Bit-flipping for Belief Propagation Decoding of Polar Codes," published in 2021 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP).

• Related publication 2: Chieh-Fang Teng and An-Yeu (Andy) Wu, "Convolutional Neural Network-Aided Tree-Based Bit-Flipping Framework for Polar Decoder Using Imitation Learning," published in 2021 IEEE Transactions on Signal Processing (TSP).

---

Required Packages

• python 3.6.5
• numpy 1.16.4
• tensorflow 1.8.0
• keras 2.2.5

Source Code

• config.py: adjust parameters

  • N : Block length
  • K : Information length
  • in_N : CRC's block length
  • in_K : CRC's information length
  • ebn0 : Desired SNR range
  • numOfWord : Desired batch size

• Generate Training Data.ipynb: generate the training data for CNN

  • data_num : Desired number of training data for each SNR

• NN_BF.ipynb: use the generated training data to train CNN and show the prediction accuracy of both CNN-BF and CS-BF

• CS_BF.ipynb: repeat the simulatin results of critical-set belief propagation, need to set some parameters as below

  • omega : Number of flipped bit
  • all_combination : 0 for flipping and 1 for both flipping and strengthening
  • T_max : Maximum number of flipping trial and is initially set as the size of CS

• Analyze_Bound (progressive multi bit).ipynb: analyze the error correction capability of exhaustive BF (flipping and strengthing) as shown in Fig. 3 in 2. Note that it is very slow...

• Analyze (progressive multi bit).ipynb: analyze the error correction capability of opposite BF (only flipping) as shown in Fig. 3 in 2.

Contact Information

Chieh-Fang Teng:
     + [email protected]
     + [email protected]