We propose a low-complexity recurrent neural network (RNN) polar decoder with codebook-based weight quantization. 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, Chen-Hsi (Derek) Wu, Andrew Kuan-Shiuan Ho, and An-Yeu (Andy) Wu, "Low-complexity Recurrent Neural Network-based Polar Decoder with Weight Quantization Mechanism," published in 2019 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP).
- Related publication 2: Chieh-Fang Teng, Chun-Hsiang Chen, and An-Yeu (Andy) Wu, "An Ultra-Low Latency 7.8–13.6 pJ/b Reconfigurable Neural Network-Assisted Polar Decoder with Multi-Code Length Support," published in 2020 IEEE Symposium on VLSI Circuits (SOVC).
- Related publication 3: Chieh-Fang Teng and An-Yeu (Andy) Wu, "A 7.8–13.6 pJ/b Ultra-Low Latency and Reconfigurable Neural Network-Assisted Polar Decoder With Multi-Code Length Support," published in 2021 IEEE Transactions on Circuits and Systems I: Regular Papers (TCAS-I).
- python 3.6.5
- numpy 1.16.4
- tensorflow 1.8.0
- Users need to customize the config.py and Polar-NN-MULT.ipynb as
N: Block lengthK: Information lengthebn0: Desired SNR rangenumOfWord: Desired batch sizebp_iter_num: The number of iteration for BPRNN: Whether using recurrent architecture (1 = yes)quantize_weight: Different mechanism for weight quantization (0 for non-quantize, 1 for normal, 2 for binarized, 3 for bin, 4 for binarized bin)bin_bit: The number of different valuebinary_prec: The number of weight precision (binary_prec must >= bin_bit)
Chieh-Fang Teng:
+ [email protected]
+ [email protected]