#Code used in paper [Learning to decode Protograph LDPC Codes]
Python environment:
python = 3.6
tensorflow = 2.0.0(use GPU for training)
numpy =1.19.1
(the versions are not too strict)
The file 'GeneratorMatrix.py' can get generator matrixs for every lifting factor by change the value of parmeter 'Z'.
The project contains five files which can train eight different neural LDPC algorithms in our paper.
file | algorithm | hyperparmeter |
---|---|---|
Neural_MS.py | neural NOMS(type1) | default |
Neural_SP.py | neural SP | default |
Neural_simplified_MS.py | simplified neural NOMS(type2) | is_weight=True, is_bias=True, others default |
Neural_simplified_MS.py | simplifeid neural NMS(type3) | is_weight=True, is_bias=False, others default |
Neural_simplified_MS.py | simplified neural OMS(type4) | is_weight=False, is_bias=True, others default |
Neural_MS_damping.py | neural NOMS with damping(type5) | single_damping=False, others default |
Neural_MS_damping.py | simplified neural NOMS with damping(type6) | single_damping=True, others default |
Neural_MS_multiloss.py | neural NOMS but not iteration-by-iteration | default |