Skip to content

Latest commit

 

History

History
46 lines (32 loc) · 2.53 KB

README.md

File metadata and controls

46 lines (32 loc) · 2.53 KB

NOLD

Here we provide codes fo papers:

Lei Chu, H. He, Robert Qiu; NOLD: A Neural-Network Optimized Low-Resolution Decoder for LDPC Codes, To appear in IEEE Journal of Communications and Networks

Code Specifications:

Requirements Numpy>=1.13.3, Python >= 3.6, tensorflow>=1.10.0

The codes include 5 modules to carry out simulations in our papers. They are "GenDecodingData", "GenTrainingData", "Floating MS", "Quantized MS" and "NeuralOptimization"

The configuration entrances of these modules are in "Configuration.py". Each module can be operated under different channels by configurating the channel parameters. Other details are as follows:

GenDecodingData: This module is executed by setting "GenConfiure" in "Configuration.py". The outputs of this module are two files containing transmitting signals data and receiving signals data in "./Decoding". These datas are used to execute the MS decoding algorithm.

GenTrainingData This module is executed by setting "TrainingConfiure" in "Configuration.py".
The outputs of this module are two files containing training data set and validation data set in "./TrainingData" and "./TestData". The training data set is used in training process and the validation data set is to test the network performance during training.

Floating MS This module is executed by setting "DecConfiure" in "Configuration.py". The output of this module is a BER result as shown in "./results"

NeuralOptimization This module is executed by setting "TrainingConfiure" in "Configuration.py".
These modules are a series of files contain the training scaling and quantization parameters for each iteration in "./parameters"

Quantized MS This module is executed by setting "DecConfiure" in "Configuration.py". The outputs of this module is a BER result in "./results" This module uses the scaling and quantization parameters in "./parameters"

Noise setting In each module, the channel is selected by setting "channel" as "AWGN", "ACGN", "RLN" in configuration. Among three noise, the "ACGN" needs the correlated parameters to be set.

The codes of this paper are partially based on the codes of provided in the paper . In our codes,the MS decoding module is derived form the SP decoding module of Dr. Liang's code (https://github.com/liangfei-info/Iterative-BP-CNN). We appreciate the pioneer work provided by the authors of the state-of-art work: .