For this project, it adopts the architecture of NMS+DIA+OSD, holding the benefits of low-complexity low-latency, high decoding performance and indepedence of noise variance estimation etc. The related manuscript entitled 'Iterative Decoding of Short BCH Codes and its Post-processing' is: https://arxiv.org/abs/2411.13876
The interactions among the involved modules: Training route: 1)Training_data_gen_63 module generates training data file; 2)BCH_63_training module optimizes the only parameter of NMS and generate training data file for DIA model. 3)DL_Training module uses the output file of step 2 to train DIA model. Testing route: 4)Tesing_data_gen_63 module generates testing data files at varied SNR points. 5)BCH_63_testing module generates testing results for output files in step 4 and generate varied files with NMS decoding failures included. 6)DL_OSD_Testing module utilzes trained DIA model of step 3 and post-processes decoding failure files of step 5 using ordered statistics decoding.
Notice: Some packages need to be installed for these modules to execute properly, say galois, pickle collections, etc. We run above modules on spyder 5.2.2 using python 3.7 of tensorflow 2.X.