gracebaoxp / signal_detection_ofdmpowerofdnn

MATLAB demonstration for the paper 'Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems'

Signal_detection_OFDMPowerofDNN

MATLAB demonstration for the paper 'Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems' @ MATLAB R2020b

Rayleigh channel deployed, for the winner2 channel (Data_Generation_WIN2.m shows WIN2 for SISO) install that toolbox but more time will be spent on channel realization, so not suggested, and commu.AWGN(fading signal, SNR) not transmitted signal, I remembered that I % commu.AWGN codes

MMSE_Channel_Tap_Block_Pilot_Demo_1.m from MIMO-OFDM wireless communication book

MMSE_Uniform_PDP.m the paper of OFDM Channel Estimation by Singular Value Decomposition

Did not upload python version, there's a python demonstration uploaded by the author

Demonstration_of_papers_DNN is main

DNN_Regression_Image_SER_Test for taining

Dont need use Test_DNN_regression and Train_DNN necessarily

The_rest includes LSTM, classification and use sequence layer as input layer, but not that relative. I just tried to see if I can implement LSTM and what effects would be if I changed to different layers

update on 2021.04.10

Hi, someone has a question on Loss of DNN_Classification_Trained.mat. The solution is that, you need to run DNN_Regression_Image_SER_Test.m to ontain an DNN_trained which is a trained NN, and save that as XXX.mat with commend save('XXX.mat', 'DNN_trained'). In the Demonstraion file, there's a load commend, then change it to load('XXX.mat'). It allows you to make changes to training options so you can try some changes in changes to see if you can improve the performance.