This repository is the source code and data for the paper
F. Siegismund-Poschmann, B. Peng and E. A. Jorswieck, "Non-Orthogonal Multiple Access Assisted by Reconfigurable Intelligent Surface Using Unsupervised Machine Learning", 31st European signal processing conference.
Run train_noma.py with the following arguments to train the model:
tsnr: transmit SNR.pmax: maximum transmit power.ris_shape: RIS shape, default: 32 x 32.lr: learning rate, default: 1e-5.
Download data from https://drive.google.com/file/d/1Vk1jgQY-wYwibYdaUbz5Az-nWIYCPSYY/view?usp=sharing
and put it in the folder data.
Run test.py to test the saved model on the validation data set.