unlimitedhummus / autoencoder_test

Autoencoder trained for transmission over AWGN channel

This repository houses an Autoencoder built using Tensorflow. It's supposed to optimize transmission over a lossy channel. The Autoencoder is Trained on the MNIST Dataset and the project should serve as a proof of concept.

The Simulation folder houses training.py, testing.py a folder for tensorboard logs and a folder where the model is saved. In training.py the model gets trained In testing.py a few predictions get made with the model and you can look at how the transmitted images compare to the original ones.

autoencoder_test's Issues

Add option to turn on and off rounding for pretraining

Add SNR

I translated the AWGNChannel from the komm Library to tensorflow, but it doesn't quite work correctly yet.

AWGNLayer

Make a working version of the AWGNLayer that can be implemented in between encoder and decoder after training.

1. convert floats to integers

split training and serving

Das Trainieren und das Anwenden des Modells auf Daten sollen unabhängig voneinander in zwei verschiedenen Programmen passieren.