This repository contains the source code to reproduce the numerical results in our paper Data-Driven Upper Bounds on Channel Capacity. The code is written in Python/TensorFlow using Jupyter notebooks and can be found in the folder code.
The following animation shows how the estimated upper bound (red curve) based on the proposed algorithm evolves as a function of the training iteration for the standard additive white Gaussian noise (AWGN) channel. The lower bound (blue curve) is based on the 2019 Paper by Fritschek, Schaefer, and Wunder, see also their github repository. The black curve is the ground-truth capacity.
Note that the optimization (both for the upper and lower bound) is performed separately for each signal-to-noise ratio (SNR).
Paper information:
- C. Häger, E. Agrell, "Data-Driven Upper Bounds on Channel Capacity", under review, 2022