We demonstrate that training of autoencoder-based communication systems on the bit-wise mutual information allows seamless integration with practical bit metric decoding receivers, as well as joint optimization of constellation shaping and labeling. Additionally, we present a fully differentiable neural iterative demapping and decoding structure which achieves significant gains on additive white Gaussian noise channels. Going one step further, we show that careful code design can lead to further performance improvements. Finally, we implement the end-to-end system on software-defined radio and train it on the actual channel.
%0 Conference Paper
%1 tcomSIPS
%A Aoudia, Fayçal Ait
%A Cammerer, Sebastian
%A Dörner, Sebastian
%A Stark, Maximilian
%A Hoydis, Jakob
%A ten Brink, Stephan
%B 2020 IEEE Workshop on Signal Processing Systems (SiPS)
%D 2020
%K autoencoder ml myown
%P 1-1
%R 10.1109/SiPS50750.2020.9195252
%T Extended Abstract: Deep Learning of the Physical Layer for BICM Systems
%U https://ieeexplore.ieee.org/document/9195252
%X We demonstrate that training of autoencoder-based communication systems on the bit-wise mutual information allows seamless integration with practical bit metric decoding receivers, as well as joint optimization of constellation shaping and labeling. Additionally, we present a fully differentiable neural iterative demapping and decoding structure which achieves significant gains on additive white Gaussian noise channels. Going one step further, we show that careful code design can lead to further performance improvements. Finally, we implement the end-to-end system on software-defined radio and train it on the actual channel.
%@ 978-1-7281-8099-1
@inproceedings{tcomSIPS,
abstract = {We demonstrate that training of autoencoder-based communication systems on the bit-wise mutual information allows seamless integration with practical bit metric decoding receivers, as well as joint optimization of constellation shaping and labeling. Additionally, we present a fully differentiable neural iterative demapping and decoding structure which achieves significant gains on additive white Gaussian noise channels. Going one step further, we show that careful code design can lead to further performance improvements. Finally, we implement the end-to-end system on software-defined radio and train it on the actual channel.},
added-at = {2020-11-10T10:50:45.000+0100},
author = {Aoudia, Fayçal Ait and Cammerer, Sebastian and Dörner, Sebastian and Stark, Maximilian and Hoydis, Jakob and ten Brink, Stephan},
biburl = {https://puma.ub.uni-stuttgart.de/bibtex/202634dbd8e697a352773bd5e654b62d4/sdnr},
booktitle = {2020 IEEE Workshop on Signal Processing Systems (SiPS)},
doi = {10.1109/SiPS50750.2020.9195252},
eventdate = {20-22 Oct. 2020},
interhash = {5648f8903ecedcf94d0dc3506d44a5bd},
intrahash = {02634dbd8e697a352773bd5e654b62d4},
isbn = {978-1-7281-8099-1},
issn = {2374-7390},
keywords = {autoencoder ml myown},
month = oct,
pages = {1-1},
timestamp = {2020-11-10T09:50:45.000+0100},
title = {Extended Abstract: Deep Learning of the Physical Layer for BICM Systems},
url = {https://ieeexplore.ieee.org/document/9195252},
year = 2020
}