ConvAE: A New Channel Autoencoder Based on Convolutional Layers and Residual Connections

Dong Jin Ji, Jinsol Park, Dong-Ho Cho

Abstract: In this letter, we propose ConvAE, a new channel autoencoder structure. ConvAE uses residual blocks with convolutional layers. This configuration increases performance while decreasing computational complexity at run-time compared with conventional channel autoencoders. The simulations using both conventional and proposed autoencoders for a 2-by-2 multiple-input multiple-output (MIMO) system under Rayleigh and Nakagami-m fading show that the ConvAE is able to attain a lower bit error rate and higher achievable rate relative to the conventional channel autoencoder schemes.

External IDs:doi:10.1109/lcomm.2019.2930287