Joint Channel Estimation and Signal Detection using Latent Space Representations in VAE

This paper presents a data-driven unsupervised Deep Learning-based joint channel estimation and signal detection method for a narrowband wireless communication system. Our proposed Deep Learning-based architecture uses a Variational Autoencoder (VAE) that can combat the effects of additive white Gaussian noise and Rayleigh fading by encoding the input into a lower dimensional representation as the latent space outputs. The lower dimensional representation is used to extract the symbol information and is classified to the corresponding symbols of the transmitted signal using a classifier. We propose two approaches for the VAE-based architecture by using a parallel 1-D VAE and a joint 2-D VAE that takes different signal representations. From our simulation results, the proposed VAE-based architectures can achieve BER performance improvements over a deep Convolutional Neural Network approach and corre-lator detector.