A Wasserstein {GAN} Autoencoder for {SCMA} Networks

In the view of the exponential growth of mobile applications, the Sparse Code Multiple Access (SCMA) is one promising code-domain Non-Orthogonal Multiple Access (NOMA) technique. The challenge in the adoption of SCMA in practical networks is twofold: designing optimal codebooks at the transmitter and decoding the data at the receiver. However, most of the works available in the literature address only one aspect. In this letter, we design an end-to-end SCMA en/deconding structure based on the integration between a state-of-the-art autoencoder architecture and a novel Wasserstein Generative Adversarial Network (WGAN) that improves the robustness to the channel noise. We compare the performance obtained by the proposed network with conventional computationally intensive solutions and a deep learning based autoencoder. The performance achieved show better performances in terms of Symbol Error Rate (SER), especially at low energy per bit to noise power spectral density ratio.