Guest talk: Jakob Hoydis "Deep-Learning for the Physical Layer with Sionna"
Deep-Learning for the Physical Layer with Sionna
Abstract: Sionna is an open-source, GPU-accelerated, and fully differentiable link-level simulator for 6G research. One of its main features is automatic gradient computation through an entire end-to-end system which allows for seamless integration and training of neural networks. After a short introduction explaining the motivation behind Sionna, its features and design principles, I will discuss several of our recent research results, including synchronization for NB-IoT as well as GNN-based channel decoding.