End-to-End Learning-Based Framework for Amplify-and-Forward Relay Networks

We study end-to-end learning-based frameworks for amplify-and-forward (AF) relay networks, with and without the channel state information (CSI) knowledge. The designed framework resembles an autoencoder (AE) where all components of neural network (NN)-based source destination nodes are optimized together in manner, signal transmission takes place AF node. Unlike literature that employs NN-based node full CSI knowledge, we consider a conventional only amplifies received using gains. Without employ power normalization-based amplification normalizes each block symbols. propose compare symbol-wise bit-wise AE by minimizing categorical binary cross-entropy loss maximizes mutual (MI), respectively. determine estimated MI examine convergence both signal-to-noise ratio (SNR). For these frameworks, design coded modulation differential modulation, depending upon availability at node, obtains symbols 2n-dimensions, n is length. To explain properties 2n-dimensional designs, utilize various metrics like minimum Euclidean distance, normalized second-order fourth-order moments, constellation figures merit. show obtain similar spherical coded-modulation designs inherently optimal bit-labeling outperforms (with faster under low SNR) considerable SNR margin.