Abstract:Considering that most existing end-to-end Autoencoders(AE) are only suitable for point-to-point communication scenarios, this paper proposes a dynamic collaborative communication system based on AE, extending the AE based on deep learning to multi-point communication systems. Three neural network subsystems are constructed, each for learning the optimal encoding, transmission, and decoding at the transmitter, relay node, and receiver, respectively, with joint training of the three to achieve the best transmission performance of the multi-point communication system. Among them, the transmitter and receiver use one-dimensional convolutional layers for signal feature extraction and learning, while the relay node supports two classic relay cooperation methods, Amplify-and-Forward (AF) and Decode-and-Forward(DF), by introducing dense layers and one-dimensional convolutional layers. Simulation experiments show that under the conditions of additive white Gaussian noise and Rayleigh fading channels, the proposed model, using two different cooperation methods, has better error performance than a single point-to-point communication system, verifying the feasibility and effectiveness of the system scheme. In addition, the system supports dynamic node topologies, and without the need for additional training, this system supports real-time changes in the number of relay nodes.
