Abstract: In multi-hop and multi-relay scenarios for distributed terminal cooperative information transmission, the iterative search method for optimal path selection has high computational complexity. In actual network, the path selection method needs to be completed in a very short time in order to timely perceive and dynamically adapt to the changes of the network environment. Therefore, in order to realize fast path selection under different node positions and different channel state information (CSI), a deep learning-based path selection method (DL_PSM) is proposed, which adopts deep neural network (DNN) model and takes the maximum received signal-to-noise ratio (RSNR) of the selected path as the optimization goal. Specifically, this DNN model uses the position of each node and CSI between nodes as input features and uses the optimal path with the largest RSNR calculated by the exhaustive algorithm as output label. Then the DNN model is trained using training samples composed of input features and labels. Simulation results show that, compared with the existing iterative search method, the DL_PSM can significantly reduce computation latency while achieving 99.3% of optimal RSNR performance.

Key words: path selection, deep learning, distributed terminal, deep neural network