Time-varying Channel Tracking and Prediction Based on Complex-valued Back-propagation Neural Network

摘要

The Complex-valued Back-propagation Neural Network (CVBPNN) algorithm is applied for time-varying channel tracking and prediction in communication system and is verified in the communication simulation system. Firstly, the CVBPNN tracking model is constructed, the hidden layer and output layer of which have non-linear and linear activation function respectively, and the delayed CSI is looked as the supervised signal for the model training to acquire model parameters. Then the tracking parameters is transferred to the prediction neural network which has the same network architecture as the tracking neural network, and the network is used to forward prediction for the output of the tracking model. Simulation results show that the proposed algorithm has less prediction error, lower computational complexity and faster running speed.