Fading Channel Prediction Based On Attention Mechanism

In wireless communication systems, predicting channel state information (CSI) is a basic task. So far, there are various methods to predict CSI. However, getting accurate CSI is challenging mainly due to rapid channel variation caused by multi-path fading, and CSI tends to become outdated. The inaccuracy of CSI will have a serious impact on the performance of wireless systems. Channel prediction can reduce the degradation of wireless communication quality caused by outdated CSI. Aiming at this problem, this paper proposes a new predictor, establishes a model combining deep recurrent neural network and attention mechanism, and uses its powerful time series prediction ability, combined with long short-term memory (LSTM) and gated recurrent unit. The performance of the new model is evaluated in terms of prediction accuracy, and the number of neurons with different prediction lengths and different hidden layers is compared. The prediction results show that the model has better prediction accuracy than the performance based on deep recurrent neural network and convolutional long short-term neural network.