Multi-Antenna Fading Channel Prediction Empowered by Artificial Intelligence

Outdated channel state information (CSI) has a severe impact on a wide variety of wireless techniques. Making use of artificial intelligence, a multi-antenna channel predictor is proposed in this paper. It can accurately forecast the future CSIs up to tens of symbols ahead in a fast fading channel. By tuning the number of neurons in the input and output layers of neural network (NN) in terms of the number of antennas, it congenitally suits a multi-antenna system. Relying on a NN with real-valued weights, this predictor is simpler but substantially outperforms existing complex-valued NN predictors.