Evaluation of neural-network-based channel estimators using measurement data

In multiantenna communication systems, side knowledge about the structure of the possible channel realizations can be exploited to improve the accuracy of the channel estimates and to reduce the computational complexity of the channel estimation procedure. To this end, it has been proposed to train a neural network based on channel realizations from the considered scenario such that the resulting estimator is specialized in the estimation of channel realizations that might occur in this particular scenario. While existing work has evaluated the performance of this approach only based on artificially generated channel realizations from a 3GPP channel model, we train and test the neural-network-based channel estimator with realistic channel realizations from a measurement campaign. The results indicate that the good performance observed in the model-based simulations carries over to more realistic experiments with measured data.