Numerical Evaluation of the Channel Estimation in 5G NR Based on Machine Learning

This paper deal with the numerical evaluation of the wireless channel estimation based on machine learning approach in the multiuser multiantenna downlink system. As sounding sequence in frequency domain, the CSI-RS sequence are used. The convolution neural network is applied for the task of channel estimation in frequency domain using LS channel estimation on CSI-RS pilots signals. Training data set is generated using the open source 5G channel model QuaDRiGa for static scenario. CNN-based channel estimation algorithm is trained with the data from LS channel estimation and the true channels to obtain the parameters of CNN. CSIRS resource is configured according to 5G NR standard. Numerical results for selected user demonstrate the better performance of the learning-based channel estimation compared the least squares channel estimation method in the frequency response domain and MSE of channel estimation. The accuracy of channel estimate is improved with increase of training epoch and with reducing the learning rate. The obtained results show that machine learning-based channel estimation methods can be used for channel estimation in modern 5G networks or other networks as an alternative to existing methods.