Deep Learning Aided Channel Estimation Approach for 5G Communication Systems

Abstract

The defining feature of the Fifth Generation (5G) mobile communication systems is going to be Multiple Input Multiple Output (MIMO) transmission scheme, which utilizes the multipath diversities to achieve beamforming and increase spectral efficiency. However, these MIMO algorithms rely on accurate channel parameters. To improve the accuracy of the channel coefficients, the study presents a Deep Learning (DL) based approach that uses the 5G Demodulation Reference Signals (DMRS) as training sequence and Deep Neural Networks (DNN) as training and prediction network in a MIMO scenario. The DNN is trained with training data obtained by applying Least Squares (LS) method to the received pilot signals and by comparing it to Clustered Delay Line (CDL) channel model. The DNN is then used to predict real-time channel coefficients. The results show that the model improves channel estimation performance by reducing the effects of noise, thus improving the Normalized Mean Square Error (NMSE) versus Signal-to-Noise Ratio (SNR) metric of the MIMO system.