Finding the Best Reference Signal Pattern for Different Channel Models in 5G

One of the challenging issues in Orthogonal Frequency Division Multiplexing (OFDM) systems is to efficiently estimate the Channel State Information (CSI) at the receiver to guarantee reliable signal detection and reception. The most commonly used channel estimation method is to insert a few known reference symbols into the time-frequency grid called the pilots. The accuracy of channel estimation depends on the pilots' density and location. In this paper, the best pilot position for 5G NR is found by using Deep learning techniques and Convolutional Neural Networks (CNN). The neural network is trained by large set of training data to learn the channel structure. Different pilot positions and patterns are tried with Demodulation-Reference Signal (DM-RS); and the pattern with less Mean Square Error (MSE) is determined by CNN. Firstly, the DM-RS symbols and indices for a given carrier and resource configuration are generated. It is then modulated and passed through different nrTDL delay profiles with different delay spread and Doppler shifts. Various patterns of conventional DM-RS resource grid with different configuration types and additional positions are simulated and the results are compared with the new Hybrid H-Type (a combination of parallel and block patterns) DM-RS pattern. It is observed from the results that the Hybrid H -Type pattern gives a 7% lesser channel estimation error compared to the conventional patterns.