Deep Learning Based Channel Estimation for MIMO-OFDM System with Modified ResNet Model

Abstract

Objectives: The effectiveness of wireless communication systems is significantly influenced by channel estimation. In order to accurately estimate the Channel Impulse Response (CIR) of the channel under varied circumstances, channel estimation is a crucial procedure in the functioning of MIMO-OFDM (Multi-Input and Multi-Output – Orthogonal Frequency Division Multiplexing) systems. Methods: The proposed Deep Learning(DL) based Convolutional Neural Network (CNN) with modified ResNet architecture channel estimation method improves the Bit Error Rate(BER) and Mean Square Error(MSE) performance compared to conventional channel estimation methods. We have compared the proposed CNN method with the Least square (LS), Minimum Mean Square Error(MMSE) and Deep Neural Network(DNN) based channel estimation methods. The results have been discussed by using BER and MSE versus SNR graphs. The simulation results are being performed on the MATLAB platform of the R2021b version. Findings: The DL-based MIMO-OFDM channel estimation can achieve better performance over multipath fading channels if Channel coefficients are perfectly estimated at the receiver. The simulation test is carried out in different test conditions by considering the different number of transmitter and receiver antennas with respect to different QAM modulation order values. Novelty: The DL-based modified ResNet architecture comprises a set of layers modified to estimate the optimal channel parameters. For achieving a great reduction of MSE and BER compared to conventional channel estimation methods, the layers of the ResNet– model are modified. Keywords: Channel estimation; MIMOOFDM system; Deep learning; Neural network; ResNet