Machine Learning Based Channel Estimation Optimization for OFDM Communication Systems

Abstract

Internet of things (IOT) networks aim for providing significantly higher data rates. Typical IOT applications like power IOT involves increasing volume of data, which requires high performance data transmission. Orthogonal Frequency Division Multiplexing (OFDM) is currently promising for IOT. Estimation of maximum doppler shift (MDS) is inevitable for the channel response estimation in OFDM systems. To improve the accuracy and efficiency of channel estimation, we propose machine learning (ML) based MDS estimation method in this paper. Our method is based on the fact that the distribution of the instantaneous frequency offset (IFO) is related to the MDS. The ML algorithm is used to learn the functional relationship between the statistic of the IFO and the MDS. To make our method feasible in the realtime communication process, we further propose MDS estimation architecture. The functional relationship is obtained through the offline training and can be directly used in the communication process, thus greatly decreasing the implementation complexity. Simulation results indicate that our method is effective in a wide range of MDS and signal to noise ratio (SNR), and greatly improves the communication performance.