DEVELOPMENT OF A HYBRID CNN-RF MODEL FOR PREDICTION OF CHANNEL STATE INFORMATION FOR WIRELESS COMMUNICATION CHANNELS

Abstract

Estimation of channel state information (CSI) is one of the major challenges in wireless communication systems. Various methods have been proposed in literature for performing the CSI estimation. However, these methods suffer from high computational complexity. In this thesis, efficient CSI prediction schemes are explored. Several salient features of the radio link have been identified which affect the CSI. Further, a hybrid model for CSI prediction is proposed, which is a combination of a Convolutional Neural Network (CNN) and a Random Forest (RF) model. An offline training mechanism is used for the proposed model. To determine the performance of the proposed hybrid CNN-RF model, a practical case study is considered where the CSI between a Raspberry Pi B4 and a D-link WiFi router is predicted using the proposed hybrid CNN-RF model. The CSI values predicted with the proposed CNN-RF model are found to have high accuracy. Different performance metrics such as mean square error, mean absolute error, root mean square error, mean absolute percentage error, and accuracy percentage are used for the comparative simulation study. In future, an online prediction scheme can be investigated for the prediction of CSI from historical data. Mathematical model for the convergence rate of the proposed hybrid CNN-RF can be further studied.