Linear and non-linear channel prediction performance for a MIMO-OFDM system

This paper presents the design and performance analysis of linear and non-linear channel prediction algorithms used in 4G communication systems. The linear prediction algorithms are based in Autoregressive (AR) model and Kalman filter; the non-linear prediction algorithms are based on neural network (NN) in a time delay and recurrent (RNN) configuration. The design and validation of the algorithms were made using a MIMO-OFDM system described using SystemC. Performance metrics such as latency and Mean Square Error (MSE) are used for comparison. Results indicate that even though latency increases in the system, with both linear and non-linear prediction, non-linear algorithms show lower MSE when trained properly. Configuration parameters of the algorithms are key to find a relationship between latency and MSE.