A machine learning based approach for the link-to-system mapping problem

Abstract: The quality of mobile communication is related to signal transmissions. Early detection of the errors in transmissions may reduce the time delay of communications. The traditional error detection methods are not accurate enough. Therefore, in this report, a machine learning based approach is proposed for the link-to-system mapping problem, which can predict the outcomes (received correctly or not) of the link-level simulations without knowing the exact signals that are being transmitted. In this method, the transmission state is assumed to be a function of the features of a channel environment like the interference and the noise, the relative motion between the transmitter and the receiver and this function is obtained using a machine learning method. The training dataset is generated by simulations of the channel environment. Logistic regression, support vector machine and neural networks are the three algorithms implemented in this thesis. Experimental results show that all three algorithms work well compared to traditional methods. Neural networks provide the best results for this problem. Furthermore, the neural network model is tested with a dataset consisting of features of ten different channel environments, which verified the generalization ability of the model.