MOS prediction by SDN controller and User Equipment to maintain good quality for VoIP over WiFi

In order to maintain good audio quality when a mobile user changes his attachment point, we present on this paper a model for predicting the mean opinion score (MOS) value by artificial intelligence based on network parameters such as end-to-end delay, jitter, packet loss and playout loss. This model is composed of two parts, the first is the prediction of the MOS by multilayer perceptron (MLP) implemented at the mobile node for prediction and trigger of handover, the second is the ranking of access points with a k-nearest neighbor algorithm (KNN) by SDN controller. The prediction of the handover is done when the value of the predicted MOS is less than 4.03, and trigger it when it equals 3.9, so the role of the controller is to provide users who foresee the need for a handover with the list of access points that generate a better quality of audio service. To assess the relevance of the model, we tested it in a WiFi network, which contains a VoIP server, and compare it with the technique based on the strength of the received signal (RSSI). The results of the simulation prove that we were able to maintain a VoIP quality that does not go down more than 4.3, and a significant reduction in the packet loss rate that no longer exceeds 1.2%.