Assessment and Improvement of the Quality of Voice-over-IP Communications

This dissertation addresses real-time communications using Voice-Over-IP (VoIP) technology, which is nowadays widely used by enterprises and individual users. The focus is on the assessment of the Quality-of-Experience (QoE) from the end-user's perspective and the development of algorithms and techniques to improve the overall QoE. One of the main contributions is a generic testing tool than can be used for any Voice and Video-Over-IP (VVoIP) application in any environment. The tool employs network emulation techniques to provide estimates for the perceived voice and video quality on different network paths. Importantly, the tool operates without the need for use of traditional quality assessment techniques which are known to be time and resources consuming as they require end-user involvement to collect audio/video sequences and network traces. Our tool emulates the audio and video tra�c and employs the E-model and video quality opinion model to estimate audio and video quality respectively, with the advantage of emulating various network conditions to run experiments in multiple scenarios. Secondly, we present a generic adaptive algorithm for switching audio codecs throughout an ongoing call. Codecs are known to have different behaviours under various network conditions; we study the behaviour of five of the most commonly used codecs (including some non-ITU codecs), deriving models for them so that E-model can be used to assess the Quality-of-Experience. Furthermore, we analyse the negative of codec switching from the end user perspective, so that this impact can be minimized as much as possible. We describe results of tests of the algorithm under different network scenarios; these results suggest that the algorithm can deliver better Quality-of-Experience than would have been achieved by employing one codec only during the call. Lastly, we study different multi-party conferencing architectures with a focus on the centralised architecture which is most commonly used. We analyse the degradation to the quality that results from the need of passing of every packet in the conference through the focal node, and to further decode and then encode those packets to be sent again to their intended destinations. An extended E-model is presented to be used with multi-party calls - we introduce a correction formula to three of the most used codecs so that E-model can be valid when used to estimate the quality of experience of multi-party conferencing calls.