VoIP Performance of LTE Networks (Aachener Beiträge zur Mobil- und Telekommunikation)

Long Term Evolution (LTE) standardized by 3GPP operates fully packet switched and does not have a circuit switched domain as older systems HSPA, UMTS and GSM. It must support a minimum number of voice calls to be recognized as IMT-Advanced system by ITU. Therefore ITU has published the IMT-Advanced Evaluation Guidelines describing evaluation scenarios and minimum quality of service demands for voice calls to be evaluated by system level simulation. Those guidelines require transmission of VoIP packets within 50 ms over the air interface and packet loss rate below 2 %.
3GPP came to the conclusion Voice-over-LTE (VoLTE) with semi-persistent scheduling satisfies the ITU demands. This was verified by independent evaluation groups. Semi-persistent scheduling reserves radio resources for VoIP packets for the duration of a talk spurt while other packets, e.g. retransmissions of lost packets, individually receive resources (dynamic scheduling). Advantage is significantly decreased signaling overhead but the method is more complex than the classical approach individually assigning radio resources for each packet. Individual assignment of radio resources is also used for data traffic of so called Over-the-Top (OTT) VoIP applications.
This thesis compares VoIP capacity of VoLTE and OTT focusing on the impact of the radio link. Radio resource assignment optimized for delay critical traffic like VoIP is assumed for OTT to allow a fair comparison. A concept developed by me to design complex VoIP schedulers from basic resource assignment algorithms is described. It is applied for VoLTE and OTT. The openWNS/IMTAPhy simulator developed at RWTH Aachen University ComNets Research Group and extended at TU Munich Chair of Communication Networks is enhanced for VoIP evaluation based on that concept. Methods to improve VoIP capacity are evaluated separately for up- and downlink. Different channel prediction strategies are applied for each direction and based on that transmission rate adaptation is performed.
Results show improved capacity through aggressive rate adaptation and resource assignment resulting in higher data rates and therefore reduced resource demand while at the same time increasing packet error rate. Too aggressive parameters significantly reduce VoIP capacity while moderate adjustments increase it by ca. 15 % in the downlink and 10 % in the uplink for VoLTE. OTT cannot benefit from these improvements if the number of control channels is limited and becomes a bottleneck. If more control channels were available than defined by the current standard OTT VoIP capacity could be improved for downlink. High uplink interference variance makes channel prediction more difficult and prevents capacity gains for OTT as achieved with VoLTE.
Besides simulation results an analytical discrete time Engset queueing model was described and solved for its waiting time distribution to determine VoIP capacity in scenarios without and with adjacent channel interference from neighbor cells.
In the Urban Macrocell scenario lowest downlink VoIP capacity results 43 users / cell / MHz for OTT and 41 users / cell / MHz for VoLTE were obtained with unlimited number of control channels. The Downlink is the bottleneck and therefore determines the overall VoIP capacity. Allowing users with bad channel condition to transmit at higher data rates than suggested by the rate adaptation algorithm reduces resource occupation but at the same time increases packet error rate. VoIP capacity is increased to 48 and 47 users / cell / MHz this way for OTT and VolTE, respectively.
From this it is concluded that dynamic scheduling used by OTT is capable to provide sufficient VoIP capacity without the complexity introduced by SPS given that sufficient control channels are available, as expected in reality.