Delay Modeling for 3G Mobile Multimedia Services QoE Estimation Ianire Taboada, Fidel Liberal, and Jose Oscar Fajardo University of the Basque Country, Spain ianire.taboada@ehu.es http://det.bi.ehu.es/NQAS Abstract. In this paper, we consider mobile multimedia services deliv- ered over a UMTS network. In this convergent scenario, low level error recovery mechanisms of the access network entail that almost all packet losses are caused by frames arriving at the receiver later than the play- out time. Our aim is to find a simplified yet realistic expression for the statistics of the end–to–end delay in order to characterize related appli- cation level losses and infer resulting Quality of Experience (QoE). We present a simple and easy to implement model based on empirically ob- tained parameters that quantitatively reflects the statistical properties of delay. We apply the model to estimate the behavior of application level losses and delay which are the key factors while forecasting QoE in VoIP. Obtained results prove the suitability of the model to be integrated in cross–layer adaptation mechanisms and its utility for dimensioning VoIP playout buffers is also tested. Keywords: UTRAN, SR–ARQ, delay, QoE, VoIP. 1 Introduction Nowadays the utilization of mobile communications for multimedia services over converged networks is becoming more frequent. For the reproduction of these streaming services, packets in the reception buffer are read at deterministically– spaced time intervals, so that those packets that do not respect the timing con- straints imposed by applications (i.e. arrive later than expected playout time) are discarded. Therefore, at the playout time of these discarded packets either “noth- ing” is reproduced or just copies of previous samples, which negatively affects user Quality of Experience (QoE). So, these application level losses happen even in allegedly lossless networks, even under not–severe degradations, whenever the end–to–end (e2e) delay of a packet is above a playout time related threshold (implemented with a buffer). Non–interactive multimedia services usually allo- cate long buffers in order to cope with these losses, so that it is the jitter and not the delay the indicator to consider. Unfortunately, VoIP services can not use so long buffers due to associated initial delay (see [1]) since it would affect the interactivity of the conversation. Even with dynamic adaptative buffers the tradeoff between delay and losses imposes a maximum size for the playout buffer. J. Del Ser et al. (Eds.): MOBILIGHT 2011, LNICST 81, pp. 67–80, 2012. c  Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012