Characterizing backhaul traffic in 3G networks using real-world speech

As third generation (3G) services become universally adopted, carriers will need to minimize recurring bandwidth expenses in their radio access network (RAN) while providing media-appropriate QoS. While the move to statistical link layer technologies is 3G is appealing for the efficient use of bandwidth and the support for other bearer services such as data, its bandwidth requirements are more difficult to characterize. This paper proposes the use of a mobile speech corpus of 1300 recorded voice conversations for analyzing bandwidth requirements. The bandwidth utilization over the UMTS Iub/Iur link layers is studied, taking into account not only voice activity statistics of the speakers from the corpus. but also the effects of background noise, echo, and signaling traffic, as well as vocoder and link update algorithms. We will demonstrate that traditional statistical models of voice activity, as commonly used, do not accurately model link usage of the backhaul links, especially ill the case of mobile-generated speech.