UBC Theses and Dissertations
Modeling and performance evaluations of teletraffic in cellular networks Yavuz, Emre Altug
The growing interest for cellular technology has motivated operators to provide a wide variety of services from conventional circuit-switched voice to packet-switched data and multimedia applications. Providing these services anytime and anywhere is challenging due to not only frequent status changes in network connectivity, but also limited resources such as bandwidth. Different priorities are assigned to services to satisfy diverse QoS requirements. Call admission control schemes have been proposed to manage resources by selectively limiting the number of admitted calls to ensure that QoS measures such as call blocking/dropping probabilities stay within acceptable limits. Exact analysis methods based on multidimensional Markov chain models are used to evaluate performance of these schemes, yet they suffer from curse of dimensionality that results in very high computational cost. Large sets of equations are avoided using approximation methods based on one dimensional Markov chain models assuming that channel occupancy times are exponentially distributed with equal mean values and all calls require equal capacities. Existing approximation methods lead to significant discrepancies when average channel occupancy times differ. We propose a novel performance evaluation approximation method, effective holding time, with low computational complexity to relax this assumption. In multi-service networks, voice is accompanied by data and multimedia applications that require distinct capacities. When capacity requirements differ, existing approximation methods based on one dimensional Markov chain models become inaccurate if not obsolete. We propose a computationally efficient approximation method, state space decomposition, to relax this assumption. Numerical results show that the proposed method provide highly accurate results that match well with exact solutions. Traffic statistics are essential to understand the distribution of idle periods of voice channels to overlay packet-switched services on circuit-switched technology and to feed simulations with realistic data. Call holding and channel occupancy times are key elements for computing performance metrics such as call blocking/dropping probabilities. We present an empirical approach to determine the distribution of call holding and channel occupancy times. We show that lognormal distribution is the closest fitted candidate to approximate channel occupancy times and call holding times for stationary/mobile users along with the number of handoffs committed by a user.
Item Citations and Data