Modeling Channel Occupancy Times for Voice Traffic in Cellular Networks

Call holding times in telephony networks are commonly approximated by exponential distributions to facilitate traffic engineering. However, for traffic engineering of cellular networks, channel occupancy times need to be modeled instead to facilitate analytical modeling or to feed network simulations. In this paper, we classify channel occupancy times and present an empirical study based on data obtained from a real cellular network to determine which probability distribution functions can approximate them better. The results are environment dependent, but no assumptions that can be influential are made, as opposed to previous analytical and simulation studies which results are highly dependent on the assumptions made by the authors. We show that all types of channel occupancy times can be approximated by lognormal distribution. For stationary users, channel occupancy times are commonly approximated by exponential distribution due to its tractability, assuming that cell residence times are also exponentially distributed. However, we show that lognormal distribution fits much better to both channel occupancy and call holding times regardless of whether users are stationary or mobile.