| In classical cellular network analysis, three assumptions were made: cell traffic (both of new calls and handoff calls) is Poissonian, call holding time is exponentially distributed, and the channel holding time is exponentially distributed. But in recent PCS teletraffic research, people have found that the channel holding time distribution may not be exponential, cell traffic may not be Poissonian. Field data has also shown that the call blocking probability is higher than what is given by Erlang-B formula. It is not clear what really causes the breakdown of the Poissonian property of the cell traffic and exponential property of channel holding time. And it is not known how the blocking probability is in G/G/m system. It is therefore, imperative to reexamine the conventional assumptions for analysis in classical cellular networks and investigate how the updated modeling assumptions affect the performance of the emerging PCS networks. Based on the exploring work of Chlamtac, Lin and Fang, it was found that the channel holding time can be derived from cell residence time which is the time a mobile stays in a cell. This research is divided into three parts. The goal of Part I is to find out how the cell traffic and blocking performance are affected by the mobility parameters under general cell residence time. We use Gamma distribution for cell residence time, by varying the high moment, i.e., the variance, to evaluate the effect on handoff traffic, and call blocking performance.; One impact of the non-Poisson cell traffic and non-exponential channel holding time on network design is resource allocation. In Part II, we address how the mobility affects the channel allocation. We propose a dynamic channel allocation scheme which adapts the number of guard channels according to the instantaneous traffic situation. We plug the results of channel holding time and blocking analysis into the adaptive channel assignment. Our results show that, when the variance of cell residence time is large, the number of guard channels can be reduced.; Besides traditional voice application, Internet data traffic is more and more paid attention. Internet traffic exhibits self-similar behavior, which can not be modeled by Poisson process and shows long-range dependence. In Part III, we target GSM/GPRS network, evaluate how the performance is affected by the mobility under integrated circuit-switched voice and packet-switched data. |
