Statistical multiplexing and connection admission control in mobile-ATM systems

A third generation TDMA based mobile radio system is considered in which all services are carried across the radio interface on ATM connections. Many future services will be variable rate and there is potential for exploiting this fact to improve the spectral efficiency of the mobile system. However, the gains which can be achieved in practice will be highly dependent on the connection admission control (CAC) procedures used. When an ATM terminal requests connection admission it supplies the CAC with only a limited amount of information concerning its bandwidth requirements. The lack of detailed information made available to the CAC can mean that calls which could be supported are instead refused admission. In this paper two different convolution-based CAC algorithms are considered. The first algorithm assumes the worst case cell arrival process which fits the source traffic descriptors declared by the terminal. It is shown that the statistical multiplexing gain achievable when this algorithm is used is very poor and sometimes nonexistent. A second algorithm is therefore considered which makes use of actual measurements of the cell arrival process of calls already admitted in order to provide an improved prediction of the performance which would be seen should the new call be admitted.