No abstract

A dynamic-user model for centralized wireless networks is studied, where users arrive with a certain file size and depart when the file is served by a central server. Although the exact analysis of dynamic-user systems can be complicated, it is shown that an approximate analysis can be performed in a timescale separation regime where the file size is much larger than the time scale of service process fluctuation. A first-order approximation result is derived that shows that when file sizes are large, a complicated service process can be replaced by a simple constant-rate service process. The accuracy of the approximation is further improved through a second order approximation result that incorporates the effect of service variability. Variability in the service process is shown to reduce the effective service rate, leading to a quantification of the conventional heuristic that service variability degrades system performance.

The design of soft handoff algorithms for cellular radio systems is considered. The design problem is posed as a tradeoff between three metrics: the rate of handoffs, the mean size of the active set, and the link quality. It is argued that the algorithm that optimizes the tradeoff between these metrics is impractical. Hence, a locally optimal (LO) handoff algorithm is derived as a practical approximation to the optimal handoff algorithm. The LO algorithm is shown to yield a significantly better tradeoff than the static threshold handoff algorithm used in second-generation code-division multiple-access (CDMA) systems. It is also shown that the dynamic threshold algorithm, which is an ad hoc algorithm proposed for third-generation CDMA systems, achieves nearly the same performance as the LO algorithm. Thus, an analytical justification is developed for the dynamic threshold algorithm. Further, handoff algorithm design is separated into independent design problems on the forward and reverse links. The forward link LO algorithm is shown to be computationally intensive but is also shown to be closely approximated by the simpler reverse link LO algorithm.

No abstract

No abstract