Performance Modeling of Hybrid Satellite/Wireless Networks Using Fixed Point Approximation and Sensitivity Analysis of the Performance Models for Network Design
Date: September 24 - September 26, 2008
In this work, we propose a new analytical method for modeling and estimating the performance of hybrid networks that are comprised of terrestrial wireless networks interconnected by a satellite overlay. We take advantage of the natural hierarchy present in the hybrid architecture to split the network into multiple levels based on the node and link characteristics. For the lower level of terrestrial wireless nodes, we use a simple approximate (throughput) loss model that couples the physical, MAC and routing layers. The model provides quantitative statistical relations between the loss parameters used to characterize multiuser interference and physical path conditions on the one hand and traffic rates between origin-destination pairs on the other. For the higher layer consisting of the terrestrial gateway nodes interconnected by the satellite overlay, we use a simple satellite channel bit error loss model coupled with a channel access protocol, to derive a similar relationship between traffic arrival rate and the link loss. We use the technique of hierarchical reduced loss network model, adapted for packet-switched networks, to create a reduced load network model for the hybrid network that connects the packet arrival rates in the different levels with the associated physical and link loss rates. We then apply a fixed point approximation approach for this set of relations to derive a solution that converges to a fixed point for the given set of parameters, while satisfying all the equations in the set. The result is an implicit model of the selected performance metric parameterized by the design variables. We apply the technique of Automatic Differentiation to the performance models obtained through the fixed point approximation and analyze the sensitivity of the performance metrics to variations in the selected network parameter. We thus develop a methodology for parameter optimization and sensitivity analysis of protocols to aid in the design of hybrid networks. We demonstrate the effectiveness of our approach through simulation results, compared with results from discrete event simulations on identical network topology.