Link quality prediction in mobile ad hoc networks

Abstract

Link quality prediction (LQP) is effective in masking topology changes in wireless mobile ad hoc networks (MANETs) as previously shown in reducing packet latency in proactive routing. However, most existing LQP algorithms have only been evaluated in simple simulations with basic radio propagation models. Therefore, we evaluated the accuracy of three existing LQP algorithms for an 802.11 MANET of nodes carried by pedestrians in two real-world, microcellular, urban environments. The best performing of these LQP algorithms failed to predict between 18% and 54% of the total observed packet loss for a prediction of two scconds ahead while between 12% and 43% of successfully received packets were falsely predicted to be lost. These results contrast sharply with the nearperfect accuracy of these algorithms shown in the standard NS-2 simulator. Our results show that LQP algorithms based on basic radio propagation models are not applicable to real-world environments. This thesis describes the design and evaluation of accurate LQP algorithms that are derived from a radio propagation model that closely resembles what is observed in microcellular, urban environments.