A major aspect of ad-hoc networks is that the nodes can move
randomly, which requires the routing protocols in ad-hoc network to quickly
respond to the network topology change in order to guarantee successful data
packet delivery. A link state prediction method can
predict the exact link breakage time of an active link before the breakage
actually occurs. So by using link state prediction, a new route can be constructed before the old route becomes unavailable,
thus avoiding data packet loss. In this thesis, we first added the link state prediction method to the reactive unicast
protocol AODV. The source can smoothly update the currently used route to avoid
any soon-to-be-broken link. Simulation results demonstrate that this pro-active
route maintenance can significantly reduce packet loss (between 32% and 72%)
with slight overhead increase (between 4% and 49%). We also examine the link state prediction method in the tree-based multicast
protocol MAODV to maintain the multicast tree in advance and avoid any branch
breakage. Simulation results show the throughput is greatly
improved to above 85% from around 70% with overhead increase below 12%.