An online approach for wireless network repair in partially-known environments

Abstract

Wireless Sensor Networks in volatile environments may suffer damage which partitions the network, and connectivity must be restored. We investigate the online problem, in which the repairing agent must discover surviving nodes and the damage to the physical and radio environment as it moves around the sensor field to execute the repair. The objectives include minimising the cost of the repair in terms of new radios and distance travelled, and minimising the time to complete the repair. We consider a number of different agent features which we can combine in different configurations. The repairing agent may prioritise either the node cost or the travel distance. The focus of the agent may be local, with a greedy choice of next partition to re-connect, or global, maintaining a plan for all partitions. To handle the developing knowledge of the network conditions, the agent may revert to full replanning when a change is discovered, or try a minimal adjustment of the current plan in order to minimise the computation effort. For each configuration, we develop a number of heuristics for creating the plans. We evaluate the approach in simulation, varying the density of the connectivity graph and the level of damage suffered. We demonstrate that the plan repair method, while producing more expensive solutions, can require significantly less computation time, depending on the choice of heuristic. Finally, we evaluate the total time to repair the network for different speeds of agent, and we show the relative importance of the agent speeds on the two focuses. In particular, algorithms which prioritise mobility cost are preferred for slow agents, while faster moving agents should prioritise the radio node cost.