Sensor selection for model-free source localization: Where less is more

The ability for a wireless network to precisely localize the radio nodes composing it is a great tool towards system optimization and is increasingly seen as a basic service requirement. In the past, model-free algorithms such as weighted centroid localization (WCL) have proved popular, especially in the context of sensor networks, due to their simplicity and robustness to temporal changes in wireless propagation properties. However, WCL algorithms are biased since they implicitly require a uniform sensor distribution around the source in all directions. In this paper, we demonstrate that

instead of employing all the sensors that result in a possibly unbalanced sensing pattern, it is better to reduce the number of sensors such that the subset of selected sensors

symmetrically distributes around the source, which in principle would need to know the source location in advance. Here, we develop a sensor selection algorithm which manages that goal while blindly. Using less than half of the sensors, a 30% reduction in localization error is demonstrated from our numerical experiments.