Sensor placement for field estimation via Poisson disk sampling

In this paper, we study the problem of sensor placement for field estimation, where the best subset of potential sensor locations is chosen to strike a balance between the number of deployed sensors and estimation accuracy. Potential sensor locations are generated by sampling a continuous field of interest. We investigate the impact of sampling strategies on sensor placement, and show that compared to other commonly-used sampling strategies, the Poisson disk sampling method can provide a more accurate (discretized) representation of the random field. Based on the sampled locations, we propose an efficient placement algorithm that scales gracefully with problem size using the alternating direction method of multipliers and the accelerated gradient descent method. Numerical results are provided to demonstrate the effectiveness of our approach for sensor placement.