Feeling Sensors´ Pulse: Accurate Noise Quantification in Participatory Sensing Network

In the participatory sensing network, the sensor noise dominates the quality of sensing data as well as the processing efficiency. Previous works focus on evaluating sensing accuracy with expectations, and fails to quantify the sensor noise with variance estimations, which will inevitably suffer from the dynamics and the incompleteness of the sensing data. In this paper, we propose FSP (Feeling Sensors´ Pulse) method, which quantifies the sensor noise using the confidence interval. Specifically, we first use EM (Expectation Maximization) based iterative estimation algorithm to compute the maximum likelihood estimation (MLE) of sensor noise. Second, on the basis of these estimations, we leverage the asymptotic normality of MLE and the Fisher information to compute the confidence interval. The extensive simulations show that, FSP can achieve 90% success rate where the true values of sensor noise fall into the 95% confidence interval, at the cost of the polynomial time complexity only.