Density estimation, hypotheses testing, and sensor classification in wireless sensor networks
In decentralized detection, the sensors first make a local decision before transmitting it to the fusion center (FC). The optimal design of the sensors' decision rule as well as the fusion rule requires knowledge of the probability distributions of the sensors' observations. This information, however, may not be available prior to deployment. Moreover, these probability distributions may vary over time or from sensor to sensor. Therefore in this paper we propose a method to estimate these distributions from the sensors' transmitted data. More specifically, we consider the problem of binary hypothesis testing in a wireless sensor network (WSN) consisting of multiple classes of sensors, where the sensors are classified according to the probability density function (PDF) of their observations under each hypothesis. The sensor nodes transmit their measurements to the FC which must detect the state of nature. In order to optimally fuse the data, the FC must also estimate the PDFs of the sensors' observations and classify the sensors. A method based on the Expectation Maximization (EM) algorithm is developed for: nonparametric estimation of the PDFs for each sensor class, classification of the sensors, and detection of the underlying hypotheses. Numerical results are presented to show that with a moderate number of samples from the sensors and three or fewer iterations of the EM algorithm, the proposed method performs remarkably well.
Publication Source (Journal or Book title)
Proceedings - IEEE Military Communications Conference MILCOM
Soltanmohammadi, E., & Naraghi-Pour, M. (2014). Density estimation, hypotheses testing, and sensor classification in wireless sensor networks. Proceedings - IEEE Military Communications Conference MILCOM, 637-642. https://doi.org/10.1109/MILCOM.2014.112