Decentralized sensor selection based on sensor observations for energy efficient hypothesis testing

We consider the sensor selection problem in a wireless sensor network attempting to solve a binary hypothesis testing problem. The selection is based only on the sensor observations and the focus is on the extreme case where the position of the sensors is not exploited except through its influence on the sensor observations. Decentralized processing approaches are desired. A subset of sensors are selected to transmit their observations to a fusion center where the hypothesis testing decision will be made. We propose three new sensor selection schemes based on observed data. The first scheme, called optimum sensor selection (OSS), uses all sensor observations to compute the metric used to rank each candidate subset. The second scheme, called selection by averaging over unseen sensors (SAUS), uses only the observations of the candidate subset to compute the ranking metric. The third approach, called GSAUS, is a distributed greedy sensor selection scheme based on SAUS. The performance of each proposed scheme is evaluated by Monte Carlo simulation for a Gaussian shift-in-mean hypothesis testing problem so that a comparison between the various sensor selection schemes can be performed. The results indicate that proper distributed selection approaches can provide performance close to the optimum centralized selection approaches and significant improvement over random selection, an approach which has been suggested in the past. A particular approach called the ordered magnitude log-likelihood ratio (OLLR) approach, which was suggested previously for a different problem formulation, looks especially attractive.