On learning for fusion over fading channels in wireless sensor networks

In order to derive optimal/suboptimal fusion rules, in general, it is assumed that statistical properties of sensors´ decisions are known to a fusion center in distributed detection for wireless sensor networks. However, if sensors are deployed to unknown environments, these statistical properties may not be available in advance and should be estimated by the fusion center. To address this problem, in this paper, we study unsupervised learning to estimate the values of the parameters that characterize statistical properties for wireless sensor networks employing a bandwidth efficient multiple access scheme, e.g., the type-based multiple access (TBMA), over Rayleigh fading channels (which would be realistic channels when there is no line-of-sight between sensors and fusion center). Through simulations, we can show that unsupervised learning can be used in deriving decision rules at the fusion center from decisions transmitted by sensors over wireless fading channels.