An information-theoretic approach to the design of a distributed cascade of sensors

We consider a new approach to the design of a cascade of distributed sensors for the detection of a target when the individual sensors in the cascade have different measurement reliabilities. To obtain specific results, we consider a Rayleigh scattering target model and independent additive Gaussian noise at each sensor. Because of the intractability of the general solution in the case where we are free to order the sensors in the cascade, we introduce a new, and more analytically tractable, information-theoretic approach to the design of the distributed sensor cascade. We compare the performance of this approach to the classical optimal design for the simple two-stage cascade. We note that although the classical design is more robust to changes in the sensor order in the cascade, if the sensors in the cascade are ordered from least reliable to most reliable, the information-theoretic approach yields performance with only a negligible degradation over the optimal classical approach. The approach holds the potential for providing near optimal designs for large cascades with only a fraction of the design complexity.