Effects of computing and communications on state fusion over long-haul sensor networks

We consider a sensor network where the state estimates are sent over long-haul connections from sensors to a fusion center to be correlated and fused to generate a global state estimate. The network loss and latency determine the state estimates received within a time-window at the fusion center, which in turn determine the quality Q_CC(τ) of the global state estimate computed within the allocated time τ. We derive a probabilistic performance bound on Q_CC(τ) as a function of the distribution of state estimates as well as the computation and network parameters, which shows a separation between the contributions of communications and computing parts. We then show that measurements based on field tests can be used to derive confidence bounds for Q_CC(τ). We complement the analytical results with component simulations and a system-level simulation of tracking a ballistic object to illustrate the qualitative effects of computing, including allocated computation time, and communications, including loss rates and transport protocols.