Minimising Loss-Induced Errors in Real Time Wireless Sensing by Avoiding Data Dependency

The use of local processing to reduce data transmission rates, and thereby power and bandwidth requirements, is common in wireless sensor networks. Achieving the minimum possible data rate, however, is not always the optimal choice when the effects of packet loss on overall measurement error are considered. This paper presents a case study from the area of wireless inertial motion capture, in which the best distributed processing strategy is shown to be that which minimises inter-packet data dependency, rather than overall data rate. Drawing on this result and further analysis, we identify questions that should be raised during the design process in order to understand the effects of packet loss on distributed signal processing tasks, decide whether the errors are acceptable for an application, and choose appropriate techniques to mitigate them if necessary.