Model-based fault detection in powerline networking

Powerline networking is increasingly becoming an important component of home networking systems. Its reliability is however still a problem. We consider the popular X10 powerline communication protocol. Our main contribution is a model-based fault detection system that achieves completeness of coverage for X10 faults. We develop experimentally a finite state automaton that models all legal sequences of X10 commands. The task of detecting every violation of this model is complicated by the presence of hidden state and unobservable illegal transitions. This problem is addressed by deducing the model state indirectly from the sequence of X10 commands that is observed on the powerline. To this end, we formulate the model state deduction task in terms of the observability of the model, a concept which arises in discrete-event dynamic systems. Based on the observability property of our X10 model, the detection of model violations is performed in our current implementation via regular expressions on observable X10 command sequences. Our final contribution is a preliminary diagnosis procedure for identifying X10 faults.