Sparse Sensing and Actuation in Stochastic Networks
Doty, Kyle Eldon
MetadataShow full item record
The two main ideas that are discussed in this dissertation are sparse sensing and actuation in stochastic networks. In the area of sparse sensing, the flag HMM is developed. The flag HMM is comprised of a structured observation process overlying an arbitrary finite-state Markov chain. The observations are such that a subset of states probabilistically emits distinct flags, while the other states are unmeasured. An explicit expression of the probability of error for a maximum likelihood state estimator is developed and used as the basis for a sensor placement algorithm. This sensor placement algorithm is then tested on randomly generated Markov chains, showing that accurate sensing can be achieved using only a few sensors. The flag HMM is then also used as the basis for sensor selection in a smart home. Many tests are run on a few smart homes, and the results are very similar to those of the randomly generated Markov chains. In the area of sparse actuation, a dynamic resource allocation (DRA) algorithm is developed for satellite communications based upon a PI controller. The DRA algorithm is designed to also allow for cognitive users. The cognitive users can proactively sense and fill any openings where data is not being sent. Multiple simulations are run with different PI gains as well as with and without the presence of cognitive users. Overall, the DRA algorithm performs well and the cognitive users have little adverse impact on the primary users. Simulations also show that the DRA algorithm performs well in the presence of selfish users and correlated traffic.