01
Μαρ
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Προπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
ΓΕΩΡΓΙΟΥ ΠΕΡΑΚΗ
με θέμα
Εφαρμογές κατανεμημένου συμπερασμού σε ασύρματα δίκτυα αισθητήρων τροφοδοτούμενα με ισχύ από το περιβάλλον
Applications of distributed inference in ambiently powered wireless sensor networks
Εξεταστική Επιτροπή
Καθηγητής Άγγελος Μπλέτσας (επιβλέπων)
Καθηγήτρια Αικατερίνη Μανιά
Καθηγητής Θρασύβουλος Σπυρόπουλος
Abstract
This thesis contains applications utilizing wireless sensor networks (WSN) powered solely by the environment, using small, credit card-sized solar panels. The network contains a number of inexpensive terminals with transmission power of 10.4 dBm and token-ring medium access, capable of distributed in-network inference. An outdoor demonstration using loopy belief propagation and two indoor demonstrations using average consensus were developed and deployed, with this wireless network. The indoor demonstration includes two different versions; a centralised version focusing on robustness and a distributed counterpart focusing on available range. The indoor demonstrations calculate the average temperature in a closed space and the outdoor demonstration measures and estimates the soil moisture on an agricultural field; specific provisions are given so that soil moisture is estimated in locations where the sensors are broken or simply not reporting due to power outage. Distributed measurements and estimation for the outdoor demo required approximately 6 minutes of message passing between ambiently powered nodes; for the indoor demo, that time was approximately 3 minutes. It was found that distributed in-network inference with resource constrained, ambiently-powered wireless terminals is possible, at the expense of increased overall delay. In addition, it was found that distributed operation demands robust time synchronization among the terminals. Future work will focus on distributed time synchronization and other medium access control schemes for resource-constrained WSNs.