Sensor network localization based on natural phenomena

Abstract

Autonomous localization is crucial for many sensor network applications. The goal of this thesis is to develop a distributed localization algorithm for the PLUG indoor sensor network by analyzing sound and light sensory data from naturally occurring background phenomena as well as synthesized emulations of background transients. Our approach has two main phases: passive and active. The system enters an active mode when, its sensed region stays relatively silent and stable, hence assumed to be unoccupied; otherwise, it stays in the passive mode. In the passive mode, each node looks for sonic transients and compares the timing of its highest sound peak to that of synchronized sound peaks from other nodes in its neighborhood in order to estimate its distance. Passive ranging achieved 50.96cm error and simulated passive localization achieved 103.06cm error with a typical node-spacing of 2m. In addition, the system exploits background transients based on light sensory data to determine room boundaries. In the active mode, each node occasionally generates recorded mimics of natural sonic transients, like pencils dropping or water glasses clinking and manipulates an attached light source. Active acoustic ranging achieved 2.1cm error and simulated active localization achieved 7.97cm error with a typical node-spacing of 2m. In addition, passive location estimation in a real deployment is found to converge as more sensory data is available; range resolutions of 2.5m and localization errors of 20.3cm were obtained after running in passive mode for 20 hours in 7m by 5m dorm hallway. The main features of author's approach are its distributed properties, the lack of any heavy infrastructure, its unobtrusive exploitation of multi-sensory background phenomena, and in active mode, making the sound signal between nodes unobtrusive by mimicking the natural sounds.