A Data-centric and Statistical-based Random Sensing Scheduler for Rare Event Detection in Distributed Wireless Sensor Networks Lam-Ling Shum and Lionel Sacks , Department of Electronic and Electrical Engineering, University College London Abstract: This paper presents a sensing scheduler that learns about the environment monitored and adjusts its sensing behaviour according to the variation in the environment. The aim is to preserve energy by minimising sensing events in normal situation, but response quickly and reliably when a rare-event happens. Temporally, the scheduler uses statistical long-term and short-term averages to tune its sensing frequency. Spatial neighbour coordination is incorporated to enhance information dissemination across the network and minimise detection delay. 1. Introduction Data handling in distributed wireless sensor network for environmental monitoring has attracted a lot of attention in recent years. The challenge is that sensor networks are often distributed systems with limited resources including processing power, memory, and stringent power consumption requirement [1][2][3][4]. The works presented in this paper focus on the design of a sensing scheduler, which alternate the sensing behaviour according to some statistical properties measured in the environment in order to achieve power efficiency. A common type of sensor network application is rare-event detection. They have a similar characteristic that in majority of time the sensor networks are monitoring some normal conditions, which is not of interest of the users. When a rare event occurs, the sensor network is required to response to the event promptly and reliably and notifies the users. Examples of such applications are bridge collision monitoring, flooding detection, landslide warning system and forest fire detection [2]. Sensing of a rare-event is the trigger for further actions including reporting and network management in the sensor network and hence, we consider the work in this paper as a first step to develop a network management algorithm that fulfils the requirements of rare- event detection in sensor networks. Traditionally, temporal samples are taken in regular intervals governed by the Nyquist frequency to avoid aliasing. In the cases when more than one frequencies are of interest and they are wide apart in temporal scale, such as wave and tidal periods in Oceanography analysis [8][9], burst sampling technique may be adopted as a mean to conserve power. Regular or burst sampling techniques are designed for recording periodic events and may not be suitable for rare-event detection. It is nor efficient in terms of power consumption that sampling frequency can be much reduced when there is no event. On the other hand if sampling frequency is set to very low interesting events may be missed. Currently, most sensing schedulers being researched for sensor networks consider mainly redundancy in radio and sensing coverage and optimise network life by putting the covered nodes to sleep [6][7][8]. The problem is approached in a spatial aspect and temporal efficiency is not tackled. Moreover, sensing coverage can only be defined for a certain type of sensors, for example, cameras, ultrasound, etc, and is not applicable to point detection sensors for measurements such as temperature and pressure. We propose a sensing schedule that response to some statistical values measured from the environment, such that sampling is sparse when the environment does not vary much, and increases according to the variation in the environment. We also incorporate neighbour coordination to combat the problem of event-detection delay due to the sparse random sampling in normal condition. 2. Temporal Design The scheduler is based on a simple 2-states model as shown in Figure 1. p s and p i are the probability of the scheduler to change from one state to the other. At sense-state, the scheduler takes a sample 1 1 A sample is an illustrative use of the scheduler. Sense-state can be adapted to other temporal data collection methods, such as averaging of samples to minimise environmental noise and maximise power efficiency of the ADCs.