Introduction
The advent of small wireless sensor networks [1] initiated a series of military-oriented applications that can efficiently replace the currently used stratagems, which are generally more expensive to maintain and operate. Intrusion detection applications, such as base perimeter monitoring and borderline inspections, attempt to uncover physical threats in the form of biological or mechanical infiltration, whereby a commando unit or a vehicle attempts to slip into a base to do damage. Wireless sensors are greatly suited for these applications because of their small inconspicuous size, ease of deployment and low cost. Scattering sensors around a base is enough to create a virtual fence that detects unwanted visitors and alert the guards. But typical IDS applications apply monolithic mechanisms such as neural networks, data-mining and knowledge-based databases in order to identify and detect suspicious activities. This results in significant overhead in learning time, processing power, energy consumption, and requires large memory spaces, a challenge for its application to small sensor nodes. In this paper we propose a novel approach for applying a lightweight, yet robust IDS designed for wireless sensor nodes based on self-organized criticality (SOC) and hidden. Markov models (HMM) [2]. SOC is a recent concept to understand the internal interactions of complex systems (like nature) and large interactive systems always self-organize into a critical state governed by a power law [3]. So determining the SOC of a certain location vis-à-vis a specific variable (say temperature), can be applied to train the probability transition matrix of an HMM in order to sort out future anomalies. HMMs are good for modeling to predict systems. It has been used for detecting inconsistencies in handwritten letter recognition, and network-based (Internet) intrusion detection [4], [5]. We would like to add that we're the first to propose an IDS based on such a joint system.