Existence of self-similarity and repeated patterns in network traffic makes it possible to predict network parameters. This phenomenon can also be seen in behavior of wireless sensor networks. Self-similarity can be used to propose security solutions in computer and sensor networks. However, sensor networks suffer from several constraints such as limited energy, storage capacity and computing power which make it impossible to take advantage of an overwhelming proportion of tested, evaluated and improved mechanisms which are used in computer networks. Kalman filter is a mathematical tool for estimating control system status with low level of computational overhead. Considering network condition in the past time, Kalman filter can estimate network parameters in future. According to accurate and low cost estimations of Kalman filter, it is compatible with constraints of sensor networks. In this research we have used this concept to present an attack detection mechanism. To this end, Kalman filter make decisions base on traffic volumes produced by different nodes. It can be deduced from results that majority of attacks and abnormal conditions can be marked using the proposed mechanism.

Wireless sensor network Attack detection Long range dependency of data Kalman filter.