Dynamic Power Management In Wireless Sensor Network Based On Wavelet Decomposition

A new method for dynamic power management (DPM) in wireless sensor, based on wavelet, Kalman filter and AutoRegressive model, is proposed in this thesis. Firstly, we review the importance and status of DPM in wireless sensor. Then, the relation between DPM and the time when the next event arrive is analyzed. Based on wavelet, Kalman filter and autoregressive model, a method for DPM is presented. In this method, the power state of wireless sensor is partitioned into several states. The time slots between the time of one event occurring and the next is forecasted by using wavelet, Kalman filter and autoregressive, and the state that sensor should enter is decided according the forecasting result. The more accurate the forecasting result is, the lower energy the wireless sensor use, and the better the effect of DPM is.The data firstly is decomposed and reconstructed in each scale (frequency). Then, high frequency coefficient is predicted by Kalman filter and low frequency coefficient by AR module. According the forecasting result, wireless sensor may turn into some low power state. Wireless sensor node that satisfies the sleep condition for the deepest state enters sleep for the time slot, and this time slot is shorter than the forecasting result. So, the probability of missed events is low.Experimental results show that the system energy consumption of sensor, by using this method, can be reduced while the probability of missed events is low. It is achieved to maximize the sensor node's lifetime. Comparing with method of Sinha, the result of the DPM based on wavelet, Kalman filter and AR is better.