Energy-Efficient Target Tracking through Wireless Sensor Networks. Cross-Layer Design and Optimization, Ph.D. thesis

In recent years, advances in signal processing have led the wireless sensor networks to be capable of mobility. The signal processing in a wireless sensor network differs from that of a traditional wireless network mainly in two important aspects. Unlike traditional wireless networks, in a sensor network the signal processing is performed in a fully distributed manner as the sensor measurements in a distributed fashion across the network collected. Additionally, due to the limited onboard resource of a sensor network it is essential to develop energy and bandwidth efficient signal processing algorithms. The thesis is devoted to discuss the state of the arte of algorithms commonly known as tracking algorithms. Although tracking algorithms have only been attracting research and development attention recently, already a wide literature and great variety of proposed approaches regarding the topic exist. The dissertation focus on statistical signal processing using Bayesian approaches for tracking targets in random motion. The first study leads to develop two simulation-based methods using sample theory and Monte Carlo (MC) realizations. Firstly, the concept of particle filter (PF) is developed and shown how it can be implemented in a bootstrap manner using sequential importance sampling/resampling (SIR) methods to perform statistical inference. Secondly, the unscented particle filter, to improve the SIR method, is developed and shown how it can be implemented using the unscented transformation and incorporating more observation than bootstrap sampling to perform more accurate estimates than PF. The second study leads to examine the efficiency of the particle filtering operating with limited energy. Indeed, particle filters have been extensively applied to track mobile targets in a wireless sensor networks, but nowadays their energy efficiency has received less attention. My thesis is that a cross-layer design, rather than the traditional layered approach, is required to prolong the lifetime of the sensor network in order to perform the tracking task. To save energy, different strategies have been adopted. Firstly, a cluster-based architecture for the sensor nodes is assumed. Hence, energy-based metrics to evaluate the energy consumption of the node selection algorithms in a cluster are introduced. Secondly, the node selection problem is formulated as a cross-layer optimization problem. Moreover, a greedy algorithm is proposed and shown to outperform the existing node selection algorithms. Finally, a trade off between the number of informative nodes and the accuracy of the tracking is drawn.