Distributed Estimation In Wireless Sensor Networks

Wireless sensor networks have attracted increasing attention nationally and international-ly during the recent years. It can been deployed in a broad range of application areas includingmilitary fields, environment monitoring, industries and agriculture control, transportation,health care monitoring, fire rescure and other business application. As a thriving distributedcalculation platform, it consists of large amount of microsensors which work cooperatively toaccomplish some complex data processing and control tasks.These microsensors have been considered to have some intrinsic characteristics:limitations of power supply, storage limitation, calculation speed, sensing ranges, communica-tion ranges and measurement resolution, et al. These limitations will raise a set of challengingproblems to make the above applications practical. One of the most important challenges isthe distributed estimation problem in wireless sensor networks discussed in our dissertation.Distributed estimation in wireless sensor networks is a comprehensive study involvingsensor technology, wireless communication technology and signal processing techniques.Different from the traditional cable network distributed platform, wireless sensor networkswhen discussing distributed estimation problems have some unique characteristics:bandwidth-limited communicatioin channels, energy-limited microsensors and diversifiednetwork topologies et al. The dissertation will discuss distributed estimation problems inwireless sensor networks systematically, including the following works:1. Distributed estimation problems in bandwidth-limited wireless sensor networks areconsidered. Based on the assumption that channels between local sensors and the fusioncenter are modeled as independent binary symmetric channels, the maximum likelihoodestimator of an unknown parameter is proposed. To validate the availability of the estimator,the Cramér-Rao lower bound is also derived. The impact of the parameters of binarysymmetric channels on the estimation performance is considered. The results show that theperformance of the proposed estimatior can approach the Cramér-Rao lower boundmonotonously.2. Based on the assumption of additive white Gaussian noise channels, the impact ofchannel coding techniques on distributed estimation performance is analyzed. Local sensorsadopt two kinds of channel coding techniques: convolutional coding and rate-compatiblepunctured convolutional coding. Two novel kinds of power scheduling based on the abovetwo coding methods are derived for minimizing the total power consumption. Simulation results show that these two schemes are energy-efficient.3. The problem of distributed estimation in a wireless sensor network with an unknownobservation noise distribution is investigated. A modified indicator Kriging estimator isdeveloped. The tradeoff between estimation performance of the proposed estimator andenergy consumption of the network is formulated as an optimization problem. A global searchalgorithm which integrates a genetic algorithm and a simulated annealing algorithm isproposed to approximate the solution of the above optimization problem. Simulations showthe proposed search algorithm is convergent.4. The problem of decentralized estimation of a parameter in a cluster-based sensornetwork is studied. Two different scenarios are investigated: one is clustering with the samenumber of cluster members and the other is clustering with different number of clustermembers. Their maximum likelihood estimators and corresponding Cramér-Rao lowerbounds are derived, respectively. The results show that the estimation performance forclustering with the same number of cluster members is close to that for cluster-free sensornetworks and its energy consumption is largely less than that for cluster-free sensor networks.Simulation results indicate the tradeoffs between estimation performance of the proposedestimators and energy consumption of the networks exist in both scenarios.