Research On Signal Fusion In Heterogeneous Sensor Network

Heterogeneous Sensor Network(HSN) consists of many heterogeneous sensor nodes with different detection performances, methods of communication and functions of monitoring. It has the characteristics of wide coverage, low energy consumption and can obtain various types of information about targets, therefore it has been widely used in target detection, location and tracking.In order to raise the detection probability of targets and improve the performance of location and tracking, some key technologies of information fusion in HSN are studied in this dissertation.(1) In order to solve the problem of fusion for single target detection in HSN with fading channels, an optimized energy allocation scheme is proposed, depending on the consumption of nodes’ performance heterogeneity. Four algorithms named OEA-LR(the optimized energy allocation- Likelihood Ratio), OEA-ALR(the optimized energy allocation- approximate Likelihood Ratio), EEA-LR(the equal energy allocation-Likelihood Ratio) and EEA-ALR(the equal energy allocation- approximate Likelihood Ratio) are proposed and analyzed via Monte Carlo simulations. The results show that OEA-LR and OEA-ALR perform better than EEA-LR and EEA-LR.(2) In order to solve the problem of fusion for multiple targets detection in HSN with fading channels, three fusion algorithms named LF-BR(likelihood function with the minimum Bayes risk),ALR-ML(approximate likelihood ratio with ML function) and LR-ML(likelihood ratio with ML function) are proposed. ALR-ML is an approximation of the LR-ML at low SNR. The simulation results show that the Bayes risk of the LR-ML is higher than that of the LF-BR, while the probability of detection of LR-ML is higher than that of LF-BR.(3) In order to solve the problem of trace point fusion for moving targets, three new fusion algorithms LN(Least Norm), TGML(Truncate Gaussian Maximum Likelihood)and GML(Gaussian Maximum Likelihood)are proposed. The three algorithms solve the problem of target position fusion under the condition of small system bias, truncate Gaussian distribution system bias and Gaussian distribution system bias, respectively. Simulation results show that the performance of these three algorithms and the number of local sensors are positively correlated. The performance of the GML and the variance of the system bias are negatively correlated. The size of bias range can not affect the performance of the TGML.(4) In order to solve the problem of moving target track generation, three interpolating algorithms, namely shape-preserving piecewise cubic Hermit interpolating, piecewise cubic spline interpolating and piecewise linear interpolating, are analyzed. Simulation results show that the shape-preserving piecewise cubic Hermit interpolating has the best conformality and feasibility among these three algorithms.