Research On Signal Fusion Method Of Distributed Radar

Distributed radar is an important field and draws intensive attention for a long time.In early distributed radar,information fusion techniques such as track fusion and plot fusion were intensively studied.In these technologies,targets are detected by distributed radar sites.With rapid development of the communication technology,signal fusion based detection is possible to improve the detection performance better,which uses raw observations from distributed radar stations to detect potential targets.In theory,signal-level fusion detection can obtain the best detection performance,but it will lead to excessive communication between local radar stations and signal fusion center.In order to reduce the communication cost,quantization fusion detection algorithm is proposed which can quantize local observations while maintaining a high detection performance.Meanwhile local observations often have different Signal-to-Noise Ratios(SNRs),which may deteriorate the detection performance.To combat with above two challenges,we first study a method that weights local observations according to their SNRs.Research contents of this thesis are summarized as follows:1.Distributed detection algorithms with quantized and unquantized data are introduced.First,single-pulse and multi-pulse signal-level fusion detection algorithms for distributed radar are introduced.A signal model of monostatic radar is given and the statistical characteristics of the test statistic are derived.A distributed multisite radar signal-level fusion detection algorithm is derived.Signal-level fusion detection algorithms,such as Cell Average Constant False Alarm Rate(CA-CFAR)and Generalized Likelihood Ratio Test(GLRT),are analyzed in different scenarios.Second,distributed detection with quantized data is also studied.Several existing quantization algorithms are introduced to find good quantization thresholds and their detection performances are compared with signals fusion based algorithms.2.Distributed detection with heterogeneous data are studied in scenarios where channel SNRs may be different in practice.Several SNR weighting algorithms are presented,based on a priori information of detection performance curves and SNR estimates.The performances of distributed CA-CFAR and GLRT are analyzed in scenarios where local tests have different statistical distributions and different SNRs,indicating that they are difficult to reach the optimal point but the performance loss is often insignificant.3.Two computationally efficient quantization fusion detection algorithms are presented,one designed with intuitive concerns and the other based on a standardization transform.The intuitive quantization method is designed in line with a concept that the points on the boundary of optimal distributed decision have identical likelihood ratios.Therefore,we impose a constraint that boundary blocks after quantization have identical likelihood ratios.The other quantization is based on a likelihood ratio transformation and then will quantize the result for a minimal mean square error of the boundary blocks.Both methods enjoy a low computational cost and can efficiently fuse heterogeneous data with a high detection performance,the SNR loss is less than 0.25 db when the SNR estimation error is 3dB.