超级军网中国军方专家在雷达领域有建树 填补国内空白
来源：百度文库 编辑：超级军网 时间：2024/04/28 22:08:36
中国军方专家在雷达领域有建树 填补国内空白
据军报报道：10月份，中国海军航空工程学院博士后关键主研人完成的雷达自动检测恒虚警率处理算法研究获军队科技进步一等奖，硕士研究生学位论文获全军优秀硕士论文。
报道称，
关键在雷达信号处理和多传感器分布式检测等领域形成了自己的独特研究方向，先后发表学术论文42篇，由清华大学学术专著出版基金资助的学术专著《雷达自动检测与恒虚警处理》，填补了国内空白。
报道透露，在清华攻读博士期间，他向国际雷达界的前沿研究方向发起冲击。由于该研究方向是解决雷达抗干扰、抗摧毁、抗隐身、抗低空所必需攻克的难题，理论研究难度大，创新点多，关键不畏艰辛，奋力攻关，所著论文被评为2002年“全国百篇优秀博士论文”。中国军方专家在雷达领域有建树 填补国内空白
据军报报道：10月份，中国海军航空工程学院博士后关键主研人完成的雷达自动检测恒虚警率处理算法研究获军队科技进步一等奖，硕士研究生学位论文获全军优秀硕士论文。
报道称，
关键在雷达信号处理和多传感器分布式检测等领域形成了自己的独特研究方向，先后发表学术论文42篇，由清华大学学术专著出版基金资助的学术专著《雷达自动检测与恒虚警处理》，填补了国内空白。
报道透露，在清华攻读博士期间，他向国际雷达界的前沿研究方向发起冲击。由于该研究方向是解决雷达抗干扰、抗摧毁、抗隐身、抗低空所必需攻克的难题，理论研究难度大，创新点多，关键不畏艰辛，奋力攻关，所著论文被评为2002年“全国百篇优秀博士论文”。
据军报报道：10月份，中国海军航空工程学院博士后关键主研人完成的雷达自动检测恒虚警率处理算法研究获军队科技进步一等奖，硕士研究生学位论文获全军优秀硕士论文。
报道称，
关键在雷达信号处理和多传感器分布式检测等领域形成了自己的独特研究方向，先后发表学术论文42篇，由清华大学学术专著出版基金资助的学术专著《雷达自动检测与恒虚警处理》，填补了国内空白。
报道透露，在清华攻读博士期间，他向国际雷达界的前沿研究方向发起冲击。由于该研究方向是解决雷达抗干扰、抗摧毁、抗隐身、抗低空所必需攻克的难题，理论研究难度大，创新点多，关键不畏艰辛，奋力攻关，所著论文被评为2002年“全国百篇优秀博士论文”。中国军方专家在雷达领域有建树 填补国内空白
据军报报道：10月份，中国海军航空工程学院博士后关键主研人完成的雷达自动检测恒虚警率处理算法研究获军队科技进步一等奖，硕士研究生学位论文获全军优秀硕士论文。
报道称，
关键在雷达信号处理和多传感器分布式检测等领域形成了自己的独特研究方向，先后发表学术论文42篇，由清华大学学术专著出版基金资助的学术专著《雷达自动检测与恒虚警处理》，填补了国内空白。
报道透露，在清华攻读博士期间，他向国际雷达界的前沿研究方向发起冲击。由于该研究方向是解决雷达抗干扰、抗摧毁、抗隐身、抗低空所必需攻克的难题，理论研究难度大，创新点多，关键不畏艰辛，奋力攻关，所著论文被评为2002年“全国百篇优秀博士论文”。
论文题目：多传感器分布式恒虚警率(CFAR)检测算法研究
作者简介：关键，男，1968年06月出生，1996年09月师从于清华大学彭应宁教授，于2000年01月获博士学位。
摘 要
多传感器分布式恒虚警率(CFAR)检测是CFAR处理和多传感器分布式检测的结合点。在非平稳杂波中，CFAR处理可以有效地抑制杂波强度增加引起的虚警概率的急剧上升，进而避免显示器画面饱和或后续数据处理设备过载。在强干扰中提取信号，不仅要求有一定的信噪比，而且要求检测器具有恒虚警率性能。因此，CFAR处理已经成为具有自动检测功能的现代雷达系统中必不可少的组成部分。在空间上分布的多个传感器构成的多传感器系统可以提高系统的生存能力、抗干扰能力(频率分集)、增加覆盖区域和监视目标数、提供更高的总的信噪比，并且可以提高系统的反应速度以及在单个传感器故障情况下的可靠性。近年来，利用多传感器的分布式信号处理格外受到重视。多传感器系统可以广泛应用在雷达组网、军事指挥、控制、通信、气象预报、医疗诊断、组织管理决策、机器人视觉，交通管制、遥感遥测等诸多领域。多传感器分布式CFAR检测兼有二者的优点，是当今国际雷达信号处理界的前沿研究方向和热点领域之一。
许多学者已经对CFAR处理和多传感器分布式检测这两个前沿领域分别进行了深入和广泛的研究。但是由于这两个领域的深度和广度，在它们的结合点上还有大量的问题尚未解决。本文深入研究了这方面的问题，首先从多传感器集中式CFAR检测入手，分析其存在的非CFAR问题，论述局部CFAR处理的必要性，提出有效的局部CFAR处理方法：局部检测统计量，证明基于局部检测统计量的分布式CFAR检测的最优性能上限，分析基于局部检测统计量的新方案在单脉冲条件下的性能，并且进一步拓展到多脉冲非相干积累的情况，然后引入反馈机制，最后利用不变检验的概念揭示CFAR处理的机理，从更高的层次上解释分布式CFAR检测做局部CFAR处理的必要性。
基于上述思路，本文主要研究了利用并行结构的分布式检测网，在高斯杂波背景中对SwerlingⅡ型起伏目标进行多传感器分布式CFAR检测的问题。有如下五个主要方面的内容：
(1) 研究了多传感器集中式CFAR检测存在的问题，提出了解决方案，论述了其必要性和最优性：
(1.1) 首先，得到了多传感器CFAR检测的CFAR处理应该在局部处理中完成的重要结论，这个结论是全文的基奠。对于集中式CFAR检测，以往认为直接利用已有的单传感器CFAR检测的结论就可以了。实际上，集中式CFAR检测存在着非CFAR的问题。本文在局部传感器背景噪声功率水平间的相对比例系数 变化情况下，推导出了集中式OS-CFAR检测在 且局部观测信噪比 条件下在非均匀杂波背景中检测性能的解析表达式。分析结果表明，集中式CFAR检测受 变化和失配的影响非常严重，既不能保持恒虚警率性能。因此，要做局部CFAR处理以形成具有CFAR性质的局部检测统计量。
(1.2) 其次，提出了一种新的局部多位量化方法。对于基于局部多位量化的分布式CFAR检测，一般则认为将基于软判决的分布式检测理论推广到分布式CFAR检测中即可。然而已有的基于软判决的分布式检测的最优化问题非常复杂，往往难以实现。因此需要寻求简单易行的、有效的、适用于CFAR检测的多位量化方法。本文提出了一种适用于分布式CFAR检测的基于检测单元采样秩的局部多位量化方法。新的多位量化方法获得了相对于二元量化方法明显增强的性能。
(1.3) 研究了多传感器分布式CFAR检测在Neyman-Pearson意义上的最优形式，确定了其检测性能的上限，为后续对新方法的性能分析提供了比较的基准。其中考虑了局部处理器检测单元采样间相关性的影响，讨论了基于局部检测统计量的分布式CFAR检测的必要性。
(2) 提出了三类新的局部检测统计量：R、S和P类，研究了基于这三类新的局部检测统计量的分布式CFAR检测的性能：
(2.1) 通过求和融合与最优融合的性能对比，确定了以求和融合做为对局部检测统计量的融合方案；推导出了基于上述三种新的局部检测统计量和求和融合的分布式CFAR检测方案：CA-R-SUM和OS-R-SUM、CA-S-SUM和OS-S-SUM以及Max-OS和Min-OS、Max-CA和Min-CA在 、 条件下在非均匀杂波背景中检测性能的解析表达式。
(2.2) 对于R类局部检测统计量还研究了在似然比(LR)融合和广义似然比(GLR)融合条件下的性能，提出了一种估计局部观测信噪比的方法，并构成了一种基于这个估计的新的GLR融合方法，利用这种融合的分布式CFAR检测方案具有与最优的LR融合方法极其相近的性能。
(2.3) 分析了上述方案在 、 以及 值失配等多种情况下的性能。结果表明：在 时，基于S类局部检测统计量的方案(如OS-S-SUM)是较好的选择；在 未知或变化的环境中，基于R类局部检测统计量的方案不受 的任何影响，因此是最好的选择。然而集中式CFAR检测和Min-OS受 变化或失配的影响非常严重。所以要求CFAR处理在局部处理中完成。若考虑数据通信量，R类和S类方案优于P类方案。
(3) 研究了分布式CFAR检测在多脉冲非相干积累条件下的性能：
(3.1) 推导出了分布式OS-CFAR检测在多脉冲视频积累条件下在非均匀杂波背景中检测性能的解析表达式。分析表明，脉冲积累数的增加使CFAR检测与最优固定阈值检测间的差别减小。因此，在多脉冲非相干视频积累条件下，多传感器分布式CFAR检测仍然是必要的。而且，脉冲积累数的增加使局部判决质量提高，利用更多的传感器可以明显改善对小信号的检测性能，因此更应该采用多传感器分布式检测。
(3.2) 提出了基于二元积累数(NBI)和基于检测单元采样秩的两类适用于多脉冲非相干积累条件下的多传感器分布式CFAR检测的局部检测统计量。对于基于检测单元采样秩的局部检测统计量，又提出两种在多脉冲条件下利用检测单元采样秩的方法：基于视频积累的秩(RVI)和基于秩的积累(IR)。推导出了评价上述几种方案检测性能的解析表达式。结果表明，基于NBI和基于RVI的方案均可获得分别相对于二元积累和视频积累的性能改善。基于IR的方案相对于二元积累的性能改善尤为明显。
(4) 研究了具有反馈机制的分布式检测，提出了一种将全局判决反馈到融合中心的新的反馈方案(FTFC)，推导出了FTFC的检测性能与反馈步数关系的解析表达式，证明了其检测性能收敛性的定理。相对于以往的将全局融合结果反馈到局部处理器(FTLP)的方案，FTFC方案不仅显著地降低了对反馈数据通信带宽的要求，而且在最小错误概率融合准则和Neyman-Pearson融合准则条件下的对比分析表明，FTFC仍具有与FTLP方案很接近的性能和收敛速度。
(5) 研究了不变检验与恒虚警率性质之间的关系，得到了利用不变检验概念证明恒虚警率性质的定理。讨论了在雷达自动检测中常见的不变统计量和不变分布，这包括了高斯杂波背景以及非高斯杂波，如Weibull、Log-normal和K分布杂波背景中的许多典型的CFAR检测算法。深刻地揭示了CFAR处理的机理，从新的角度和更高的层次上为多传感器CFAR检测的CFAR处理在局部处理中完成的必要性提供了新的解释和理论支持。
关键词：多传感器，分布式检测，集中式检测，CFAR，局部检测统计量，局部CFAR处理，多位量化，融合，多脉冲非相干积累，反馈，不变检验，Swerling II型起伏目标，高斯杂波，并行网络
Abstract
Distributed constant false alarm rate (CFAR) detection with multisensor is the combination of CFAR processing and distributed detection with multisensor. In condition of strong nonstationary clutter, CFAR processing can restrain sharp increase of PFA (probability of false alarm). It also can avoid screen saturation and the successive data processing overburdening. Therefore, not only a certain signal-to-noise ratio (SNR) is needed, but also the CFAR function is required so as to obtain a signal in strong interference. The CFAR processing is a necessary component in any modern radar system with automatic detection. The multisensor system composed of spatially distributed sensors can be improved in the aspects of survivability, anti-interference ability, reaction speed, and reliability in the case of sensor malfunction. It would also be advantageous to increase the covered region and the number of monitored targets, and to provide higher overall SNR of observations. In recent years, the multisensor system has spurred much attention. It can be widely used in the radar network, military command, control, communication, weather forecast, medical diagnosis, organization and management and decision, robot vision, traffic control, remote sensing etc.. The distributed CFAR detection possesses the advantages of both CFAR processing and the distributed detection with multisensor. It has become one of the most challenging research branches in the international field of radar signal processing.
Many researchers have studied respectively the CFAR processing and the distributed detection deeply and widely. Yet many problems remain unsolved in the field of their combination. This dissertation studied some of these problems in depth according to the following rationale. First, it begins with the study of the centralized CFAR detection with multiple sensors. The non-CFAR problem is analyzed; the necessity for local CFAR processing is established. The effective local CFAR processing scheme is proposed, i.e. the local test statistic (LTS). The upper limit of the optimal performance is proved for the distributed CFAR detection based on LTS. Then, the new schemes based on LTS are analyzed in condition of signal pulse, and then the analysis is expanded to the case of multiple pulses. Next, the feedback mechanism is introduced. At last, the invariant test is used to reveal the essence of CFAR processing. It explains the necessity of local CFAR processing in the distributed CFAR detection at a higher level.
According to the above train of thought, this dissertation considers mainly the case that the distributed CFAR detection in parallel topology is used to detect a Swerling II fluctuating target in Gaussian background. The five main sections are as follows:
(1) The problem in centralized CFAR detection with multiple sensors was studied. A feasible scheme and its importance were developed.
(1.1) First, it was concluded that CFAR must be done in local processing in multisensor CFAR detection. This conclusion is the basis of this dissertation. It was thought in the past that the theory of CFAR detection with single sensor could be used directly. In fact, there exists the non-CFAR problem in the centralized CFAR detection. Under the condition of varying , the ratio between noise levels in local sensors, this dissertation derives the analytic expression of performance of the centralized OS-CFAR detection for and local signal-to-noise ratio (SNR) in nonhomogenous background. Results show that the centralized CFAR detection is heavily influenced by the variation and mismatching of ; hence, the CFAR ability can’t be guaranteed. Therefore, the local CFAR processing must be used to generate the local test statistic with the CFAR property.
(1.2) Second, a new local multilevel quantization scheme was proposed. For distributed CFAR detection based on local multilevel decision, it was considered generally that the theory of distributed detection based on soft decision could be expanded to distributed CFAR detection. However, the optimization is very complex for the present distributed detection schemes based on soft decision and it is usually difficult to complement. Therefore, simple and effective multilevel quantization that is suitable for CFAR detection is need. This dissertation proposes a new local multilevel quantization scheme, which is suitable for distributed CFAR detection. It is shown to be superior to the common binary quantization.
(1.3) Third, the optimal form in the Neyman-Pearson sense of distributed CFAR detection with multiple sensors was studied. The upper limit of detection performance was determined. It provided a basis for comparison with the successive performance analysis, where, correlation between samples of test cells in local processors was considered. The necessity for distributed CFAR detection based on local test statistics (LTS) was discussed.
(2) Three new types of LTS were proposed, and the distributed CFAR detection based on these new LTSs was studied.
(2.1) After comparison of sum fusion and optimal fusion, sum fusion was selected as the fusion scheme for local test statistics. For the distributed CFAR detection based on the above three local statistics and sum fusion: CA-R-SUM and OS-R-SUM, CA-S- SUM and OS-S-SUM, Max-OS and Min-OS, Max-CA and Min-CA, their analytic expressions of detection performance were derived for and in nonhomogenous background.
(2.2) For the R type LTS, the likelihood ratio (LR) fusion and generalized likelihood ratio (GLR) fusion were studied. A method of estimating local SNR of observation is proposed, and a new GLR fusion was proposed based on this estimation. The distributed CFAR detection with this new GLR fusion performed as well as the optimal fusion.
(2.3) The above schemes were analyzed under the conditions of and and many mismatching cases of . Results show that the scheme based on the S type LTS is a better choice for (such as OS-S-SUM). In the environment of varying or unknown , the R type scheme isn’t influenced by at all. Hence, it is the best choice. However, the centralized CFAR detection and Min-OS are heavily influenced by the variation and mismatching of . Therefore, the CFAR processing should be done locally. If the communication bandwidth is considered, The R and S type schemes are better than the P type scheme.
(3) The distributed CFAR detection under the condition of noncoherent integration of multiple pulses was studied.
(3.1) The analytic expression of performance of the distributed OS-CFAR detection is derived for video integration of multiple pulses in nonhomogenous background. Results show that increasing the number of integrated pulses decreases the difference between CFAR detection and fixed threshold detection. Therefore, distributed CFAR detection is still necessary for noncoherent integration of multiple pulses. Moreover, the more number of integrated pulses improve the quality of local decisions, and the use of many more sensors can enhance the detection of small signals sharply. So multisensor distributed detection should be adopted.
(3.2) Two LTSs based on the number of binary integration (NBI) and the sample rank of the test cell are proposed; they are suitable for distributed CFAR detection with multiple pulses integration. Concerning the local test statistic based on the sample rank of the test cell, two schemes are proposed to use the sample rank of the test cell: the rank based on video integration (RVI) and the integration based on rank (IR). Their analytic expressions are derived. Results show that the two new schemes based on NBI and RVI are superior to the binary integration and video integration, respectively. The improvement of the scheme based on IR over the binary integration is very obvious.
(4) Distributed detection with feedback is analyzed. A new feedback scheme (FTFC, feedback to fusion center) is proposed. The expression of its detection performance and feedback step is derived. A theorem about its convergence property of detection performance is proven. Compared with the old FTLP scheme (feedback to local processor), the FTFC not only lowers the requirement about bandwidth of the feedback data communication, but also possesses the performance and convergence speed very similar to FTLP under the two fusion criterions of minimum error probability and Neyman-Pearson.
(5) The relationship between the invariant test and the CFAR property is studied. A theorem is founded about the proof of CFAR property using the concept of invariant test. The common invariant statistics and invariant distributions are discussed in radar detection background. It involves many typical CFAR detection algorithms in Gaussian clutter and non-Gaussian clutter, such as Weibull, Log-normal and K distributed clutter background. This part profoundly opens out the mechanism of CFAR processing. It provides a new explanation of and theoretical support for the necessity of the local CFAR processing in multisensor distributed CFAR at a new point of view and a higher level.
Keywords: multiple sensors, distributed detection, centralized detection, CFAR, local test statistic, local CFAR processing, multilevel quantization, fusion, noncoherent integration of multiple pulse, feedback, invariant test, Swerling II fluctuating target, Gaussian clutter, parallel network
作者简介：关键，男，1968年06月出生，1996年09月师从于清华大学彭应宁教授，于2000年01月获博士学位。
摘 要
多传感器分布式恒虚警率(CFAR)检测是CFAR处理和多传感器分布式检测的结合点。在非平稳杂波中，CFAR处理可以有效地抑制杂波强度增加引起的虚警概率的急剧上升，进而避免显示器画面饱和或后续数据处理设备过载。在强干扰中提取信号，不仅要求有一定的信噪比，而且要求检测器具有恒虚警率性能。因此，CFAR处理已经成为具有自动检测功能的现代雷达系统中必不可少的组成部分。在空间上分布的多个传感器构成的多传感器系统可以提高系统的生存能力、抗干扰能力(频率分集)、增加覆盖区域和监视目标数、提供更高的总的信噪比，并且可以提高系统的反应速度以及在单个传感器故障情况下的可靠性。近年来，利用多传感器的分布式信号处理格外受到重视。多传感器系统可以广泛应用在雷达组网、军事指挥、控制、通信、气象预报、医疗诊断、组织管理决策、机器人视觉，交通管制、遥感遥测等诸多领域。多传感器分布式CFAR检测兼有二者的优点，是当今国际雷达信号处理界的前沿研究方向和热点领域之一。
许多学者已经对CFAR处理和多传感器分布式检测这两个前沿领域分别进行了深入和广泛的研究。但是由于这两个领域的深度和广度，在它们的结合点上还有大量的问题尚未解决。本文深入研究了这方面的问题，首先从多传感器集中式CFAR检测入手，分析其存在的非CFAR问题，论述局部CFAR处理的必要性，提出有效的局部CFAR处理方法：局部检测统计量，证明基于局部检测统计量的分布式CFAR检测的最优性能上限，分析基于局部检测统计量的新方案在单脉冲条件下的性能，并且进一步拓展到多脉冲非相干积累的情况，然后引入反馈机制，最后利用不变检验的概念揭示CFAR处理的机理，从更高的层次上解释分布式CFAR检测做局部CFAR处理的必要性。
基于上述思路，本文主要研究了利用并行结构的分布式检测网，在高斯杂波背景中对SwerlingⅡ型起伏目标进行多传感器分布式CFAR检测的问题。有如下五个主要方面的内容：
(1) 研究了多传感器集中式CFAR检测存在的问题，提出了解决方案，论述了其必要性和最优性：
(1.1) 首先，得到了多传感器CFAR检测的CFAR处理应该在局部处理中完成的重要结论，这个结论是全文的基奠。对于集中式CFAR检测，以往认为直接利用已有的单传感器CFAR检测的结论就可以了。实际上，集中式CFAR检测存在着非CFAR的问题。本文在局部传感器背景噪声功率水平间的相对比例系数 变化情况下，推导出了集中式OS-CFAR检测在 且局部观测信噪比 条件下在非均匀杂波背景中检测性能的解析表达式。分析结果表明，集中式CFAR检测受 变化和失配的影响非常严重，既不能保持恒虚警率性能。因此，要做局部CFAR处理以形成具有CFAR性质的局部检测统计量。
(1.2) 其次，提出了一种新的局部多位量化方法。对于基于局部多位量化的分布式CFAR检测，一般则认为将基于软判决的分布式检测理论推广到分布式CFAR检测中即可。然而已有的基于软判决的分布式检测的最优化问题非常复杂，往往难以实现。因此需要寻求简单易行的、有效的、适用于CFAR检测的多位量化方法。本文提出了一种适用于分布式CFAR检测的基于检测单元采样秩的局部多位量化方法。新的多位量化方法获得了相对于二元量化方法明显增强的性能。
(1.3) 研究了多传感器分布式CFAR检测在Neyman-Pearson意义上的最优形式，确定了其检测性能的上限，为后续对新方法的性能分析提供了比较的基准。其中考虑了局部处理器检测单元采样间相关性的影响，讨论了基于局部检测统计量的分布式CFAR检测的必要性。
(2) 提出了三类新的局部检测统计量：R、S和P类，研究了基于这三类新的局部检测统计量的分布式CFAR检测的性能：
(2.1) 通过求和融合与最优融合的性能对比，确定了以求和融合做为对局部检测统计量的融合方案；推导出了基于上述三种新的局部检测统计量和求和融合的分布式CFAR检测方案：CA-R-SUM和OS-R-SUM、CA-S-SUM和OS-S-SUM以及Max-OS和Min-OS、Max-CA和Min-CA在 、 条件下在非均匀杂波背景中检测性能的解析表达式。
(2.2) 对于R类局部检测统计量还研究了在似然比(LR)融合和广义似然比(GLR)融合条件下的性能，提出了一种估计局部观测信噪比的方法，并构成了一种基于这个估计的新的GLR融合方法，利用这种融合的分布式CFAR检测方案具有与最优的LR融合方法极其相近的性能。
(2.3) 分析了上述方案在 、 以及 值失配等多种情况下的性能。结果表明：在 时，基于S类局部检测统计量的方案(如OS-S-SUM)是较好的选择；在 未知或变化的环境中，基于R类局部检测统计量的方案不受 的任何影响，因此是最好的选择。然而集中式CFAR检测和Min-OS受 变化或失配的影响非常严重。所以要求CFAR处理在局部处理中完成。若考虑数据通信量，R类和S类方案优于P类方案。
(3) 研究了分布式CFAR检测在多脉冲非相干积累条件下的性能：
(3.1) 推导出了分布式OS-CFAR检测在多脉冲视频积累条件下在非均匀杂波背景中检测性能的解析表达式。分析表明，脉冲积累数的增加使CFAR检测与最优固定阈值检测间的差别减小。因此，在多脉冲非相干视频积累条件下，多传感器分布式CFAR检测仍然是必要的。而且，脉冲积累数的增加使局部判决质量提高，利用更多的传感器可以明显改善对小信号的检测性能，因此更应该采用多传感器分布式检测。
(3.2) 提出了基于二元积累数(NBI)和基于检测单元采样秩的两类适用于多脉冲非相干积累条件下的多传感器分布式CFAR检测的局部检测统计量。对于基于检测单元采样秩的局部检测统计量，又提出两种在多脉冲条件下利用检测单元采样秩的方法：基于视频积累的秩(RVI)和基于秩的积累(IR)。推导出了评价上述几种方案检测性能的解析表达式。结果表明，基于NBI和基于RVI的方案均可获得分别相对于二元积累和视频积累的性能改善。基于IR的方案相对于二元积累的性能改善尤为明显。
(4) 研究了具有反馈机制的分布式检测，提出了一种将全局判决反馈到融合中心的新的反馈方案(FTFC)，推导出了FTFC的检测性能与反馈步数关系的解析表达式，证明了其检测性能收敛性的定理。相对于以往的将全局融合结果反馈到局部处理器(FTLP)的方案，FTFC方案不仅显著地降低了对反馈数据通信带宽的要求，而且在最小错误概率融合准则和Neyman-Pearson融合准则条件下的对比分析表明，FTFC仍具有与FTLP方案很接近的性能和收敛速度。
(5) 研究了不变检验与恒虚警率性质之间的关系，得到了利用不变检验概念证明恒虚警率性质的定理。讨论了在雷达自动检测中常见的不变统计量和不变分布，这包括了高斯杂波背景以及非高斯杂波，如Weibull、Log-normal和K分布杂波背景中的许多典型的CFAR检测算法。深刻地揭示了CFAR处理的机理，从新的角度和更高的层次上为多传感器CFAR检测的CFAR处理在局部处理中完成的必要性提供了新的解释和理论支持。
关键词：多传感器，分布式检测，集中式检测，CFAR，局部检测统计量，局部CFAR处理，多位量化，融合，多脉冲非相干积累，反馈，不变检验，Swerling II型起伏目标，高斯杂波，并行网络
Abstract
Distributed constant false alarm rate (CFAR) detection with multisensor is the combination of CFAR processing and distributed detection with multisensor. In condition of strong nonstationary clutter, CFAR processing can restrain sharp increase of PFA (probability of false alarm). It also can avoid screen saturation and the successive data processing overburdening. Therefore, not only a certain signal-to-noise ratio (SNR) is needed, but also the CFAR function is required so as to obtain a signal in strong interference. The CFAR processing is a necessary component in any modern radar system with automatic detection. The multisensor system composed of spatially distributed sensors can be improved in the aspects of survivability, anti-interference ability, reaction speed, and reliability in the case of sensor malfunction. It would also be advantageous to increase the covered region and the number of monitored targets, and to provide higher overall SNR of observations. In recent years, the multisensor system has spurred much attention. It can be widely used in the radar network, military command, control, communication, weather forecast, medical diagnosis, organization and management and decision, robot vision, traffic control, remote sensing etc.. The distributed CFAR detection possesses the advantages of both CFAR processing and the distributed detection with multisensor. It has become one of the most challenging research branches in the international field of radar signal processing.
Many researchers have studied respectively the CFAR processing and the distributed detection deeply and widely. Yet many problems remain unsolved in the field of their combination. This dissertation studied some of these problems in depth according to the following rationale. First, it begins with the study of the centralized CFAR detection with multiple sensors. The non-CFAR problem is analyzed; the necessity for local CFAR processing is established. The effective local CFAR processing scheme is proposed, i.e. the local test statistic (LTS). The upper limit of the optimal performance is proved for the distributed CFAR detection based on LTS. Then, the new schemes based on LTS are analyzed in condition of signal pulse, and then the analysis is expanded to the case of multiple pulses. Next, the feedback mechanism is introduced. At last, the invariant test is used to reveal the essence of CFAR processing. It explains the necessity of local CFAR processing in the distributed CFAR detection at a higher level.
According to the above train of thought, this dissertation considers mainly the case that the distributed CFAR detection in parallel topology is used to detect a Swerling II fluctuating target in Gaussian background. The five main sections are as follows:
(1) The problem in centralized CFAR detection with multiple sensors was studied. A feasible scheme and its importance were developed.
(1.1) First, it was concluded that CFAR must be done in local processing in multisensor CFAR detection. This conclusion is the basis of this dissertation. It was thought in the past that the theory of CFAR detection with single sensor could be used directly. In fact, there exists the non-CFAR problem in the centralized CFAR detection. Under the condition of varying , the ratio between noise levels in local sensors, this dissertation derives the analytic expression of performance of the centralized OS-CFAR detection for and local signal-to-noise ratio (SNR) in nonhomogenous background. Results show that the centralized CFAR detection is heavily influenced by the variation and mismatching of ; hence, the CFAR ability can’t be guaranteed. Therefore, the local CFAR processing must be used to generate the local test statistic with the CFAR property.
(1.2) Second, a new local multilevel quantization scheme was proposed. For distributed CFAR detection based on local multilevel decision, it was considered generally that the theory of distributed detection based on soft decision could be expanded to distributed CFAR detection. However, the optimization is very complex for the present distributed detection schemes based on soft decision and it is usually difficult to complement. Therefore, simple and effective multilevel quantization that is suitable for CFAR detection is need. This dissertation proposes a new local multilevel quantization scheme, which is suitable for distributed CFAR detection. It is shown to be superior to the common binary quantization.
(1.3) Third, the optimal form in the Neyman-Pearson sense of distributed CFAR detection with multiple sensors was studied. The upper limit of detection performance was determined. It provided a basis for comparison with the successive performance analysis, where, correlation between samples of test cells in local processors was considered. The necessity for distributed CFAR detection based on local test statistics (LTS) was discussed.
(2) Three new types of LTS were proposed, and the distributed CFAR detection based on these new LTSs was studied.
(2.1) After comparison of sum fusion and optimal fusion, sum fusion was selected as the fusion scheme for local test statistics. For the distributed CFAR detection based on the above three local statistics and sum fusion: CA-R-SUM and OS-R-SUM, CA-S- SUM and OS-S-SUM, Max-OS and Min-OS, Max-CA and Min-CA, their analytic expressions of detection performance were derived for and in nonhomogenous background.
(2.2) For the R type LTS, the likelihood ratio (LR) fusion and generalized likelihood ratio (GLR) fusion were studied. A method of estimating local SNR of observation is proposed, and a new GLR fusion was proposed based on this estimation. The distributed CFAR detection with this new GLR fusion performed as well as the optimal fusion.
(2.3) The above schemes were analyzed under the conditions of and and many mismatching cases of . Results show that the scheme based on the S type LTS is a better choice for (such as OS-S-SUM). In the environment of varying or unknown , the R type scheme isn’t influenced by at all. Hence, it is the best choice. However, the centralized CFAR detection and Min-OS are heavily influenced by the variation and mismatching of . Therefore, the CFAR processing should be done locally. If the communication bandwidth is considered, The R and S type schemes are better than the P type scheme.
(3) The distributed CFAR detection under the condition of noncoherent integration of multiple pulses was studied.
(3.1) The analytic expression of performance of the distributed OS-CFAR detection is derived for video integration of multiple pulses in nonhomogenous background. Results show that increasing the number of integrated pulses decreases the difference between CFAR detection and fixed threshold detection. Therefore, distributed CFAR detection is still necessary for noncoherent integration of multiple pulses. Moreover, the more number of integrated pulses improve the quality of local decisions, and the use of many more sensors can enhance the detection of small signals sharply. So multisensor distributed detection should be adopted.
(3.2) Two LTSs based on the number of binary integration (NBI) and the sample rank of the test cell are proposed; they are suitable for distributed CFAR detection with multiple pulses integration. Concerning the local test statistic based on the sample rank of the test cell, two schemes are proposed to use the sample rank of the test cell: the rank based on video integration (RVI) and the integration based on rank (IR). Their analytic expressions are derived. Results show that the two new schemes based on NBI and RVI are superior to the binary integration and video integration, respectively. The improvement of the scheme based on IR over the binary integration is very obvious.
(4) Distributed detection with feedback is analyzed. A new feedback scheme (FTFC, feedback to fusion center) is proposed. The expression of its detection performance and feedback step is derived. A theorem about its convergence property of detection performance is proven. Compared with the old FTLP scheme (feedback to local processor), the FTFC not only lowers the requirement about bandwidth of the feedback data communication, but also possesses the performance and convergence speed very similar to FTLP under the two fusion criterions of minimum error probability and Neyman-Pearson.
(5) The relationship between the invariant test and the CFAR property is studied. A theorem is founded about the proof of CFAR property using the concept of invariant test. The common invariant statistics and invariant distributions are discussed in radar detection background. It involves many typical CFAR detection algorithms in Gaussian clutter and non-Gaussian clutter, such as Weibull, Log-normal and K distributed clutter background. This part profoundly opens out the mechanism of CFAR processing. It provides a new explanation of and theoretical support for the necessity of the local CFAR processing in multisensor distributed CFAR at a new point of view and a higher level.
Keywords: multiple sensors, distributed detection, centralized detection, CFAR, local test statistic, local CFAR processing, multilevel quantization, fusion, noncoherent integration of multiple pulse, feedback, invariant test, Swerling II fluctuating target, Gaussian clutter, parallel network
有关文章。。。。。。
１９９６年９月免试进入电子工程系攻读博士学位的。时光飞逝，转眼间在清华已经度过了整整六年的光阴。在此期间，我获得了博士学位，博士学位论文还被评为全国优秀博士学位论文，博士后研究工作现在也将近尾声。高标准的要求、紧张忙碌的工作、学习和生活氛围使我受益颇多。
在导师彭应宁教授和何友教授的指导下，我选择了多传感器分布式检测作为研究方向，这是增强现代雷达系统四抗能力的关键技术，具有较强的前瞻性。我发现分布式检测中的恒虚警率ＣＦＡＲ处理还存在很大的可开发空间。国际上已发表的研究成果还很少，并且是限制在二元局部判决下的，这是制约分布式检测系统性能提高的一个瓶颈。实际上，具有高速通信能力的雷达网已经可以实现，在这种条件下改进局部处理器和融合中心的通信数据是提高系统检测性能的一条非常有效的途径。于是，我提出了基于局部检测统计量的分布式ＣＦＡＲ处理方法和三种算法，进行了大量的详细分析，并将研究结果发表在ＩＥＥＥ Ｔｒａｎｓａｃｔｉｏｎｓ ｏｎ ＡＥＳ 和欧洲Ｅｌｓｅｖｉｅｒ学会的Ｓｉｇｎａｌ Ｐｒｏｃｅｓｓｉｎｇ上，均被ＳＣＩ收录。此外，还利用不变检验理论研究了恒虚警处理，揭示了恒虚警处理的机理，研究结果也发表在ＩＥＥＥ Ｔｒａｎｓａｃｔｉｏｎｓ ｏｎ ＡＥＳ上，并被ＳＣＩ收录。博士学位论文答辩委员会认为这些研究成果具有重要理论和实用价值，对该学科领域的发展起到了一定的推动作用。
在清华学习和工作的六年经历锻炼了我多方面的能力，培养了更强的创新精神，给了我更多自信和勇气。我现在不仅能够独立开展科研工作，还作为主要负责人承担着国家自然科学基金课题等多项重要科研项目，在教学和科研等各方面工作中起着骨干力量的作用。
在攻读博士期间，我还与导师合作出版了一部获得清华大学学术专著出版基金资助的专著《雷达自动检测与恒虚警处理》。在该书的修改和出版过程中，我与已年逾七旬的陆大教授和清华大学出版社的王仁康老师接触很多，我被他们所展现的老一代清华人踏实认真、不求名利、默默无闻和甘于奉献的精神所感染，这也是我在清华的六年里在人格和修养方面的一个宝贵收获。
１９９６年９月免试进入电子工程系攻读博士学位的。时光飞逝，转眼间在清华已经度过了整整六年的光阴。在此期间，我获得了博士学位，博士学位论文还被评为全国优秀博士学位论文，博士后研究工作现在也将近尾声。高标准的要求、紧张忙碌的工作、学习和生活氛围使我受益颇多。
在导师彭应宁教授和何友教授的指导下，我选择了多传感器分布式检测作为研究方向，这是增强现代雷达系统四抗能力的关键技术，具有较强的前瞻性。我发现分布式检测中的恒虚警率ＣＦＡＲ处理还存在很大的可开发空间。国际上已发表的研究成果还很少，并且是限制在二元局部判决下的，这是制约分布式检测系统性能提高的一个瓶颈。实际上，具有高速通信能力的雷达网已经可以实现，在这种条件下改进局部处理器和融合中心的通信数据是提高系统检测性能的一条非常有效的途径。于是，我提出了基于局部检测统计量的分布式ＣＦＡＲ处理方法和三种算法，进行了大量的详细分析，并将研究结果发表在ＩＥＥＥ Ｔｒａｎｓａｃｔｉｏｎｓ ｏｎ ＡＥＳ 和欧洲Ｅｌｓｅｖｉｅｒ学会的Ｓｉｇｎａｌ Ｐｒｏｃｅｓｓｉｎｇ上，均被ＳＣＩ收录。此外，还利用不变检验理论研究了恒虚警处理，揭示了恒虚警处理的机理，研究结果也发表在ＩＥＥＥ Ｔｒａｎｓａｃｔｉｏｎｓ ｏｎ ＡＥＳ上，并被ＳＣＩ收录。博士学位论文答辩委员会认为这些研究成果具有重要理论和实用价值，对该学科领域的发展起到了一定的推动作用。
在清华学习和工作的六年经历锻炼了我多方面的能力，培养了更强的创新精神，给了我更多自信和勇气。我现在不仅能够独立开展科研工作，还作为主要负责人承担着国家自然科学基金课题等多项重要科研项目，在教学和科研等各方面工作中起着骨干力量的作用。
在攻读博士期间，我还与导师合作出版了一部获得清华大学学术专著出版基金资助的专著《雷达自动检测与恒虚警处理》。在该书的修改和出版过程中，我与已年逾七旬的陆大教授和清华大学出版社的王仁康老师接触很多，我被他们所展现的老一代清华人踏实认真、不求名利、默默无闻和甘于奉献的精神所感染，这也是我在清华的六年里在人格和修养方面的一个宝贵收获。
好，中国人的骄傲。敬礼！
能人还是有的！
支持一下。
仅仅是文章而已哦，文人习气不改么？
顶 直接copy搞成偶的毕业设计 嘎嘎
以下是引用五彩稻草人在2004-5-12 23:59:00的发言：
仅仅是文章而已哦，文人习气不改么？
不知不可轻言哟！
骄傲中！
值得高兴
zhe 这是每个中华儿女应该的
<B>以下是引用<I>judie</I>在2004-5-13 1:09:00的发言：</B>
顶 直接copy搞成偶的毕业设计 嘎嘎
<P>我们的毕设要写1.5万字呢,这点恐怕不够呵呵~~~~~</P>
支持！
没有泄密吧??
<P>为什么还要买捷克雷达》</P>
令人振奋的消息
看不懂，呵呵！
希望尽快实用化
好
呵呵，看着和天书一般，不过还是要顶
好啊[em03]
冷静啊,同志们
<B>以下是引用<I>judie</I>在2004-5-13 1:09:00的发言：</B>
顶 直接copy搞成偶的毕业设计 嘎嘎
<P>小心拿不到毕业证。
<P>好,好,好,</P><P>就是要自己搞,争取早些不要老毛子的</P>