Recently, various algorithms for radar signal detection that rely heavily upon complicated processing and/or antenna architectures have been the subject of much interest. These techniques owe their genesis to several factors. One is revolutionary technological advances in high-speed signal processing hardware and digital array radar technology. Another is the stress on requirements often imposed by defence applications in areas such as airborne early warning and homeland security. This book explores these emerging research thrusts in radar detection with advanced radar systems capable of operating in challenging scenarios with a plurality of interference sources, both man-made and natural. Topics covered include: adaptive radar detection in Gaussian interference with unknown spectral properties; invariance theory as an instrument to force the Constant False Alarm Rate (CFAR) property at the design stage; one- and two-stage detectors and their performances; operating scenarios where a small number of training data for spectral estimation is available; Bayesian radar detection to account for prior information in the interference covariance matrix; and radar detection in the presence of non-Gaussian interference. Detector design techniques based on a variety of criteria are thoroughly presented and CFAR issues are discussed. Performance analyses representative of practical airborne, as well as ground-based and shipborne, radar situations are shown. Results on real radar data are also discussed. Modern Radar Detection Theory provides a comprehensive reference on the latest developments in adaptive radar detection for researchers, advanced students and engineers working on statistical signal processing and its applications to radar systems.
Inspec keywords: radar antennas; radar detection; radar clutter
Other keywords: antenna beam; radar clutter; coherent processing interval; radar interference; radar detection
Subjects: Signal detection; Radar equipment, systems and applications; Single antennas; Radar theory
Radar systems are faced with the problem of discriminating the useful target echoes from the interference background, which hides the target making its detection difficult. Thus, the taskof a radar processor is to decide whether or not the collected returns contain useful signal components. This problem can be formulated in terms of a binary hypothesis test [13] H0 : data contain interference only' H1 : data contain useful target and interference, where1 H0 is referred to as the null hypothesis and H1 as the alternative hypothesis.
This chapter has considered classic radar detection in the presence of white Gaussian noise. The proposed approach first performs a compression of the observation space through reduction by sufficiency. Then detector design is attacked in the reduced dimensional space. The optimum NP detector is unfortunately not implementable for both the coherent and the noncoherent pulse trains. Thus, the robust GLRT-based approach is exploited to design practically implementable receivers that are canonical, namely independent of any prior assigned to the unknown parameters. This construction, depending on the coherency of the train, leads to either the classic coherent radar detector or linear and square-law integrators. Hence, a performance analysis of the obtained detection structures is conducted through analytic equations, where possible, or exploiting numerical simulations.
Coherent processing of various forms of multidimensional signals is commonplace in radar applications. Space-time adaptive processing in radars is a well-established example of coherent processing involving the domains of space (multiple receiving antenna elements separated spatially) and time (multiple pulse returns at each antenna element). The problem of detecting a subspace signal in a given test data vector can be formulated as a statistical hypothesis testing problem. An approach that has proven effective in dealing with nuisance parameters are invariant hypothesis tests. The general approach is to identify a set of matrices such that the linear transformation of the data by any member of the set leaves the original hypothesis testing problem unchanged, although the original nuisance parameters themselves are changed as a result. In this chapter, the author extended the three signal detectors above to a subspace signal model. Analytical expressions derived include results of signal mismatch errors. The analysis is applied to an example to illustrate the use of subspace detectors to mitigate detection loss resulting from signal mismatch errors.
This chapter has provided a survey on the two-stage detection of point-like targets embedded in homogeneous Gaussian disturbance. The family of two-stage detectors belongs to the more general class of tunable receivers, which allow to modify their directivity tuning proper parameters. They are obtained cascading two detectors with opposite behaviors in terms of directivity. The presence of a signal is declared if and only if each stage is above the respective threshold.
This chapter aims to discuss recent advances on Bayesian KA-STAP techniques. It unfolds as the classical STAP signal model in Section 5.2 evolves into a framework of KA-STAP model including a knowledge-aided homogeneous model, a knowledge-aided partially homogeneous model, and a knowledge-aided compound Gaussian model in Section 5.3. Then in Section 5.4, a hierarchical two-layered STAP model is discussed, which provides a new way to describe the non-homogeneity between the test and training data. Section 5.5 is devoted to parametric Bayesian detectors that integrate structural space-time information, i.e., a multi-channel autoregressive (AR) process, for the interference model and the consequent Bayesian estimation. The resulting Bayesian parametric detectors allow a fast implementation and further reduction in the amount of training data needed for reliable detection.
The problem of radar target detection in the background interference (plus noise) environment is the central problem in statistical radar theory. As a result, there are a number of well-established optimal (in the Neyman-Pearson sense) solutions for Gaussian signals and interference models with known interference covariance matrices, as presented in the first two chapters of this book.
As a radar system operates, it generally receives clutter returns from the environment that must be distinguished from targets of interest. If one assumes that the clutter returns obey complex multivariate Gaussian statistics, then a straightforward application of statistical detection theory leads to an optimal detector in the form of a well-known matched filter (Chap. 2 of this book). The occurrence of Gaussian statistics is often justified on the basis of the central limit theorem (CLT) applied to a phenomenological scattering picture that models the radar return as arising from contributions of a large number of scatterers in the radar resolution cell. In this case, the univariate intensity tail distribution is exponential. For early, low-resolution radars, this model was adequate.
As developed in this chapter, the detection performances are strongly linked to the covariance matrix estimation process. Several estimation methods have been studied through the statistical properties of the estimators. Then, they have been used in various detection problems on simulated data and real datasets. These results have enlightened the interest of using advanced estimation methods. Notice that there was not an exhaustive presentation of the different covariance matrix estimation approaches. Recently, to tackle the problem of few secondary data, as well as to deal with robustness matters, improved regularization techniques have been introduced [71-74]. One can also mentioned the promising framework of the Random Matrix Theory with some recent results in robust covariance matrix estimation for signal processing applications.
In this chapter, experimental data from a Ka-band high-resolution radar system are used to evaluate the performance of several schemes for coherent detection of distributed targets. The importance of this analysis relies on the possibility of exploiting detectors for range-spread targets in modern applications for homeland security. A careful border control is crucial to avoid illegal immigration or drug trade by means of very small boats (such as inflatable or wooden boats). It is thus of great interest to assess the capability of a high-resolution radar system to detect such kind of small targets which, at the resolutions of the data available in this chapter (0.10m, 0.20m, and 1 m), appear range-spread.