New Methodologies for Understanding Radar Data
2: Cognitive Radar Department, Fraunhofer Institute for High Frequency Physics and Radar Techniques (FHR), Germany
Research in the domain of radar signal understanding has seen interesting advances in recent years, mainly due to the developments around cognitive radar and the use of modern machine learning algorithms. This book brings together these strands of research into a coherent and holistic picture, presenting a consolidated approach to understanding radar signals. The book begins with an introduction, which provides some historical and philosophical context to developing methodologies for understanding radar signals, introduces new techniques, and outlines the book's approach to the topic. The book is then divided into three parts: the first focusing on statistical and conventional methods for interpreting radar data; the second addressing compressed sensing and cognitive methods for understanding radar data; and the third covering machine learning methods for understanding radar and remote sensing data. New Methodologies for Understanding Radar Data provides a complete, systematic guide to this multi-faceted topic for advanced researchers and professionals in radar engineering and signal processing.
Inspec keywords: learning (artificial intelligence); radar imaging; synthetic aperture radar; parameter estimation; radar signal processing
Other keywords: least squares approximations; oceanographic techniques; sea ice; radar signal processing; FM radar; synthetic aperture radar; radar imaging; radar data; learning (artificial intelligence); nonlinear estimation; parameter estimation
Subjects: General and management topics; Radar equipment, systems and applications; General electrical engineering topics; Simulation, modelling and identification; Machine learning (artificial intelligence); Optical, image and video signal processing; Signal processing and detection; Computer vision and image processing techniques
- Book DOI: 10.1049/SBRA542E
- Chapter DOI: 10.1049/SBRA542E
- ISBN: 9781839531880
- e-ISBN: 9781839531897
- Page count: 499
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
1 New methods for radar signal understanding
- + Show details - Hide details
-
p.
1
–13
(13)
We discuss the significance of the efforts towards understanding radar signals. Different parts of this endeavour have been termed using different phrases such as signal classification, scene classification, target recognition, parameter extraction. However, we can identify the common thread. All these efforts have one thing in common. They all help radar engineers interpreting the shadows in Plato's Allegory of the Cave! The research in the domain of radar signal understanding has seen some interesting new developments. This has been mainly due to the developments around cognitive radar and the use of modern machine learning algorithms. It was the editors' belief that an edited book to consolidate such efforts will be extremely beneficial. Maybe it is time to put all these pieces of research under one umbrella of 'radar signal understanding' to give a coherent and holistic picture. It was exciting to note that the IET appreciated this vision and many of our well-known colleagues consented to contribute chapters to this humble endeavour.
-
2 Understanding FM radar signal through parameter estimation
- + Show details - Hide details
-
p.
15
–46
(32)
This chapter is a short review through modern digital signal processing techniques used for frequency-modulated radar waveform parameter estimation. The theory related to novel algorithms developed in recent years is briefly described, with particular emphasis on their usefulness in radar, ER, and spectrum sensing applications. The main goal of the chapter is to present promising algorithms that serve to estimate parameters of radar signals with the assumption that the waveforms are of a completely unknown nature and the phase terms are not given. Primarily, the chapter is devoted to time-frequency (TF) signal processing methods which are usually dedicated to non-stationary waveforms, which radar signals are. However, the chapter also presents the matched filtering-based approach, which precisely estimates radar signal parameters enabling the understanding of data contained in the waveform. All the methods provided are supported with real-life signal analysis, and the outcomes are illustrated and described.
-
3 Non-linear estimation algorithms to analyse radar signals
- + Show details - Hide details
-
p.
47
–95
(49)
In this chapter, a description of different nonlinear estimation techniques, to enhance the estimation performance of the radar system, is given. We discuss some nonlinear adaptive estimation techniques along with the nonlinear extension of Kalman filter-based estimation techniques. The chapter proceeds with a description of the general radar signal model. Subsequently, the conventional batch processing estimation method is discussed. Next, the adaptive online nonlinear estimators and nonlinear estimators based on the Kalman filter are discussed. The chapter ends with a summary drawn from the discussion of the various adaptive online estimation techniques.
-
4 Understanding radar signals in a highly heterogeneous clutter environment
- + Show details - Hide details
-
p.
97
–114
(18)
Ground-penetrating radar (GPR) systems operate very often in environments with highly heterogeneous clutter sources and typically low signal-to-clutter ratios (SCRs). These clutter sources are generally divided, according to their origin, as surface and subsurface clutters. Surface clutter is related, particularly, to the reflection produced by the air-soil interface and also by objects located on the surface near the antennas. Subsurface clutter can be originated from all kinds of soil inhomogeneities and by other objects present in the targeted area. Besides dealing with low SCRs, GPR systems have generally limited bandwidth due to soil attenuation. This implies that the obtained resolution is usually not enough to differentiate between targets and clutter sources with similar characteristics. In this chapter, we will show examples as to how clutter affects GPR data and how multistatic acquisitions and polarimetry can be applied in GPR to improve detection and classification attempting to compensate these limitations in resolution and SCR.
-
5 Signature characteristics and discrimination of small airborne targets with a staring radar
- + Show details - Hide details
-
p.
115
–149
(35)
In this chapter, we briefly describe the staring radar concept and show how it can be used to detect and discriminate drones against a severe background of other targets such as birds and ground clutter. In particular, the time-varying responses from drones and competing targets are examined, highlighting inherent differences that can be exploited for discriminating between detected/tracked objects. The relationship between a staring radar mode of operation, target echo attributes and their exploitation for classification is examined. Throughout, examples from real-world measurements will be used to illustrate the form and dynamic nature of complex echoes as well as demonstrating the target-recognition performance with a supervised-learning-based approach.
-
6 SAR coherent change detection
- + Show details - Hide details
-
p.
151
–196
(46)
In recent years, many surveillance and reconnaissance applications have exploited the continued development of synthetic aperture radar (SAR) imagery techniques. In particular, the coherent change detection (CCD) technique has enabled the identification of subtle changes in a scene between two imaging passes. This chapter explores the topic of SAR CCD using physics-based derivations to assess various aspects of CCD performance in the default case of monostatic imaging over flat terrain. It then goes on to discuss the impact of the more complex imaging scenario encountered when the terrain is non-flat and introduces advanced processing schemes which can maintain optimum CCD performance in this situation.
-
7 Modern GPR target recognition methods
- + Show details - Hide details
-
p.
197
–252
(56)
In some humanitarian, commercial, and military applications, information on imaging, detection, and localization of shallow-buried targets is highly desirable. Often there is a necessity to sense and retrieve this information reliably and safely in, for instance, landmine recognition, archaeological excavations, planetary expeditions, and construction engineering. While a wide variety of technologies are available for subsurface exploration, namely radiometric, seismic, and electromagnetic (EM), only a ground penetration radar (GPR) provides non-invasive, safe, efficient, and high-resolution sensing. This advantage has led to significant advances in research on GPR acquisition and information retrieval for the past two decades.
-
8 Jet engine recognition via sparse decomposition of ISAR images
- + Show details - Hide details
-
p.
253
–284
(32)
Recognition and identification of targets is an important part of radar systems. Due to the high resolution, imaging radars are a well suitable choice for this task. This chapter presents a framework, which is based on sparse decomposition of radar images, to identify different kinds of scattering mechanisms. The aim of this framework is the identification of specific parts of certain targets. The basic procedure of the framework is shown exemplarily by a simulation using different canonical shapes, i.e. points, squares, circles, crosses and lines, which will be separated by the framework. As a practical application, the recognition of jet engines in radar images is presented. For this application, a scattering model based on waveguides is introduced and an example using data from the Tracking and Imaging Radar (TIRA) of Fraunhofer FHR is shown.
-
9 Waveform design for efficient radar signal recovery
- + Show details - Hide details
-
p.
285
–317
(33)
In this chapter, we focus on the design of transmit waveform and the resulting signal recovery. Waveform design plays a key role in determining the useful properties of the targets and environment. In particular, optimizing waveforms to have good correlation properties may improve the detection performance of weak targets. An emitted probing signal with low autocorrelation sidelobes maximizes the signal-to--interference-plus-noise-ratio (SINR), when complemented by a matched filter at the receiver, while significantly mitigating the signals from adjacent range bins. In the classical pulsed radar systems, a short code is transmitted per pulse repetition interval (PRI), followed by a long silence. In this case, the aperiodic auto-correlation function of the transmitted code should be optimized to have low peak sidelobe level (PSL) and integrated sidelobe level (ISL) for pulse compression. In the emerging phase-modulated continuous wave (PMCW) automotive radar applications, the radar sensor can have zero range sidelobes, provided that they use particular phase sequences with perfect periodic auto-correlation function. Additionally, waveform optimization is coupled with the advent of multiple input, multiple output (MIMO) radar systems which employ several probing signals at the transmitter; designing set of sequences with good correlation properties is an example. Furthermore, imposing practical constraints like unimodularity (or constant modulus) or finite or discrete phase (potentially binary) alphabet at the design stage enables the designs from the waveform optimization favourable for implementation. These constraints are important since the emergent applications tend to be power constrained necessitating power efficient operation of the sensing modules. A popular family of optimization frameworks that have found success in the design of radar waveforms considering these constraints is the coordinate descent (CD) method; its use in waveform design for efficient recovery will be discussed.
-
10 Sea ice concentration estimation techniques using machine learning: an end-to-end workflow for estimating concentration maps from SAR images
- + Show details - Hide details
-
p.
319
–337
(19)
Sea ice concentration (SIC) is an important metric used to characterize polar sea ice behavior. Understanding this behavior and accurately representing it is of critical importance for climate science research and also has important uses in the context of maritime navigation. An end-to-end workflow for generating learned concentration estimation models from synthetic aperture radar (SAR) data, trained on existing passive microwave (PMW) data, is presented here. A novel objective function was introduced to account for uncertainty in the PMW measurements, which can be extended to account for arbitrary sources of error in the training data, and a recent set of in situ observations was used to evaluate the reliability of the chosen PMW concentration estimation model. Google Colaboratory was used as the development platform, and all notebooks, training data, and trained models are available on GitHub. This chapter is an overview of the most interesting aspects of this investigation, and a detailed report is also available on GitHub.
-
11 Cognitive radar for ground moving target imaging
- + Show details - Hide details
-
p.
339
–388
(50)
In future decades there will be a growing demand for multi-function radar systems operating in resource-constrained and congested and contested spectrum environment. Due to the complexity of several missions, radar systems are required to perform multiple functions simultaneously, such as surveillance, clutter/interference cancellation, moving target detection (MTD), imaging and recognition. This leads to the development of multichannel, multi-static, multi-frequency and multi-polarization, in a single word, multi-dimensional, radar systems which enrich the received signal information content and increase the degrees of freedom that can be used to perform the above-mentioned radar tasks. Such systems need to share the spectrum with other concurrent and contrasting systems and, therefore, must be able to adapt to the environmental conditions. Moreover, to meet demanding requirements and to reduce costs, size and weight, modern radar systems should share the same hardware and software with other systems. In light of recent developments in the matter of cognitive radar, recent radar demonstrators and prototypes have started to embrace this new technology as it provides the best performances in dynamically changing environments. As it happens in the human brain, as well as in other mammals, a cognitive radar system learns about the environment through the interaction with it and acts consequently by transmitting optimal wave-forms and adopting optimal signal processing on receive. This process can be considered the core of a cognitive system. The concept of cognition has been introduced and extended to radar systems for the first time by Simon Haykin in 2006. One of Haykin's citations is emblematic of the concept of cognitive systems.
-
12 GAN4SAR: generative adversarial networks for synthetic aperture radar imaging of targets signature
- + Show details - Hide details
-
p.
389
–412
(24)
This chapter describes the generation of Synthetic Aperture Radar (SAR) images of man-made targets using modern deep learning methods called Generative Adversarial Networks (GANs). These relatively new techniques have shown impressive results in the optical image field, enabling, for instance, the generation of very convincing fake images, i.e. images that look very much like the ones the generative algorithm has been trained on. Thus, we wish to investigate how these methods can be used or adapted for similar data generation in radar imaging. We follow the progressive development path of GANs and explore some applications for SAR data augmentation and image domains transformations, while monitoring their expected performances for Automatic/Assisted Target Recognition (AsTR) purposes.
-
13 Micro-Doppler
- + Show details - Hide details
-
p.
413
–436
(24)
In this chapter, the fundamentals and the physical background of micro-Doppler signatures and their dependencies on radar modes of operation and target types are explained and illustrated by some examples.
-
14 Kinematic and linguistic interpretation of human motion via RF signal analysis
- + Show details - Hide details
-
p.
437
–467
(31)
This chapter provides an overview of machine learning for RF signal classification, especially as applied to human motion recognition. Current innovations that integrate a knowledge-aided, physics-enabled approach to DNN design are discussed in the context of recognition of activities of daily living. A newly emerging application of RF sensing to the study of American Sign Language (ASL) is also presented.
-
Back Matter
- + Show details - Hide details
-
p.
(1)