This book introduces advanced sparsity-driven models and methods and their applications in radar tasks such as detection, imaging and classification. Compressed sensing (CS) is one of the most active topics in the signal processing area. By exploiting and promoting the sparsity of the signals of interest, CS offers a new framework for reducing data without compromising the performance of signal recovery, or for enhancing resolution without increasing measurements. An introductory chapter outlines the fundamentals of sparse signal recovery. The following topics are then systematically and comprehensively addressed: hybrid greedy pursuit algorithms for enhancing radar imaging quality; two-level block sparsity model for multichannel radar signals; parametric sparse representation for radar imaging with model uncertainty; Poisson-disk sampling for high-resolution and wide-swath SAR imaging; when advanced sparse models meet coarsely quantized radar data; sparsity-aware micro-Doppler analysis for radar target classification; and distributed detection of sparse signals in radar networks via locally most powerful test. Finally, a concluding chapter summarises key points from the preceding chapters and offers concise perspectives. The book focuses on how to apply the CS-based models and algorithms to solve practical problems in radar, for the radar and signal processing research communities.
Inspec keywords: radar detection; greedy algorithms; image enhancement; image sampling; image classification; synthetic aperture radar; radar target recognition; compressed sensing; image coding; radar imaging; image resolution; iterative methods; testing; Doppler radar; image representation; radar resolution
Other keywords: Poisson disk sampling; parametric sparse representation; wide-swath SAR imaging; locally most powerful test; radar networks; advanced sparse signal models; model uncertainty; advanced sparsity-driven models; radar target classification; radar imaging quality enhancement; high-resolution SAR imaging; hybrid greedy pursuit algorithms; coarsely quantized radar data; multichannel radar signals; two-level block sparsity model; radar applications; sparse signals distributed detection; sparsity aware microDoppler analysis
Subjects: Optical, image and video signal processing; Optimisation techniques; Radar theory; Signal processing and detection; Radar equipment, systems and applications; Interpolation and function approximation (numerical analysis); General electrical engineering topics
This book focuses on the application of sparsity-driven signal processing in radar tasks. In the rest of this book, we will present some advanced sparsity-driven models and methods in specific radar scenarios, instead of discussing the general case and the applications in other areas. Although the sparse signal processing has been studied for several decades, it is found that the direct applications of the basic models and algorithms based on the sparsity of signals to radar data may be less than optimal and even unsatisfactory. The reason is that radar signal processing algorithms are generally applicationdependent. We will first review the fundamentals of sparse signal recovery in this chapter and then present some advanced sparsity-driven models and algorithms specially designed for radar tasks in the rest of this book. The radar tasks discussed in this book include clutter suppression, signal detection, radar imaging, target parameter estimation, and target recognition.
Compared to convex optimization methods, the greedy algorithms are more computationally efficient and therefore more suitable for real-time radar processing. In this chapter, we will analyze the strengths and limitations of typical greedy algorithms such as orthogonal matching pursuit (OMP and subspace pursuit (SP) and then present two advanced greedy algorithms, that is, the hybrid matching pursuit (HMP and the look-ahead hybrid matching pursuit (LAHMP), for enhancing the quality of radar imaging. By combining the strengths of OMP and SP, the HMP algorithm ensures the orthogonality among the selected basis -signals like OMP and keeps the reevaluation of the selected basis-signals like SP. The LAHMP algorithm is an extension of HMP, based on embedding the look-ahead operation into HMP. The basis-signal selection strategy is refined in LAHMP by evaluating the effect of the basis signal selection at each iteration on the recovery error at subsequent iterations. Experimental results on real radar data demonstrate that these two advanced greedy algorithms can improve the concentration of dominant coefficients in the imaging result around the true target locations and the suppression of clutter and artifacts outside the target areas, at the cost of increased computational complexity.
In this chapter, we presented an advanced sparse signal model referred to as two level block sparsity model and introduced its applications in multichannel radar signal processing such as TWRI and STAP. By enforcing both the clustered sparsity of each single -channel signal and the joint sparsity pattern of the signals across all the channels, the two -level block sparsity model can help in clustering the dominant components and suppressing the artifacts. In the case of TWRI, the two level block sparsity model was directly applied to radar image formation in free space and through -wall scenarios. In the case of STAP, the two -level block sparsity model was utilized to fi rst reconstruct the angle -Doppler domain and then estimate CCM. The experimental results on simulations and real radar data have demonstrated the positive effect of the two -level block sparsity model on improving the quality of TWRI and enhancing the detection performance of STAP. The applications of the two -level block sparsity model can also be extended to other multichannel radar systems, such as multiple -input multiple -output (MIMO) radar, multistatic radar, and distributed radar. To further develop the two -level block sparsity model in various applications, it is worth studying more accurate formulation of the prior knowledge about the clustered structure and more simplified approaches for selection of model parameters.
In this chapter, a PSR method was presented to eliminate the negative effect of model uncertainty on radar imaging quality. A parameter vector formulating the model uncertainty is embedded into the basis -signal dictionary so that the dictionary is adjustable with the varying value of the parameter vector. Both the spare radar image and the parametric dictionary can be solved by the alternating iterations or the parameter searching scheme. The PSR method has been applied to SAR refocusing of moving targets, SAR motion compensation, and ISAR imaging of maneuvering aircrafts. Experimental results demonstrate that the PSR method is superior over some widely used algorithms in terms of the radar imaging quality at an affordable computational complexity. The model uncertainty considered in this chapter is mainly caused by the unknown relative motion between the radar and the target. It is worth investigating the applications of the PSR method in other radar tasks with model uncertainty. For example, in the case of through-wall radar imaging or ground penetrating radar imaging, the model uncertainty lies in the unknown characteristics of the propagation channel. By designing a proper parametric dictionary that is closely related to the propagation characteristics, the PSR is expected to be able to improve the quality of through-wall radar imaging and ground penetrating radar imaging. One may also extend the PSR method to the problems of radar detection and classification with model uncertainty, in which the parametric dictionary is expected to make the features corresponding to different hypotheses more separable.
Modern applications of synthetic aperture radar (SAR) demand both high spatial resolution and wide imaging swath. Unfortunately, traditional SAR imaging algorithms based on the Nyquist sampling theorem and the matched filtering can hardly achieve high-resolution and wide-swath simultaneously, since there is a trade-off between these two demands. Existing wide-swath SAR imaging systems usually adopt the flexible digital beamforming or waveform coding techniques with phased arrays, at the cost of increased system complexity. It is worth investigating how to achieve wider imaging swath of single-antenna SAR while keeping high resolution. If this solution is found, the imaging swath of the phased array SAR systems can be further increased.
In this chapter, we presented two algorithms, that is, the PQIHT algorithm and the E-BIHT algorithm, for enhancing the radar imaging quality with coarsely quantized data. The first algorithm, PQIHT, is basically the combination of the original QIHT algorithm and the PSR framework. It aims to eliminate the negative effect of the model uncertainty caused by the target motion and the quantization error on the SAR imaging quality. Experimental results demonstrate that the PQIHT algorithm can achieve moving target refocusing for coarsely quantized data and even 1 -bit data, with little sacrifice of the image quality compared to that generated from precise data. The second algorithm, E-BIHT, is based on the combination of the original BIHT algorithm and the two -level block sparsity model. Experimental results demonstrate that the E-BIHT algorithm can enhance the quality of radar imaging of stationary targets with 1 -bit data, by effectively removing the isolated artifacts and clustering the dominant pixels, at the cost of the increase in computational complexity. It is worth emphasizing that similar extensions can be also applied to other CS algorithms based on coarsely quantized data. The combination of the advance sparse signal models with 1 -bit CS algorithms has great potential in simplifying hardware implementation without severe degradation in radar imaging quality.
In this chapter, two sparsity-driven algorithms of micro-Doppler analysis were presented for radar classification of rigid-body and nonrigid body targets, respectively. The first algorithm aimed to accurately estimate the micro-Doppler parameters of a rigid-body target. A parametric dictionary, which is dependent on the unknown angular speed of the target, was designed to decompose the radar echo into several dominant micro -Doppler components. By doing so, the problem of micro-Doppler parameter estimation was converted into the problem of sparse signal recovery with a parametric dictionary. To avoid the time-consuming full search, the POMP algorithm was presented by embedding the pruning process into the OMP procedure. Simulation results have demonstrated that the POMP algorithm can yield more accurate micro -Doppler parameter estimates and better time - frequency resolution in comparison with some well-recognized algorithms based on WVD and Hough transform. The second algorithm, referred to as the Gabor- Hausdorff algorithm, was presented for micro -Doppler feature extraction and applied to radar recognition of nonrigid body targets such as hand gestures. Taking advantage of the sparse properties of radar echoes reflected from dynamic hand gestures, the Gabor decomposition was used to extract the time -frequency locations and corresponding coefficients of the dominant signal components. The extracted micro-Doppler features were inputted into modified-Hausdorff-distance based NN classifier to determine the type of dynamic hand gestures. Experimental results based on real radar data have shown that the Gabor-Hausdorff algorithm outperforms the PCA-based and the DCNN-based methods in conditions of small training dataset.
In this chapter, the problem of the detection of sparse stochastic signals with radar sensor networks was studied. The BG distribution was imposed on the sparse signals and accordingly the problem of distributed detection of sparse signals was converted into the problem of close and one-sided hypothesis testing. The original LMPT detector was presented to detect sparse signals from high -precision measurements without any requirement of signal reconstruction. Simulation results demonstrated that to achieve the same detection performance, the original LMPT detector has a much lower computational burden than the DOMP-based detector. We further presented the quantized LMPT detector to solve the problem of the distributed detection of sparse signals with quantized measurements. To ensure the optimal detection performance, a method for the design of the quantizers at the local sensors was presented. Theoretical analysis of the performance of both original and quantized LMPT detectors was consistent with the simulation results. Simulation results also demonstrated that (1) the 1 -bit LMPT detector with 3.3L measurements approximately achieves the same detection performance as the original LMPT detector with L high -precision measurements; and (2) the detection performance of the 3 -bit LMPT detector is very close to that of the original LMPT detector.
The motivation for writing the book came from an experiment, in which direct applications of some compressed sensing (CS) algorithms to the data collected by some real radar systems failed to achieve satisfactory performance. It was realized that something beyond the simple sparsity is necessary for signal processing in various radar tasks. Moreover, it was found that there were a number of books on the fundamentals of CS but few books were devoted to explaining how to apply the CS-based methods to solve the practical problems in the radar area. This book aims to introduce the advanced sparsity-driven models and methods that were designed for radar tasks such as detection, imaging, and classification, mainly based on the author's publications in the last decade. Besides the theoretical analysis, a number of simulations and experiments on real radar data were provided through this book to intuitively illustrate the effect of the advanced sparsity-driven models and methods.