As one of the main sensor in autonomous driving, radar has great advantages over other sensors, especially its capabilities during adverse weather condition and Doppler information extraction. Performance of the radar in terms of accuracy and target resolution strongly depends
...
As one of the main sensor in autonomous driving, radar has great advantages over other sensors, especially its capabilities during adverse weather condition and Doppler information extraction. Performance of the radar in terms of accuracy and target resolution strongly depends on radar waveforms transmitted and signal processing algorithms applied. To achieve high range resolution, an ultra-wideband (UWB) signal has to be used for sensing, which introduces difficulties to achieve high Doppler and direction-of-arrival (DOA) estimation simultaneously due to the range migration. To address this problem, in this thesis new signal processing algorithms are proposed, which pave the way to improved performance of the automotive radar sensor. As the frequency-modulated continuous-wave (FMCW) radar are widely used in short-range and middle-range applications due to its low cost and simplicity, FMCW waveform is the main research subject. The FMCW signal model is derived and analysed in Chapter 2 which for the first time takes both the range migration and wideband DOA problems into account at the same time. The point-like moving targets are considered in Chapter 3 where their Doppler velocities are within the maximum unambiguous velocity of the radar. A novel improved multiple signal classification (MUSIC) algorithm with the dynamic noise-subspace method is proposed to address both the range migration and wideband DOA problems. The algorithm releases the great potentials of the conventional MUSIC algorithm in the presence of the range migration. Moreover, an efficient algorithm based-on Rayleigh-Ritz step is introduced for the proposed method resulting in a considerable reduction of computational requirements without any performance degradation. Comparison with the conventional narrow-band MUSIC, Keystone-MUSIC, inversion-MUSIC and corresponding Cramer-Rao bounds (CRB) using simulations, reveals the superiority of the method proposed in terms of accuracy, resolution and efficiency. The problems similar to those considered in Chapter 3 but in the presence of the Doppler ambiguity are considered in Chapter 4. A spectral norm-based algorithm is proposed to address the coupling terms for a single moving point-like target. The algorithm for the first time abandons the integration-based method for ambiguous velocity estimation. The spectral-norm based algorithm provides a new tool to resolve the ambiguity problem which outperforms the conventional integration-based algorithm by avoiding the off-grid problem with limited data size. Moreover, combined with the modified CLEAN techniques and Greedy algorithm, the proposed algorithm can be extended to multiple moving targets. Furthermore, the power iteration algorithm is smartly adopted for an efficient implementation of the proposed method. After addressing the point-like targets, the moving extended targets are studied in Chapter 5 especially when multiple extended targets cannot be separated both in range and beam profile. The Doppler difference is used to recognise them and inverse synthetic aperture radar (ISAR) concept is adopted to split and image the targets separately. The conventional entropy minimisation approach is applied to the signal model for not only the Fourier spectrum but also the eigenspectrum as well for the first time. The Fourier spectrum has a relatively high resolution in higher-order motion (e.g. acceleration) while eigenspectrum has a better resolution in Doppler separation. The advantages of both spectra are utilised to separate multiple extended targets by a simple but powerful combination. Via numerical simulation, the applicability of the algorithm in the automotive application is demonstrated. Last in Chapter 6, by processing the experimental data from automotive radar, we present a novel and fast imaging algorithm for slow-moving targets which provides super-resolution on DOA. The range information is processed via fast Fourier transform (FFT) for efficiency while the DOA is estimated by the MUSIC algorithm for super-resolution. Since the MUSIC spectrum is pseudo-spectrum and can not represent the correct dynamic range of the imaging results, a novel normalisation method is introduced to vividly indicate the energies of different targets. In comparison with conventional FFT-BF, a cleaner range-azimuth image is obtained with the proposed algorithm demonstrating higher angular resolution and without strong side-lobes. Although the research presented in this thesis is served for automotive application, some of the algorithms and ideas can be easily generalised for a broad spectrum of diverse applications. @en