Antenna Array Processing: Autocalibration and Fast High-Resolution Methods for Automotive Radar.
Technische Universität, Darmstadt
[Ph.D. Thesis], (2012)
Available under Creative Commons Attribution Non-commercial No Derivatives, 2.5.
Download (1MB) | Preview
|Item Type:||Ph.D. Thesis|
|Title:||Antenna Array Processing: Autocalibration and Fast High-Resolution Methods for Automotive Radar|
In this thesis, advanced techniques for antenna array processing are addressed. The problem of autocalibration is considered and a novel method for a two-dimensional array is developed. Moreover, practicable methods for high-resolution direction-of-arrival (DOA) estimation and detection in automotive radar are proposed.
A precise model of the array response is required to maintain the performance of DOA estimation. When the sensor environment is time-varying, this can only be achieved with autocalibration. The fundamental problem of autocalibration of an unknown phase response for uniform rectangular arrays is considered. For the case with a single source, a simple and robust least squares algorithm for joint two-dimensional DOA estimation and phase calibration is developed. An identification problem is determined and a suitable constraint is proposed. Simulation results show that the performance of the proposed estimator is close to the approximate CRB for both DOA estimation and phase calibration. The proposed algorithm for phase autocalibration is extended for the case with multiple sources. Simulation results demonstrate that the proposed algorithm enhances the resolution performance in the presence of phase errors.
In automotive applications, modern driver assistance systems such as adaptive cruise control (ACC) or collision avoidance require an accurate description of the environment of a vehicle. For target localization in terms of range, relative velocity and DOA, a pulsed radar system with an array of receive antennas is considered. After pulse compression and Doppler processing, one obtains processing cells according to range and relative velocity, each represented by a single snapshot. In most cases, multiple targets can be distinguished by their range and/or relative velocity, so that each processing cell only contains a single target. However, there are situations, in which several targets are superposed in a processing cell. In the mentioned applications, this can occur in the presence of horizontal multipath with a close guardrail, which results in a ghost target. If the propagation paths cannot be resolved by conventional methods, this results in a false localization of the observed vehicle and high-resolution DOA estimation becomes necessary. The potential two-target model in the difficult case with a single snapshot is considered. An optimal generalized likelihood ratio test is applied, which involves the calculation of the computationally intensive maximum likelihood (ML) estimate of two targets. This approach provides good results with real data from experiments with a single and two corner reflectors. To achieve real-time capability, the computational cost has to be reduced substantially. Therefore, suitable criteria are presented to pre-select the processing cells, for which the ML estimator of two targets is necessary. When the targets are resolved in the spatial spectrum, the resulting DOA estimates are generally biased. For this case, a strategy for bias correction with low computational complexity is proposed. Results obtained from simulations and real data show that the performance of the developed algorithm is close to ML estimation, but at a significantly lower computational cost. When the spatial spectrum only shows a single significant peak, either a single target is present or two targets are unresolved. For this case, a computationally simple test is developed to decide whether the model with a single target is appropriate. Consequently, ML estimation of two targets is carried out only if the single-target model is rejected. This strategy is able to substantially save computations, when situations with more than one target per processing cell are unlikely. Finally, a practicable implementation for the ML estimator of two targets is developed, which is based on a simplified objective function and a delimited search range. The required projection operators are data-independent and can be pre-calculated off-line, which enables a trade-off between computational complexity and storage space. In simulations, the developed approach is shown to perform similarly to selected computationally efficient algorithms, but allows a straightforward and non-iterative implementation. The practical value of the proposed approach is further demonstrated using real data from a typical situation of an ACC application.
|Place of Publication:||Darmstadt|
|Classification DDC:||600 Technik, Medizin, angewandte Wissenschaften > 620 Ingenieurwissenschaften|
|Divisions:||18 Department of Electrical Engineering and Information Technology
18 Department of Electrical Engineering and Information Technology > Institute for Telecommunications > Signal Processing
|Date Deposited:||24 Aug 2012 09:58|
|Last Modified:||07 Dec 2012 12:05|
|Referees:||Zoubir, Prof. Dr.- Abdelhak M. and Yang, Prof. Dr.- Bin|
|Refereed:||6 June 2012|