Research On Colocated MIMO Radar Signal Processing

In recent years, multiple-input multiple-output (MIMO) radar has been receiving increasing attention from researchers and engineers. MIMO radar is characterized by using multiple antennas to simultaneously transmit diverse waveforms and by utilizing multiple antennas to receive the reflected signals. The transmitted waveforms are commonly chosen to be orthogonal or linearly independent so that they can be extracted in the receiver by matched filtering. Compared with conventional radar systems, MIMO radar possesses significant potential advantages for target fading mitigation, resolution improvement, and interference suppression. Fully exploiting these potentials can lead to substantially enhanced target detection, parameter estimation, as well as target tracking and recognition performance. This thesis considers the many issues in colocated MIMO radar signal processing, including target parameter estimation, adaptive beamforming and space-time adaptive processing (STAP). The main contributions of this thesis are summarized as follows:1. By utilizing the special structure of the received signals, an iterative least square (I-LS) method is proposed for direction-of-arrival (DOA) estimation in monostatic MIMO radar system. This method optimizes two target steering matrices associated with the transmitting and receiving arrays, and the target's bearing can be easily obtained based on prior knowledge of the array geometries. Furthermore, by jointly using the transmitting and receiving array apertures, angular estimation accuracy can be significantly improved. Compared with conventional high-resolution DOA estimation methods, the I-LS method can directly obtain the DOA estimates without spectral peak searching, enhance the estimation accuracy by compressing the noise via dimension reduction, and reduce the computational burden remarkably without any eigendecomposition of the high-dimensional covariance matrix.2. A novel method based on multistage decomposition of the observed-data matrices is introduced for target localization and Doppler frequency estimation in bistatic MIMO radar. By utilizing the biorthogonality of matrices, a reasonable cost function is constructed. Via solving the cost function and a systematical multistage decomposition process, the two-dimensional angles and Doppler frequency of each target are estimated in turn, which can be paired automatically. Compared with the ESPRIT-based methods, the proposed method can simultaneously estimate the target location and Doppler frequency, eliminate the effect of the non-ideally orthogonal transmitted waveforms on target localization, and is also applicable to sensor arrays without an invariance structure.3. A new beamformer with a Kronecker product structure is developed for MIMO radar transmit-receive adaptive beamforming. The new method separates the two-dimensional Capon beamforming weight vector into a Kronecker product of the transmitting and receiving weight vectors. Therefore, the final long weight vector can be reconstructed via optimizing the two low-dimensional weight vectors. The fast convergence of the proposed iterative algorithm is proved. Since the dimension of the two separated weight vectors is significantly reduced, a better estimation of the observed-data covariance matrix can be obtained with a small number of training data samples. Hence the proposed Kronecker beamformer shows a substantially better performance than the sample-matrix-inversion (SMI) method in the case of short data record. Moreover, the inverse of the high-dimensional covariance matrix is avoided in our method, and therefore, the computational complexity is significantly reduced.4. The Kronecker product beamformer is further improved to result in a multistage Capon beamformer. It is shown that the fully adaptive Capon beamforming weight vector can be reshaped into a weight matrix. By performing the singular value decomposition on the weight matrix, the Capon beamforming weight vector can be expressed as the sum of Kronecker products for several pairs of singular vectors. We introduce a sequential way to solve for these singular vectors stage by stage, where a two-sided dimension reduction procedure is also adopted in each stage to successively compress the observed-data matrix for reducing the complexity. The multistage Capon beamformer has a good modular structure, due to which the beamformer can stop at any stage to naturally achieve rank reduction. Therefore, finding a low-rank approximation to the observed-data covariance matrix in conventional reduced-rank method is no longer required in the new method. From this point of view, the multistage Capon beamformer is the generalization of the Kronecker beamformer, which essentially carries out a rank-1 approximation to the weight matrix. Compared with the Kronecker beamformer, the multistage Capon beamformer has more degrees of freedom for signal processing, and therefore provides substantially improved performance on interference suppression.5. A simplified clutter eigencanceler for airborne MIMO radar is introduced. By using prior knowledge of the space-time distribution of the clutter, the clutter subspace is constructed off-line to replace the clutter subspace obtained via matrix eigendecomposition, and then the optimal weight vector is simplified as the product of a known projection matrix and the expected space-time steering vector. The complicated estimation and eigendecomposition of the covariance matrix are no longer required in the new method. Besides, the proposed beamformer not only eliminates the performance degradation caused by the inaccurate estimation of the covariance matrix, but also reduces the computational cost. Under the condition of ideal clutter model, the proposed method can provide near optimal clutter suppression performance.6. On the basis of the classic mDT algorithm, a computationally efficient two-stage reduced-dimension STAP method is developed. Firstly, the Doppler filtering is performed to reduce the data dimension in temporal domain. Secondly, the two-dimensional transmit-receive beamforming problem involved in mDT algorithm is divided into two one-dimensional beamforming problems, for which a bi-quadratic cost function is constructed. Finally, a bi-iterative algorithm for minimizing the bi-quadratic cost function is utilized to iteratively optimize the two short weight vectors. Because the weight vectors need to be solved has much lower dimension, particularly for small number of data samples, the two-stage reduced-dimension STAP method has substantially better clutter suppression performance and can be carried out at a smaller computational cost than the mDT method.7. By using the three-dimensional structure of the received data, a beam-Doppler space reduced-dimension STAP method is proposed for airborne MIMO radar. The received data in element-pulse domain is firstly transformed to the angle-Doppler domain using the two-dimensional spatial beamforming and temporal Doppler filtering. Then, the joint-domain adaptive processing is performed by using a group of three-dimensional beams around the angle-Doppler bin of interest. Because only the low-dimensional covariance matrix of the localized clutter needs to be estimated and reversed in our method, significantly reduction in the computational load and training requirement can be achieved.