OTH radar has important applications in the field of low-altitude defense, warning far from the sea, detection for stealth targets; also is an important means of the sea environmental data collection and remote sensing of sea state. However, the frequency range of the HF radar is lower correspondingly, 3-30 MHz, leading to a wider beam width and poorer spatial resolution ability which affect its application scope seriously. In addition, the high-frequency radar has complex monitoring environment, strong external noise and interference, leading to the targets is not easy to be detected from strong noise background. In order to extract the target information from the radar echo signals, longer coherent accumlative time is required, even up to hundreds of seconds. These lead to less snapshots after accumlation for DOA estimation, which bring the huge challenge to signal processing methods. Super-resolution technology is the effective means of solving the low resolution of the radar, but the traditional super-resolution approaches are vulnerable to obsevation conditon constraints, such as the number of sensors, received signal multiple effects, coherent accumulative time etc. These factors could be the failure reasons of the traditional super-resolution. The emergence of compressed sensing provides a complete theoretical support for reconstruction of sparse signals, but also brings new vitality for spatial spect rum estimation theory. From the traditional subspace decomposition algorithm and compressed sensing theory to start, we establish a complete process flow to low resolution problem of HF radar, has made the following findings:1. An joint compressed sensing and subspace decomposition approximation algorithm is proposed to solve the problem of DOA estimation. There usually exists multipath interference and insufficient spatial sampling for HF radar received signals, and this will lead to the observed signals is non-full rank matrix. The traditional subspace decomposition algorithm can not solve this problem directly. Compressed sensing evades solving the second-order statistics of observed signal because of its unique recovery way, and copes with the coherent sig nals and insufficient snapshots problem effectively. However, it has poor sensitivity to multiple measurement vectors. In this paper, we proposed a joint approximated algorithm by virtue of stability of subspace decomposition approach and reconstruction process of compressed sensing method, which is applied for parameter estimation under any number of snapshots, and is robust to noise.2. A dictionay optimization algorithm is proposed for high atom correlation in observed matrix in low SNR. Super-resolution technology is to break the inherent theory limit. In order to estimate the parameters of the target as accurately as possible, we need the observed matrix constructed represents the characteristis of sparse signals accurately, therefore, more intensive grids are needed. For DOA estimation, however, dictionary atom itself has a high correlation, intensive grids furtherly deteriorate the structrual constrains on the observed matrix. Meanwhile, the high correlation will cause a great deviation of reconstruction algorithm and degrade the algorithm stability. In this paper, for parameter estimation of low SNR, we proposed a dictionary opimization algorithm, which can restrain noise effectively and reduce the accumulative correlation coefficient of observed matrix by construting a sensing matrix. In consequence, the convex optimization algorithm and greedy algorithm due to introducing the sensing matrix have a better performance, especially for close-space distance signals.3. A grid opimization algorithm is proposed for energy leakage due to poor grids. Compressed sensing method generally discretes a continuous domain signal to obtain the based vectors. If the non-zero elements of the signals exactly on the grids, the signal can be represented by the group of bases. However, in practical applications, the parameters of signal generally vary continuously, such as DOA approximation. The angle of targets is random, which can not accurately fall on our preseted grid points. In this case, there exists energy leakage(also called base mismatch). More serious problem is that the signal dissatisfies the sparse characteristic under the bases, then premise is not established for compressed sensing theory. This paper discusses two ways to optimize grids, which are discrete domain processing and continuous domain processing corresondingly. Based on the outstanding grids will let the approximated signal to be more sparse, we establish a iterative sparse optimization algorithm, which gives less estimated bias under less sensors compared to other classical grid opimization algorithms. |