FastICA算法在雷达快扫描中的应用

Application of FastICA Algorithm in the Fast-scanning Radar

对于现代多普勒雷达来说,相关数据的计算精度依赖于估计单体内的独立样本数,独立样本数越多,精确度越大。为了增强雷达精确度和提高扫描速度,文中采用独立成分分析技术来增加独立样本数。独立成分分析算法通过迭代估计逐步逼近分离矩阵,从而得到所需的相对独立的因子。其中,快速不动点算法能在根本上显著提高计算方法的在线学习速度和可靠性,不仅收敛速度较快,而且收敛性不易破坏。利用基于极大似然估计的快速不动点算法先后对瑞利信号和模拟雷达信号进行了独立成分分析处理,并与常用的主成分分析技术进行了对比,获得了较好的试验结果,证实了独立成分分析技术的良好处理效果,为进一步应用于实际的多普勒雷达信号的处理奠定了基础。

To Modern Doppler radar, calculation accuracy of the data depends on the number of independent samples, the more in- dependent samples the more accurate in the calculation. In order to enhance the accuracy and improve the speed of radar scanning, in this paper, independent component analysis technique is used to increase the number of independent samples. ICA algorithm is on the basis of kurtosis, maximum likelihood estimation or other mathematical methods. It estimates separating matrix gradually through the iterative approximation so that the independent factors can be achieved. Among those algorithm, fast fixed-point algo- rithm can improve the speed of calculating by online learning and enhance its reliability. In this paper, a fast fixed point algorithm based on the maximum likelihood estimation algorithm is applied to rayleigh signals and simulative radar signals respectively and the results are compared with principal component analysis which is commonly used. The results of the texts prove a good perform- ance of ICA which fully verifys that ICA techniques could more accurately estimate independent signal from the mixed-signal, greatly increasing the number of independent samples to reduce the spectral estimation errors. This has laid the foundation for the practical application of Doppler radar signal processing.