要点一种基于二次特征提取的神经网络信号源个数估计方法.docVIP

一种基于二次特征提取的神经网络信号源个数估计方法( 张兴良* 王可人 樊甫华 （电子工程学院信息工程系，合肥 230037） 摘要：信源数目估计对阵列信号空间谱估计非常关键，但容易受到相关噪声环境和相关信号源的影响。为此，提出一种基于二次特征提取的源数估计算法。首先，利用阵列信号协方差矩阵的特征值和特征向量，提取6组二次特征参数；然后，利用这些参数对神经网络进行训练；最后，利用训练好的神经网络进行信源数目估计。由于稳健性强，该算法非常适合在复杂电磁环境下使用。仿真试验结果表明，该算法在低信噪比、相关噪声和相关信号源条件下均具有良好的估计性能。因此，该算法应用前景广阔。 关键字：空间谱估计；信源数目；神经网络；特征提取 中图分类号：TN911.23 文献标识码：A Detection of the Number of Sources Based on Secondary Feature Extraction Using Neural Network Zhang Xingliang Wang Keren Fan Fu-hua (Department of Information Engineering, Electronic Engineering Institute, Hefei 230037, China) Abstract: Detection of the number of sources is extremely essential to spatial spectrum estimation in array signal processing. However, it is vulnerable to the correlated noise and the coherent sources fields. Aiming at resolving above problem, a new algorithm for detection of the number of sources based on secondary feature Extraction is proposed. Firstly, six sets of secondary feature parameters are calculated with the eigenvalues and the eigenvectors of the array signal covariance matrix. Then, these parameters are used to train neural network. Last, the number of sources is detected by the trained neural network. The new algorithm is suitable to work in complex electromagnetic environment, for it has high robustness. The results of simulation experiments show that the new algorithm has good performance in the low signal-to-noise ratios, the correlated noise and the coherent sources fields. Therefore, the new algorithm has a wide application prospect. Key words: Spatial spectrum estimation; Number of sources; Neural network; Feature extraction 1 引言 20世纪70年代以来，以MUSIC算法[1]为代表的子空间类算法突破“瑞利限”的限制，实现了谱估计的重大目前已广泛应用雷达、通信和等诸多领域，优点早已在实践中得到证明。算法Wu对此进行了改进[10]。文献[11]通过对角加载数据协方差阵减轻相关色噪声对信号源数目估计的影响，但加载量的理论最优值在实际工程中一般难以确定。文献[12]提出的特征值聚类算法具有一定的普适性，但该算法没有对噪声进行去相关处理，因此对高斯白噪声背景的估计效果最佳，对空间相关色噪声背景的估计效果改进不明显。信号源数目估计的另一难题是相关信号源入射的估计问题，当入射信号相关时会导致阵列信号协方差矩阵的特征值和特征向量发生“混淆”，甚至“少秩”，即其秩比信号不相关时要少。相关信号源数估计的方法[13-15]主要有空间平滑处理法、