基于特征选择和支持向量机的数字调制识别方法

Digital Modulation Classification Based on Feature Selection and Support Vector Machines

针对基于决策树的数字调制识别方法在低信噪比和小样本情况下的不足,提出了一种改进的基于特征选择和支持向量机的数字调制识别算法。首先选择信号训练样本的循环谱截面作为备选特征集合,然后利用基于支持向量机的特征选择方法保留有效特征参数并训练分类器,最后将待识别信号选择后的特征输入支持向量机分类器,完成对ASK、MSK、PSK、QAM等4类信号的识别。仿真表明,本文算法在低信噪比和小样本情况下的识别性能优于基于决策树的调制识别方法。

An improved digital modulation classification algorithm based on feature selection and support vector machines is proposed to deal with the disadvantage of digital modulation classification based on decision trees under lower SNR and small samples. Firstly, some available cyclic spectrum sections of signal training samples are chosen as optional feature sets. Secondly, valid features are selected and support vector machine classifier with these features is trained. Finally, the valid fea- tures of signal testing samples are imported into the classifier and four modulation signals are classi- fied, such as ASK, MSK, PSK and QAM. Simulation result shows that the algorithm performs better than the algorithm based on decision trees under lower SNR and small samples.