联合深度学习和专家先验特征的信号调制识别

Signal Modulation Recognition Based on Joint Deep Learning and Expert Prior Features

自动调制识别在智能信号分析中起着关键作用,目前研究方法主要有专家先验特征和深度学习。这两者存在各自的优劣势,但难以有效结合实现优势互补,因此提出一种联合深度学习和专家先验特征的信号调制识别算法,提高复杂信道干扰下多种调制类别信号的识别准确率。将设计的神经网络代替决策树的分类判决门限,并结合输入的专家先验特征实现分层分类。实验表明,该算法性能要优于现有方法,在多径衰落信道下也可以取得较高的识别准确率。

Automatic modulation recognition plays a key role in intelligent signal analysis.At pres-ent,the research methods mainly include expert prior features and deep learning.Both of them have their own advantages and disadvantages,but it is difficult to effectively combine them to complement each other.Therefore,a signal modulation recognition method based on joint deep learning and ex-pert prior features is proposed to improve the recognition accuracy of multiple modulation signals un-der complex channel interference.The neural network is designed to replace the classification threshold of decision tree to realize hierarchical classification,combining expert prior features input.The experiment shows that the proposed method has better performance than the existing methods and can achieve high recognition accuracy in multipath fading channels.