基于栈式稀疏自编码器的新型干扰识别

为了有效应对频谱弥散干扰(SMSP)和切片组合干扰(C&I)两种新型干扰,提出了一种基于栈式稀疏自编码器的识别算法。该算法首先对干扰与否的雷达接收信号进行双谱分析;然后对双谱特征进行降维,得到高维样本。预训练阶段,构造稀疏自编码器神经网络模型进行无标签样本的预训练;然后根据有标签数据对该模型参数进行有监督微调;最后利用Softmax分类器完成新型干扰的识别。仿真实验证明该方法有较高的识别率,特别是相较于其他文献方法,该方法受信噪比影响最小且识别效果最佳。说明了深度学习方法应用于雷达新型干扰信号识别领域的可行性和优越性。

Jamming Identification Algorithms of Advanced Jamming Based on Stacked Sparse Autoencoder

A novel jamming identification method based on stacked sparse autoencoder is proposed to effectively deal with two new jamming of smeared spectrum (SMSP) and chopping and interleaving(C&I). In this method,bispectrum analysis of received radar signal under three cases is given firstly, and after a series of dimensionality reduction,high-dimensional samples are obtained. In the phase of pre-training, network parameters are fine-tuned with label information. Next, stacked sparse autoencoder model is trained with unlabeled samples. Finally, the classifier is used to recognize the active jamming. The experimental results show that the proposed approach can achieve satisfying recognition. Particularly compared with other methods, the method is less affected by SNR and has a better result. Deep learning is feasible and superiority in the field of radar advanced jamming signal recognition.