摘要
针对复杂信道环境下的调制信号识别问题,文中提出了一种基于信号预处理与卷积神经网络的端到端识别算法。在信号预处理阶段,采用分层小波去噪与自适应滤波技术,有效抑制噪声并增强信号特征,同时通过鲁棒归一化与频域能量均衡提高信号质量。在时频特征提取环节,结合自适应窗长的短时傅里叶变换与快速小波变换,提升了时频图的分辨率与计算效率。在卷积神经网络模型的设计中,还引入了深度可分离卷积与频带注意力机制,降低了模型的复杂度,增强了模型对关键频段的特征提取能力。
For the problem of modulation signal recognition in complex channel environments,this paper proposes an endto-end recognition algorithm based on signal preprocessing and convolutional neural networks.In the signal preprocessing stage,hierarchical wavelet denoising and adaptive filtering techniques are used to effectively suppress noise and enhance signal features.At the same time,robust normalization and frequency-domain energy balancing are used to improve signal quality.In the time-frequency feature extraction step,short-time Fourier transform and fast wavelet transform with adaptive window length are combined to enhance the resolution and computational efficiency of the time-frequency diagram.In the design of the convolutional neural network model,deep separable convolution and frequency band attention mechanisms are also introduced,which reduce the complexity of the model and enhance the model̓s ability to extract features from key frequency bands.
作者
吴明仙
邢娟
WU Mingxian;XING Juan(Shangqiu University,Shangqiu,Henan 476000,China)
出处
《移动信息》
2025年第8期77-79,86,共4页
Mobile Information
关键词
信号预处理
CNN
调制信号
识别算法
Signal pretreatment
CNN
Modulating signal
Recognition algorithm
