Wang Shoubin, Li Huan, Zhang Xiaolong, Jiang Hao, Shen Lei
The 36th Research Institute of China Electronics Technology Corporation, Jiaxing 314033, China.
College of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China.
Sensors (Basel). 2025 Jun 27;25(13):4007. doi: 10.3390/s25134007.
The existing multi-carrier composite modulation recognition methods have failed to effectively integrate inner and outer modulation characteristics, thereby limiting the potential for improving recognition performance under low signal-to-noise ratio (SNR) conditions. To address this issue, this paper proposes a multi-carrier composite signal modulation recognition algorithm based on a multi-dimensional time-frequency superimposed spectrum (MD-TFSS) with integrated inner and outer features, which can recognize composite modulation signals in the set {BPSK-PM, QPSK-PM, BPSK-QPSK-PM, BPSK-BPSK-PM, QPSK-QPSK-PM}. The proposed method constructs a dual spectrum through multiplying an inner modulation spectrum and a squared spectrum, then combines the inner modulation dual spectrum with the outer modulation time-frequency diagram in dual-channel mode to form MD-TFSS features. Based on the MD-TFSS, a blind recognition algorithm is implemented using the dual-channel input ECA-ResNet18 (DECA-ResNet18) incorporating the ECA attention mechanism. The proposed algorithm first converts the complex features of multi-carrier composite modulation signals into visually interpretable image features (including the quantity and concentration of bright spots and lines) through the MD-TFSS, achieving intuitive representation of multiple modulation characteristics. Meanwhile, the dual-channel input mechanism enables collaborative expression of outer modulation time-frequency diagram and inner modulation dual spectrum features, ensuring tight integration of inner and outer characteristics while avoiding feature isolation issues in traditional multi-diagram concatenation methods. Secondly, the DECA-ResNet18 network dynamically allocates weights through an adaptive regulation mechanism based on input feature differences, autonomously adjusting channel attention levels to effectively capture complementary characteristics from both inner and outer modulation features, thereby enhancing recognition accuracy and generalization capability for multi-carrier composite modulation signals. Theoretical analysis and simulation results demonstrate that, compared with the existing methods that use isolated outer and inner features or conventional multi-feature diagram construction approaches, the proposed algorithm achieves superior recognition performance under low SNR conditions. Additionally, DECA-ResNet18 demonstrates enhanced recognition performance for multi-carrier composite modulated signals compared to the traditional ResNet18.
现有的多载波复合调制识别方法未能有效整合内外调制特性，从而限制了在低信噪比（SNR）条件下提高识别性能的潜力。为解决这一问题，本文提出了一种基于多维时频叠加谱（MD-TFSS）的多载波复合信号调制识别算法，该算法集成了内外特征，能够识别集合{BPSK-PM、QPSK-PM、BPSK-QPSK-PM、BPSK-BPSK-PM、QPSK-QPSK-PM}中的复合调制信号。该方法通过将内调制谱与平方谱相乘构建双谱，然后以双通道模式将内调制双谱与外调制时频图相结合，形成MD-TFSS特征。基于MD-TFSS，利用结合了ECA注意力机制的双通道输入ECA-ResNet18（DECA-ResNet18）实现了盲识别算法。所提算法首先通过MD-TFSS将多载波复合调制信号的复数特征转换为视觉上可解释的图像特征（包括亮点和线条的数量和浓度），实现多种调制特性的直观表示。同时，双通道输入机制使外调制时频图和内调制双谱特征能够协同表达，确保内外特征紧密整合，同时避免传统多图拼接方法中的特征孤立问题。其次，DECA-ResNet18网络通过基于输入特征差异的自适应调节机制动态分配权重，自主调整通道注意力水平，以有效捕捉内外调制特征的互补特性，从而提高多载波复合调制信号的识别精度和泛化能力。理论分析和仿真结果表明，与现有使用孤立内外特征或传统多特征图构建方法的方法相比，所提算法在低SNR条件下具有卓越的识别性能。此外，与传统的ResNet18相比，DECA-ResNet18对多载波复合调制信号表现出更强的识别性能。
