计算机科学
电子工程
调制(音乐)
量子
电子线路
物理
信号处理
算法
人工智能
信号(编程语言)
人工神经网络
频率调制
钥匙(锁)
噪音(视频)
方案(数学)
解调
传播学
数学
调幅
量化(信号处理)
鉴定(生物学)
量子系统
编码(内存)
拓扑(电路)
作者
Ximing Yu,Zhiyang Zhang,Xiaogang Tang
出处
期刊:
日期:2026-07-01
卷期号:: 1261-1268
标识
DOI:10.1109/cisat71030.2026.11682421
摘要
To address the issues of noise propagation and phase feature loss in multi-sensor cooperative reception for communication signal modulation recognition, this paper proposes a modulation recognition method based on variational quantum circuits. A dual-sensor cooperative reception model is constructed with a max-pooling front-end to preserve instantaneous phase features. CNOT gates are introduced into the variational quantum circuit to achieve quantum entanglement fusion of multi-channel features, which empirically suppresses the adverse effects of noise in the fused representation. To mitigate the Barren Plateau problem (exponential gradient decay) during training, an identity sequence block initialization strategy is adopted to maintain gradient flow. Experimental results show that in the low SNR range from -10 dB to 6 dB, the proposed method achieves 5.09% higher recognition accuracy than the classical baseline. In the 2 dB to 14 dB SNR range, accuracy remains stable between 90% and 92%, peaking at 92.29% at 8 dB. A shallow (2-layer) architecture outperforms a deeper (12-layer) one, which suffers from over-expressivity and generalization degradation at high SNR. Ablation studies validate the effectiveness of identity initialization and quantum entanglement. This work revealing a quantum floor effect at -10 dB to -6 dB where HQCNN outperforms classical DeepCNN, ResNet-1D, and CLDNN, though a significant performance gap persists at high SNRs due to current quantum depth limitations imposed by the Barren Plateau problem.
科研通智能强力驱动
Strongly Powered by AbleSci AI