基于1D-CNN的射频强度温度传感研究  

Research on RF Intensity Temperature Sensing based on 1D-CNN

作　　者：丁美琪 桂林[2] 王子怡 尚荻森 钱敏 李乾坤 DING Meiqi;GUI Lin;WANG Ziyi;SHANG Disen;QIAN Min;LI Qiankun(School of Intelligent Manufacturing and Control Engineering,Shanghai Polytechnic University,Shanghai 201209,China;School of Computer and Information Engineering,Shanghai Polytechnic University,Shanghai 201209,China)

机构地区：[1]上海第二工业大学智能制造与控制工程学院,上海201209 [2]上海第二工业大学计算机与信息工程学院,上海201209

出　　处：《光通信研究》2025年第2期99-104,共6页Study on Optical Communications

摘　　要：【目的】文章研究了基于一维卷积神经网络(1D-CNN)的微波光子滤波器(MPF)在射频(RF)强度温度传感中的应用,以提高温度传感的精度和效率。【方法】文章实验搭建了基于马赫-曾德尔干涉仪(MZI)结构的MPF系统,通过改变环境温度,采集了在陷波深度为8.1 dB条件下20~70℃的RF谱数据,每个温度条件下采集30组数据,然后采用贪心策略设计并优化1D-CNN结构,确定网络层数、卷积核大小、池化核大小以及激活函数类型。利用训练集数据对模型进行训练,并使用测试集数据进行验证,优化模型参数以获得最佳性能。利用其非线性映射能力从RF谱数据中提取特征,实现RF强度与温度变化的高精度解调。最后采用均方根误差(RMSE)作为评价指标,并将1D-CNN的性能与传统算法(最值法、质心法和高斯拟合法)进行对比,分析其在不同温度条件下的性能。【结果】实验结果表明,基于1D-CNN预测模型的RMSE达到了10-3量级,而传统算法的RMSE通常在10-1量级。与传统高斯拟合算法相比,基于1D-CNN的算法解调速度提高了2.72倍。1D-CNN在不同温度条件下均表现出较高的稳定性和较低的误差。【结论】1D-CNN在处理复杂的非线性关系和特征提取方面具有显著优势,不仅在计算效率和鲁棒性方面表现优越,还能有效应对噪声和环境的干扰。文章的研究为MPF在RF强度温度传感领域的应用提供了新的思路和方法。【Objective】In order to improve the accuracy and efficiency of temperature sensing,the application of Microwave Photonic Filter(MPF)based on One-Dimensional Convolutional Neural Network(1D-CNN)in Radio Frequency(RF)intensit temperature sensing is studied.【Methods】The MPF system based on Mach-Zehnder Interferometer(MZI)structure is built experimentally,and the RF spectral data of 20~70℃under the condition of notch depth of 8.1 dB are collected by changing the ambient temperature.30 sets of data are collected under each temperature condition.Then the 1D-CNN structure is designed and optimized by greedy strategy to determine the number of network layers,the size of the convolutional kernel,the size of the pooled kernel and the type of activation function.The model is trained with the training set data and validated with the test set data to optimize the model parameters for optimal performance.Its nonlinear mapping capability is used to extract features from RF spectral data to achieve high-precision demodulation of RF intensity and temperature changes.Finally,the Root Mean Square Error(RMSE)is used as the evaluation index,and the performance of 1D-CNN is compared with the traditional algorithms(maximum-value method,centroid method and Gaussian fitting method)to analyze its performance under different temperature conditions.【Results】The experimental results show that the RMSE of the prediction model based on 1D-CNN reaches the order of 10-3,while the RMSE of the traditional algorithms is usually in the order of 10-1.Compared with the traditional Gaussian fitting algorithm,the demodulation speed of the 1D-CNN-based algorithm is improved by 2.72 times.1D-CNN shows high stability and low error under different temperature conditions.【Conclusion】1D-CNN has significant advantages in dealing with complex nonlinear relationships and feature extraction,not only superior in computational efficiency and robustness,but also effective in dealing with noise and environmental interference.The research in this pape

关 键 词：一维卷积神经网络 微波光子滤波器 光纤传感 温度传感 射频强度 

分 类 号：TN256[电子电信—物理电子学]