【摘要】 模数转换器(ADC)测试主要包括静态参数和动态参数两个测试过程。随着性能的提升，ADC的测试复杂度和成本也急剧增加。替代测试，即通过分析两类参数间的关系来实现一个测试过程得到两类参数，已被证明是降低ADC测试复杂度和成本的主要方案之一。本文通过构建基于人工神经网络的参数预测模型来实现替代测试，模型以总谐波失真为预测目标，以静态性能参数为输入特征。针对高维的ADC非线性曲线，文章结合统计分析和主成分分析设计了专用的特征提取方法，在降低特征维度的同时尽可能地减少了信息损失。模型在测试集上的预测结果与参考值的均方误差和拟合优度分别达到了1.15 dB和0.6,显著优于相关对比模型。此外，在SHAP解释器的框架下分析了上述模型的预测目标和特征变量之间的依赖关系，并得到了有意义的结果。更多还原

【Abstract】 The test of analog-to-digital converter(ADC) mainly includes two test processes of static parameters and dynamic parameters. With the improvement in the performance, the testing complexity and cost of ADC increases dramatically. Alternative test, which means obtaining two types of parameters from only one test process by analyzing the relationship between static and dynamic parameters, has been proven to be a major solution to reducing the complexity and cost of ADC test. In this article, the alternative testing is achieved by constructing a regression model based on artificial neural network. The model takes total Harmonic distortion(THD) as the prediction target, and takes the static performance parameters as the input features. For high-dimensional ADC nonlinear curves, statistical analysis and principal component analysis are combined to design a special feature extraction method, which greatly reduces the feature dimension and the loss of information. The prediction results on the test set show that the mean absolute error and R-squared between the predicted THD and the reference value reach 1.15 dB and 0.6, respectively, which are significantly better than those of other comparison models. In addition, SHAP(shapley additive explanations) model interpreter is used to analyze the dependencies between the prediction target and feature variables of the model, and meaningful results are obtained.更多还原