论文部分内容阅读
在混合信号自动测试系统中,周期性模拟信号波形的识别是ATE自学习建库和自动测试、分析的关键。提出一种运用空间分割竞争网络结合Kohonen学习规则实现有规则模拟信号波形的识别方法,该方法用空间块归属的概念代替超平面分类,具有空间上离散分类的能力。重点讨论了通过软边界处理的Kohonen训练规则,该学习方法可将识别特征值分布特性通过软边界技术快捷地训练到寻址层隶属度矩阵中,克服由大样本训练实现空间块隶属度统计的不足,减少训练样本、提高学习速度。通过以信号十次谐波之和、信号一周面积及信号正负面积之比为特征,识别几类常见信号波形为例,表明该方法对波形识别的实用性。
In the mixed signal automatic test system, the periodic analog signal waveform recognition is the key to ATE self-learning database building and automatic testing and analysis. This paper proposes a method to identify the waveform of a regular analog signal using the algorithm of Kohonen learning by using the network of spatial segmentation competition. This method replaces hyperplane classification with the concept of spatial block ownership, and has the ability of discretely classifying in space. This paper focuses on the Kohonen training rules through the soft boundary processing. This learning method can quickly identify the distribution of recognition eigenvalue to the addressing layer membership matrix by using the soft boundary technique, and overcomes the problem of large block membership Insufficient, reduce training samples, improve learning speed. By taking the sum of the ten harmonics of the signal, the area of the signal one week and the area of the signal plus to the signal as the characteristics, several common signal waveforms are identified as an example, which shows the practicability of the method for waveform recognition.