Test Generation Based on SVM for Analog System with a Multi-training Sets Method

This paper proposes a novel analog test generation based on the SVM (support vector machine) with a multitraining sets method. The test generation method in this paper generates test signals directly from the sample space of the output responses of the analog DUT. The training sets are randomly generated in the test generation, so we use a multitraining sets method to avoid the influence of choosing different training sets. This method can avoid achieving the worst misclassification rate. The experiment confirms that it can enhance the efficiency of the test generation.