Artificial Neural Network-based Transient Response Approach To Fault Diagnosis Of Mixed Signal Circuits

The fault diagnosis of mixed-signal circuits is an advanced subject paid close attention extensively. With development of electronic technology, especially VLSI and mixed signal circuits, it brings up new challenge to fault diagnosis of mixed signal circuits. Because most electronic devices include digital components and analog components now, testing of mixed signal circuits has important actual meaning. Some research on mixed signal has been made abroad but there is little literature relevant to the subject domestic. It is a new subject for fault diagnosis by using of Artificial Neural Network. In recent years, researchers paid more attention to ANN than before. The application of ANN has turned into a succeed aspect for ANN.A new unified method to test mixed signal circuits is presented in the paper. Using Transient Response Analysis, the method needn't partition the mixed signal circuits into digital units and analog units and solves the difficulty that the characteristics between analog circuits and digital circuits are too different to test. In traditional DC analysis several nodes in circuit need be sampled. By using Transient Response Analysis, different transient response curves correspond to different states of the circuit and time domain character of output node is sampled. Obviously the method is better than traditional approaches.On the basis of above-mentioned methods, the application of Transient Response Analysis to diagnosis of analog circuits is further studied. In measurement, nodes are difficult to test and voltage(current) values of nodes are usually got by supplementary means. Only input node and output node need to be tested after applying the new method. Thus the testing process is simplified greatly , testing expenses is reduced andthe time of testing is saved.Then time domain characters are drew from transient response curves and treated with SOFM network. Besides the robustness and associated memory of ANN,SOFM has unique advantage: if the input data changes with time according to a certain statistical distribution, system can adapt to the change automatically and can continue simulate the data distribution of the input mode present. The speed of SOFM network is fast, even in the course of training. So SOFM network method is a kind of verypotential diagnosis method. Last the results of simulation to actual analog circuits and mixed signal circuits indicate that it is a good approach to diagnosis circuits correctly. It is proved be feasible under lots of experiment.