Abstract: Analog circuit has its tolerance on component parameters, continuous response and nonlinearity, and the limitation of the number of test points, it is difficult to achieve expected results in practical engineering based on classical circuit fault diagnosis. While neural network has the characteristics of fault tolerance, generalization ability and non-linear processing, this paper extracts the fault feature in radar circuit rapidly and validly, constructs the samples of neural network, and solves the problems in fault diagnosis of analog circuits with frequency characteristics of circuits.

Key words: analog circuit, fault diagnosis, feature extraction, neural network, frequency characteristics