A DSP testing approach by modeling the circuit response as a Markov chain

An approach based on signal processing and time domain analysis is proposed to test linear and non-linear analog circuits. A step signal is applied as an input stimulus and the response is measured at the output. A stochastic model is used to represent the response of a DUT to the step input. The process itself is modeled as a monodesmic, ergodic Markov chain. Different metrics are computed for the faulty and non-faulty circuit in order to detect the faulty components and to estimate noise. Because there exists a relationship between the measured metrics and the functionality of the DUT (Device Under Test), we can characterize the response of the circuit and compact it as a signature. This signature serves to differentiate between the faulty and non-faulty circuit. Different signatures can be obtained for different injected faults and a fault dictionary can be constructed. A comparative study, between a TMS 320C40X DSP processor and a Sparc10 Sun Station, is done in order to evaluate the computation time of the proposed algorithm