Fault diagnosis for a delta-sigma converter by a neural network

The diagnosis of faults in a first order /spl Delta/-/spl sigma/converter is described. The circuit behaviour of fault-free circuits and circuits containing single faults were simulated and characterized by the output bitstream patterns. The latter were compared with that of the ideal fault-free circuit. A Simplified fuzzy ARTMAP was trained with metrics derived from the bitstreams and their assigned class. A diagnostic accuracy of 93% was achieved using just two of the metrics. The technique might be useful for the diagnosis of other circuits.