On the detection of a sudden change in system parameters

This correspondence deals with the problem of detecting a sudden change in system parameters by means of noisy observations made on the system. The solution given here is dependent on the classical Bayes criterion. However, by forming an auxiliary sequence of O's and l's as the observations are taken, it is shown, at most, three log-likelihood numbers need to be updated recursively at any stage. An example is given to illustrate the principle.