DETECTION OF CHANGES IN THE SPECTRUM OF A MULTIDIMENSIONAL PROCESS

We present an algorithm for the sequential detection of changes in the spectrum of a multidimensional process. We investigate the asymptotic properties of the statistic that we use in the case of a real Gaussian process. The algorithm of detection is based on a sequential likelihood-ratio test. Simulations show the verv good behavior of the algorithm in the case of Gaussian and non-Gaussian processes. In both cases, change are detected with good accuracy, while the number of false alarms is small.