ROBUST ESTIMATION OF SINUSOIDAL SIGNALS WITH COLORED NOISE USING DECENTRALIZED PROCESSING

A new technique is developed for the estimation of the number of signals and their central frequencies using decentralized processing when it is known a priori that the observations consist of a finite number of sinusoidal signals corrupted by an additive colored random noise process with unknown correlations. Such a noise sequence may be caused by jamming from a hostile agent. Our decentralized processing scheme is one in which each sensor estimates the frequencies and their covariance matrix, and sends the results to the fusion center. At the fusion center, since the estimates from the sensors have a mixture density which is possibly not Gaussian, a robust technique is utilized to combine the estimates. Even when the numbers of frequencies transmitted by the various sensors are identical, determining corresponding frequencies from each sensor is not a straightforward task. Also, outliers caused by line splitting or by spurious frequencies are hard to detect. These problems can be resolved by two methods: the so-called refitting method and ranking method. Algorithms for the ranking method and refitting method are presented in detail. Simulation results confirm that the new methods perform well for combining multiple frequency estimates. Moreover, the overall performance does not deteriorate quickly if some of the sensors are destroyed. © 1990 IEEE