Performance of large distributed detection networks and the sign detector

Abstract

A comparison is made of the performance of fusion and serial distributed networks and the centralized system when there are many local sensors. For the fusion network, the author compares its asymptotic efficacy to an approximation of the probability of error that is shown to predict performance more accurately when the signal is not vanishingly small. This result also applies to the sign detector, which is special case of the fusion network. For the serial network, no analytic performance equation is available for general noise. However, for Laplacian noise, the author shows that the performance is bounded as the number of processors increases. She compares the performance of the two networks numerically, using the framework of detecting a known signal in generalized Gaussian noise. The fusion network always performs better than the serial network for many sensors. The fusion network performs best in Laplacian noise and worst in nearly uniform noise. The serial network performs best in nearly uniform noise and worst in Laplacian noise. >