Distributed signal processing with application to constant false alarm rate radar target detection

Distributed Signal Detection (DSD) schemes are needed when system performance factors such as speed, reliability, and constraint over the communication bandwidth are taken into account. The inherent redundancy possible with multiple sensors, the availability of high speed communication networks, and increased computational capabilities have spurred great research interest in this topic. In DSD techniques, each sensor sends either a binary decision or a condensed form of information (statistics) about the observations available at the sensor to the fusion center where a final decision about the presence of a target is made. In this dissertation, two problems in DSD are considered. The first concerns distributed constant false alarm rate (CFAR) radar target detection and the second problem concerns the performance loss computation in order to assess the quantization effect in DSD.; A new CFAR test using distributed sensors is developed. The sensor modeling assumes that the returns of the test cells of different sensors are all independent and identically distributed. In the proposed scheme each sensor transmits its test sample and a designated order statistic of its surrounding observations to the fusion center. At the fusion center the sum of the test cells' samples is compared to a constant multiplied by a function of the order statistics. For a two-sensor network, the functions considered are the minimum of the order statistics (mOS) and the maximum of the order statistics (MOS). For detecting Rayleigh fluctuating target in Gaussian noise, closed form expressions for the false alarm and detection probabilities are obtained. The numerical results indicate that the performance of the MOS detector is very close to that of a centralized OS-CFAR, and that it performs considerably better than the OS-CFAR detector with AND or OR fusion rule. Extension to an N-sensor network is also considered, and general equations for the false alarm probabilities under homogeneous and nonhomogeneous background noise are presented.; The loss associated with a distributed signal detection scheme as compared to a central procedure is evaluated with respect to the average probability of error. For a system of N sensors, closed form probability of error expressions as a function of the number of sensors which send their decisions to the fusion center, are derived. Different distributions of the observations considered are gamma, exponential, normal and Poisson. For a network of five sensors, the average error probabilities for the above cases are computed. Numerical results show that the loss due to quantization is more significant for gamma (with large shape parameter) and normal densities than is for the exponential density. Also, the loss increases as the ratio of the Signal-power to the Noise-power (SNR)increases.