A new robust neural network method for coherent interference rejection in adaptive array systems.

摘要

This dissertation considers the techniques for adaptive noise cancellation that exploit prior information about the observation noise so that the detection performance of the desired target signal can be increased. In the first section, the LMS algorithm and beamformer application for narrowband interference rejection is analyzed, and problems associated with this method are described in detail. Computer simulations are performed to verify the results that were predicted based on the theories.; In addition, a description of the fundamental architectures of the back propagation (BP) algorithm is provided, and a new BP beamformer that will overcome some of the problems associated with the conventional LMS beamformer is investigated. The simulation compares the performances of the adaptive array systems for the conventional LMS beamformer with the multilayer neural network, using the standard BP learning algorithm.; The simulation comparisons have indicated that if the signals are not simply distributed or are highly correlated, the implementation of the conventional LMS technique may be inadequate. On the other hand, the BP beamformer achieves desirable performances even in some of the worst situations where the interference is at the same frequency as the target signal.; It has been shown that the BP beamformer architectures are robust, and learning can be performed rapidly for simple decision patterns. However, if complex decision patterns are required, convergence speed for learning time can be excessively long, and performances are often unpredictable for initial weights and set-up parameters. These shortcomings of the BP beamformer have led to the development of the counter propagation (CP) beamformer, which is a new type of neural network architecture.; The remaining section of the dissertation describes a new CP neural network technique that may improve the performances of the standard BP beamformer. Computer simulations compare the CP beamformer with the BP beamformer in relatively complex decision problems. These results illustrate the performances of a certain number of complex target classifications, with emphasis on the speed of convergence and the effectiveness of the output responses for interference rejection.