Training of Adaptive Antennas Using Simulated Data
Tayfun Özdemir, Christos George Christodoulou, M.J. Miranda · 2005
GPS is vulnerable to multipath interference and intentional jamming, and many variants of multipath mitigation systems have been proposed and deployed. Such GPS receivers make use of adaptive arrays that employ spatial processing to place nulls in the direction of interfering signals. Although this approach is adequate for narrowband signals, it may be inadequate for broadband operation, especially when multipath is present. A new approach, based on machine learning and support vector machines (SVM) has been developed in order to enhance the spatial and temporal capabilities of the existing adaptive array antennas used by GPS receivers. The new approach makes the antenna array intelligent, so that when one of the antenna elements in the array fails, the performance of the GPS array degrades gracefully for both narrowband and broadband signals. The beamforming algorithms need to be improved using inter-element coupling models and trained by data contaminated with multipath interference. Experimental data is expensive to obtain. Therefore, VirAntenn/spl trade/ antenna array simulation software from Virtual EM Inc. has been interfaced with a high frequency propagation modeler (HFPM) (also by Virtual EM) to produce the training data. Inter-element coupling is provided by VirAntenn/spl trade/ alone, while multipath interference has been simulated by integrating the two software.