| Abstract: | Since the Global Positioning System (GPS) satellites broadcast signals travel a long distance, the received signals are attenuated below the thermal noise level. Such weak signals are seriously subject to intentional or unintentional interferences from hostile or friendly noise sources. We propose an NN-based predictor for GPS anti-jamming applications. This new method exploits the adaptive notch filter as a cascade filter and the simple structure of Sigma-Pi neural network (Σ-Π NN). The Σ-Π NN can be trained quickly while avoiding the huge exponential computation for updating weights and thresholds in each layer, which allows easy hardware implementation. Simulation results show that its signal-to-noise ratio (SNR) improvement factor exceeds the factors of conventional multilayer perceptron, recurrent neural network, and other compound methods in single-tone and multi-tone continuous wave interference environments. Besides improving SNR, the anti-jamming performances are evaluated by computing root mean squared (RMS) prediction error and space vehicle (SV) observation number. The proposed algorithm provides the desired SV observation number, even for more than four vehicles, increases SNR improvement by about 46 % on average, and reduces RMS by about 27 % in average in both jamming mitigation processes. |
