Title: Airborne GPS Interference Cancellation Algorithm Based on Deep Learning
Author(s): Qiong Yang, Yi Zhang, Baowang Lian, Chengkai Tang
Published in: Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017)
September 25 - 29, 2017
Oregon Convention Center
Portland, Oregon
Pages: 1695 - 1700
Cite this article: Yang, Qiong, Zhang, Yi, Lian, Baowang, Tang, Chengkai, "Airborne GPS Interference Cancellation Algorithm Based on Deep Learning," Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017), Portland, Oregon, September 2017, pp. 1695-1700.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: This paper investigates one method of simultaneous suppression of various types of interference for airborne GPS receiver. We present a multi types interference cancellation algorithm based on deep learning for airborne dual antennas GPS receiver. Since the satellite navigation signals are easily obscured by the huge fuselage, we install an antenna on the top and abdomen of aircraft respectively. The abdomen antenna of aircraft can only receive the interference signals from the ground as the reference channel input. Meanwhile, the top antenna of fuselage gets the satellite navigation signals from the sky plus the interference signals from the ground as the main channel input. Then the adaptive interference cancellation is realized by the neural network based on deep learning model. The main channel inputs subtract the reference channel inputs which are adjusted by the Deep Neural Network (DNN) and get the output error. In turn, the output errors feedback adjust the weights of the deep neural network. Finally, the convergence output errors are the exact satellite navigation signals after interference cancellation. The deep neural network interference canceller has a higher hidden layer to improve the fitting accuracy of network. The training method of DNN uses the Momentum Elastic Averaging Stochastic Gradient Descent(MEASGD)algorithm to speed up the convergence rate of training large scale data. The DNN only need the current observation data and has strong nonlinear mapping ability with the advantages of strong self-learning ability, low computational cost and good real-time processing performance. The simulation results show that the proposed method avoid the interference type recognition and direction of arrival interference estimation, meanwhile it breaks through the limitation on the maximum interference quantity of array anti-jamming methods. The receivers suppress various types of interference with a low complexity, low cost, universal method successfully at the same time.