Prediction of GPS user position using LM and CGF Algorithms

Global Positioning System (GPS) is a satellite based navigation system used worldwide to find the user position on or near the earth. The system provides navigation ability to military, civil, and commercial GPS receivers around the world with good accuracy. However, if the signals transmitted by the GPS satellites are lost temporarily due to varying atmospheric conditions, satellite health problems or due to any other reason then the position information could not be estimated by the receiver. To solve this problem, Artificial Neural Network (ANN) can be incorporated at the receiver to predict the position. Navigation with improved accuracy, integrity, availability and reliability can be made possible with ANN. In this paper a dynamic neural network called Nonlinear AutoRegressive with eXogenous inputs (NARX) network trained with prominent methods such as Levenberg Marquardt (LM) and Conjugate Gradient with Flectcher Reeves updates (CGF) algorithm is used for the prediction of the GPS user position. The performance of CGF algorithm is found better for the prediction of user's latitude and longitude and the performance of LM algorithm is found better for the prediction of user's height. Further for different orders of the network, performance is evaluated. It is found that the Mean Square Error (MSE) in predicted user position decreases with the increase in the order of the neural network.