Comparative analysis of DGPS predicted corrections using dynamic neural networks

Differential Global Positioning System (DGPS) is a technique to improve the accuracy of the GPS positioning. In DGPS, error correction signal is transmitted to the surrounding rovers. Any correction loss during transmission may lead to navigation inaccuracy. This problem can be minimized by incorporating Dynamic Neural Networks (DNNs) at the rovers. DNNs can be used to predict the present and future DGPS correction values by utilizing the past correction values. This paper presents the prediction of error correction values using DNNs such as Focused Time Delay Neural Network (FTDNN), Distributed Time Delay Neural Network (DTDNN), Nonlinear Auto Regressive with eXogenous input Neural Network (NARXNN), Nonlinear Auto Regressive Neural Network (NARNN) and Layer Recurrent Neural Network (LRNN). The results show that the Mean Square Error (MSE) in predicted correction values due to third order LRNN is the least (2.5316e- 05 m).