A practical approach for accurate positioning with L1 GPS receivers using neural networks

Abstract

The ability to determine an accurate global position has many useful commercial and military applications. Because of the L1 GPS receiver's error sources, it is essential to model them. In this paper, a new approach is presented for improving low cost receivers positioning accuracy with Differential GPS (DGPS) corrections real time prediction using pi-sigma, sigma-pi, recurrent, and parallel recurrent neural networks. Methods validity is verified with experimental data from an actual data collection, before and after Selective Availability (SA) error. The result is a highly effective estimation technique for accurate real time positioning; so that prediction RMS errors were less than 0.40 meter after prediction, independent of SA error. The experimental test results with real data emphasize that total performance of RNN is better than PSNN and SPNN considering trade off between accuracy and speed for DGPS corrections prediction.

The performance of proposed Parallel Recurrent Neural Network (PRNN) is compared with RNN in DGPS corrections real time prediction. The experimental results demonstrate which the PRNN has great approximation ability and suitability than RNN; so that the PRNN prediction total RMS error respect to the RNN is improved from 2.7348 to 1.7576 meters for 10 seconds ahead prediction and from 4.0397 to 2.5937 meters for 30 second ahead prediction, respectively.