A comparative study between performance of recurrent neural network and Kalman filter for DGPS corrections prediction

• Published in: Proceedings 7th International Conference on Signal Processing, 2004. Proceedings. ICSP '04. 2004 Vol. 1; pp. 356 - 359 vol.1
• Main Author: Mosavi, M.R.
• Format: Conference Proceeding
• Language: English
• Published: Beijing IEEE 2004; Piscataway NJ Pub. House of Electronics Industry
• Subjects: Applied sciences; Artificial intelligence; Computer science; control theory; systems; Connectionism. Neural networks; Detection, estimation, filtering, equalization, prediction; Equations; Error correction; Exact sciences and technology; Filters; Gain measurement; Global Positioning System; Information, signal and communications theory; Neural networks; Noise measurement; Predictive models; Radiolocalization and radionavigation; Recurrent neural networks; Satellite broadcasting; Satellite telecommunications. Space telecommunications; Signal and communications theory; Signal, noise; Telecommunications; Telecommunications and information theory; Performance evaluation; Hopfield neural nets; Recurrent neural nets; GPS system; Predictor; Differential system; Kalman filter; Radiolocalisation; Predictor corrector method; Accuracy; Satellite navigation; Learning algorithm; Comparative study
• ISBN: 9780780384064; 0780384067
• DOI: 10.1109/ICOSP.2004.1452655

Recurrent neural networks (RNNs) trained by the real time recurrent learning algorithm are capable of storing sequential information from the past. This property makes these networks ideal for DGPS corrections prediction. In this paper, performance of RNN is compared with Kalman filter predictor (KP) based on a state dependant model for DGPS corrections prediction. The experimental tests results with real data are stated and discussed in this paper. The results on real data indicate that the KP can predict DGPS corrections with better accuracy, but more slow.