GNSS monitoring: Ionospheric model for GBAS

We use AI for CNS operations and infrastructure monitoring.

To support air navigation service providers when submitting a Ground-Based Augmentation System (GBAS) approval to their regulator, but also to guarantee the constant validity of the ionospheric model, EUROCONTROL is continuously monitoring the ionosphere over Europe. We are using AI techniques to improve the ionospheric gradient detection.

Background

A Ground-Based Augmentation System (GBAS) is a civil-aviation safety-critical system that supports local augmentation –at airport level– of the primary GNSS constellation(s). By providing enhanced levels of service it enables all phases of approach, landing, departure and surface operations notably under low visibility conditions.

One of the main sources of GBAS GNSS ranging error is the propagation delay induced by a particular layer of the earth’s atmosphere: the ionosphere. To overcome this integrity issue an ionospheric model was developed, based on ionospheric data collected over the last decade.

The detection of ionosphere spatial gradient is complex and challenging: a significant amount of GNSS raw data needs to be processed and the current processing techniques provide many “fake” gradients coming from measurement artefacts. As a result, a manual validation of the processing outcomes is required, which represent a significant amount of effort from skilled operators. As output of the current automatic processing, more than 90% of the detected gradients are classified as measurement artefact during the manual validation.

How it works

To improve the quality of the automatic gradient detection and reduce the manual effort, Artificial-Intelligence-based algorithm were tested. Since the beginning of this project (October 2012), a total of 16084 potential gradients were manually validated out of which, 1641 were true ionosphere gradients. These gradients represent a valuable database that can support a machine-learning algorithm. The database has been prepared with the objective to overcome two limitations: the number of entries is limited (16000 is small in the machine-learning world), and the database is biased (10% of “True’” entry for 90% of “False” entry). These limitations have been mitigated through data augmentation by applying small variation techniques (scaling & reversing).

The machine-learning techniques were used in an “offline” neural network. Once optimized, the best trained neural network was frozen and implemented in the GNSS data processing, improving further the GBAS ionospheric model, without compromising the safety criticality of the approved operation.

Benefits

The results showed that the best architecture is a mix of the two technics (a convolutional network and a recurrent (LSTM) network), providing a classification between true gradients and measurement artefact 3 times better than the former processing technique.

Based on this first application, AI is considered to assist further in the CNS monitoring tasks under the responsibility of the Network Manager.