Temporal analysis of high-rate GPS satellite differential code biases for signal monitoring and anomaly detection

Kurzfassung

The increasing number of safety-critical applications using GNSS demands the continuous assessment of signals to ensure the reliability, accuracy and integrity of navigation solutions. In this context, monitoring systems play a key role. In principle, monitoring can be used for various kinds of aspects: the assessment of the nominal system performance and service quality, monitoring for failure detection in GNSS subsystems, for environmental analysis like radio frequency interference (RFI) detection, multipath characterization as well as for the analysis of signal propagation effects due to the atmosphere. In particular, monitoring strategies for system-based irregularities and signal threats, which are more difficult to detect and characterize, are of fundamental importance for integrity assessments for safety critical applications. Indeed, if the impact of such irregularities is not properly characterized, the computed protection levels may turn out to be optimistic and the true integrity risk of system would not be properly defined. Signal distortions originated in the satellite payload are an important error source that may cause a degradation in the performance of specific applications and represent a threat for safety critical applications if such distortions are non-nominal, also called evil waveforms. In general, nominal and non-nominal signal distortions cause biases in the received signal leading to positioning errors that range from several decimeters to several meters. Current approaches for the characterization and detection of nominal and non-nominal signal distortions employ correlator outputs from specialized receivers. In the context of satellite-based augmentation systems (SBAS), monitoring schemes for threat detection already exist, like the US-American Wide Area Augmentation System (WAAS) and the European Geostationary Navigation Overlay System (EGNOS). Nevertheless, such systems have been deployed to monitor application-specific metrics and are thus limited in the scope of detection of signal changes. In addition, the detection capabilities are limited to a certain area, e.g., North America, Europe and North Africa and require the availability of specialized receivers to provide raw GNSS measurements, such as correlator outputs. In last few years, the impact of signal distortions in GPS signals on code observations caused by the activation and deactivation of the so-called flex power in IIR-M and IIF satellites has been analyzed. Furthermore, in recent studies, high-rate estimates of GNSS differential code biases have been used for the detection of anomalies resulting from the presence of non-nominal signal distortions using unsupervised machine learning methods. This study continues the latest efforts at DLR in the development of a system for a global and continuous signal monitoring and search for GNSS signal changing events that may represent a general threat, i.e. not only for specific applications. The presented strategy builds up on the experience from previous studies using high-rate estimates of differential code biases for anomaly detection computed using publicly available observations from the globally-distributed network of the International GNSS Service (IGS). Other than previous schemes based on unsupervised machine learning methods using data batches, partially disregarding the time-dependency of the data, this study presents a strategy that is based on a temporal analysis of high-rate estimates of differential code biases. In a first step, observations in the Receiver Independent Exchange Format (RINEX) from stations in the IGS network are used to estimate satellite differential code biases from different combinations of signal components. In a second stage, selected data periods for each signal combination are chosen in order to fit/train autoregressive and machine learning models. Finally, the trained models are employed for the computation of forecasts in the selected period, which are then combined in an ensemble prediction and compared with data from the test data set (observed values) for the computation of anomaly detection metrics. For each forecast, statistical error bounds are computed, which are similarly employed in the evaluation of the anomaly detection metrics. By using an analysis based on the time series of differential code bias estimates, the present strategy aims at providing more reliable detection capabilities in comparison with previous approaches using a batch processing of data. The proposed strategy has been tested using data sets from the GPS IIF SVN66 and SVN73 satellites from May-July 2021 and August-September 2022, which exhibited apparent anomalies due to non-nominal signal distortions on 1 July 2021 and 29 September 2022, respectively. The obtained results show that the presented scheme can reliably characterize the time series of satellite differential code bias estimates as well as detect non-nominal signatures that suggest the presence of anomalous signal distortions.