GNSS interference monitoring and detection based on the Swedish CORS network SWEPOS

1 Introduction

With the modernization of the Global Positioning System (GPS) and the advent of other similar systems such as the Russian Globalnaya Navigatsionnaya Sputnikovaya Sistema (GLONASS), the European Galileo, and the Chinese BeiDou Navigation Satellite System (BDS), the availability of signals of opportunity, Global Navigation Satellite System (GNSS), and GNSS-dependent technologies are evolving at an unprecedented rate. GNSS is providing applications ranging from surveying, remote sensing and machine control, traffic monitoring, and synchronization of power and communication networks (Nordin et al. 2009, Petrov et al. 2016). In addition, transportation and critical activities such as emergency vehicles use GNSS to provide accurate destination and route information (Raza et al. 2022). GNSS is also an important tool for aircraft, trains, and ships. Large-scale applications that require GNSS, such as smart cities, smartphones, Internet of Things devices, and self-driving cars, are also emerging (Minetto et al. 2020). The latter applications, such as the positioning of self-driving cars, require GNSS with better quality, continuity, and integrity (Jost 2022).

However, GNSS signals have insufficient power when received by ground receivers, making them vulnerable to various sources of interference. Sources of interference to GNSS signals are generally classified as unintentional and intentional. Ionospheric scintillation, other systems with similar frequencies to GNSS, and broadcast and telecommunications transmitters are examples of unintentional interference. Interference can also be intentional through jamming and spoofing. Jamming can degrade or completely weaken the GNSS signal reception, and spoofing can cause a receiver to appear in the wrong position.

The threat of GNSS interference has grown in recent years. GNSS and GNSS-dependent infrastructure are vulnerable to these threats. For example, the EU-funded research project STRIKE3 reports that nearly 500,000 interference incidents have been reported, of which more than 10% were reported as deliberate attacks (Bhuiyan et al. 2019). Nearly 10,000 GPS attacks have also been reported throughout the Russian Federation (C4ADS 2019). Other recent reports of intentional jamming include Russian jamming of GPS in Norway in 2019 (Reuters 2019) and reports of GPS jamming of aircraft flying over Kaliningrad, Russia, in March 2022 (Reuters 2002). A recent study by Roberts et al. (2022) also showed a significant correlation between the occurrence of GNSS signal interference and geopolitical activities and regional conflicts. There are also reports of unintentional interference threats (Wolff et al. 2014, Rødningsby et al. 2022), as there are authorized users such as radio amateurs who share frequency bands with GNSS and may interfere with some GNSS signals, such as with Galileo E6 (José et al. 2015).

These threats have led to a growing interest in GNSS developments and interference detection studies for continuous monitoring of GNSS bands. The general purpose of these developments is to protect critical GNSS infrastructures from emerging (un)intentional threats. The same GNSS infrastructure can also be used for threat signal situational awareness. For example, Finland has developed a system to monitor GNSS signal quality and detect interference using the Finish permanent GNSS network FinnRef (Nikolskiy et al. 2020). The potential of using the Swedish Continuously Operating Reference Stations (CORS) network SWEPOS for interference monitoring and detection has also been investigated (Alexandersson et al. 2021, Fors et al. 2021). Extensive research has been conducted by the Swedish Defense Research Agency (FOI) in collaboration with Lantmäteriet and Swedavia (Alexandersson et al. 2021, Fors et al. 2021) by performing controlled simulated interference waves on three different GNSS receivers, which are typically used by the SWEPOS reference stations. In addition, RF Oculus (Linder et al. 2019), a software-defined radio-based detection of GPS L1 signal interference, was installed at three reference stations close to the airport. Data from SWEPOS GNSS receivers were investigated and compared with the RF Oculus, and the study evaluated different detection algorithms for GNSS data based on the receivers. Using GNSS data from selected SWEPOS stations and comparing them with RF Oculus, the study concluded that effective interference detection can be achieved using basic GNSS data from SWEPOS stations, such as signal-to-noise ratio (SNR) measurements.

This article presents a GNSS interference detection system developed in SWEPOS that uses SNR data and monitors all GNSS signals with over 500 stations throughout Sweden. The detection method has been demonstrated using simulated data and real-world interference events. The purpose of building a SWEPOS-based GNSS band interference monitoring system is to provide comprehensive interference situational awareness for the whole reference network. Situational awareness is the first and critical step in risk management, such as localization, mitigating, and containing sources of disturbance. This will play an important role in ensuring quality and continuity and integrity in service and will, in turn, benefit the future critical applications and society to provide a clean GNSS spectrum. The impact of radio frequency interference (RFI) in positioning and other applications is not a goal of this work but is planned for future work; however, there are studies that show the impact of RFI, such as interference on carrier phase-based positioning (Larsen et al. 2021).

This article is organized and divided into six sections. Section 2 briefs the Swedish CORS network SWEPOS. Section 3 gives an overview of GNSS data-based inteference detections with a main focus on SNR. The SWEPOS interference detection is discussed in Section 4. In Section 5, the detection method is tested using simulated interference waves, and actual detected interference incidents are presented. Section 6 concludes this article with a discussion and possible future works.

2 SWEPOS

SWEPOS is a network of permanent GNSS reference stations in Sweden, operated and managed by Lantmäteriet, established in the 1990s in cooperation with the Onsala Space Observatory at Chalmers University. Since its establishment, SWEPOS has evolved in terms of network structure and modernization and now has nearly 500 stations spaced in the 10–70 km range (Figure 1). The network is equipped with latest receivers that enable it track all GNSS signals.

Figure 1

SWEPOS GNSS ground station observation network. Blue dots indicate the sites owned and operated by SWEPOS, while orange dots are third-party sites, owned and operated by others, such as those in Norway, Finland, and Denmark. Third-party stations are also incorporated into the routine for automatic interference detection, bringing the total number of monitored stations to 540.

SWEPOS offers a wide range of services, from dual-frequency data for geoscientific and meteorological studies to real-time kinematic (RTK) corrections for real-time applications. The SWEPOS Network-RTK service provides a real-time, high-accuracy correction service with nearly 10,000 users. It also provides real-time differential GNSS services for sub-meter accuracy. GNSS observations and navigation information are also stored hourly and daily in Receiver Independent Exchange Format (RINEX) for post-processing-related applications, as well as providing data to international initiatives and organizations such as the International GNSS Service (IGS) and the EUREF Permanent GNSS Network (EPN). In addition, SWEPOS is the foundation and backbone of the Swedish national geodetic reference network SWEREF99 (Jivall and Lidberg 2000).

SWEPOS ensures quality and continuity of it is service by continuously monitoring data quality in real-time, near-real-time, and post-processing modes. The interference detection presented in this study is part of this effort and is a near real-time monitoring of data quality using hourly RINEX data.

3 Interference detection

Several interference detection methods using GNSS data have been discussed in the literature. The most common are automatic gain control (AGC) and SNR-based detection systems. GNSS receivers are equipped with AGC to maintain gradual changes in the received signal power. When interference occurs, it affects the AGC performance and can be used to control and detect interfering signals (Bastide et al. 2003, Akos 2012). AGC-based interference detection has been reported to be sensitive to pulsed signals (Ndili and Enge 1998). However, one of the challenges in interference detection using AGC is threshold setting.

SNR is also widely used to monitor interference in GNSS signals and detect RFI (Calcagno et al. 2010, Borio and Gioia 2015, Balaei et al. 2006, Axell 2014, Axell et al. 2015, Roberts et al. 2022).

3.1 SNR

SNR is a ratio of carrier signal power to carrier noise power, commonly used to indicate the signal strength received by receivers. It is a very accessible parameter as it is reported by all commercial GNSS receivers and is also included in the RINEX data. In this study, SNR data are extracted from RINEX hourly data using G-Nut/Anubis (Vaclavovic and Dousa 2016), a quality control tool for multi-GNSS observations.

If there is RFI, for example, if an interfering object is near the GNSS receiver, the noise level of the GNSS measurement will increase and the SNR value will decrease. The decrease in SNR value can be monitored and used to detect interference in the signal. However, factors other than RFI can cause unpredictable events in the SNR. The known factors include ionospheric scintillations (Jiwon et al. 2007), multipath-induced SNR oscillations (Penina et al. 2005, Benton and Mitchell 2011), GNSS antenna gain-related satellite elevation dependency, and GNSS power flex (Steigenberger et al. 2019, Esenbuga and Hauschild 2020). In addition to the aforementioned factors, station equipment such as architecture and characteristics of the receiver, antenna type, antenna cable, and antenna splitters influence GNSS SNR.

Modern commercial receivers handle long-delay multipath well. However, short-delay multipath affects GNSS measurements and causes quasi-periodic oscillations in SNR (Benton and Mitchell 2011). Multipath and changes in the station environment can be unpredictable for rovers, for example, when receivers are moving in urban areas. However, station surroundings and multipath effects of a reference station such as the SWEPOS network can be partially monitored. Filtering the effects of multipath is essential for a detection system that aims to monitor SNR drops caused by interference. In this work, SNR values below 2 0 ∘ C elevation are discarded to reduce multipath effects.

GNSS satellites transmit signals with constant power. However, the GPS BLOCKs IIF and IIR-M satellites redistribute power over individual signals, which is called flex power (Steigenberger et al. 2019, Esenbuga and Hauschild 2020). The flex power causes drops in SNR and affects the estimation of GPS-derived products, such as the differential code bias estimation (Esenbuga and Hauschild 2020). Flex power-induced SNR drops need to be modeled to avoid false alarms.

Figure 2 shows the elevation-azimuth diagram of BLOCK IIR-M and IIF satellites for the encrypted P(Y)-code on GPS L2 stacked over all SWEPOS stations. Color code infers SNR values. It can be seen that, for ground stations within Sweden, the SNR drops due to the flex power changes occur when the satellites reach 225–360 degrees and 0–30 degrees azimuth. Although most SNR changes occur at lower elevation angles ( < 30 degrees), SNR drops occur at elevation angles up to 55 degrees for azimuths 250–290 degrees. These factors should also be taken into account when establishing SNR-based detection system.

Figure 2

Elevation-azimuth diagram of SNR for the encrypted P(Y)-code on GPS L2 for BLOCK IIF and IIR-M satellites stacked over the entire SWEPOS network of 500+ stations.

3.2 SNR model

Satellite elevation angle is the most dominant factor changing SNR, which is driven by antenna gain patterns. In addition, the length of the antenna cable, the antenna splitter, the receiver architecture, and the station environment all play a role in reducing SNR. As those factors would vary for each station and some of the factors could be frequency dependent, SNR is modeled for each frequency and each receiver. This is determined by a second-order polynomial best-fit regression model that takes into account the dependence of the SNR on the elevation angle of the satellites and other factors such as station equipment and GPS power flex. To account for the SNR drops resulting from flex power, separate polynomial fits are created for the different azimuth angles depicted in Figure 2. Figure 3 shows the raw SNR data (green points) for all tracked satellites over a day period for GPS L1 and the corresponding polynomial fit (red curve) plotted against the satellites’ elevation angle. The coefficients from the derived regression model are then used to calculate SNR residuals (model minus measured) for all satellites tracked.

Figure 3

SNR of the GPS L1 C/A code for all tracked GPS satellites, plotted against the satellites’ elevation angles over a day. Green points represent the raw data, while red points represent the polynomial fit SNR model.

Since the SNR is modeled for each receiver and frequency, factors such as equipment and frequency-related effects on the SNR are removed, and thus, the residuals (model minus observed data) represent the effects and disturbances that are not modeled, for example, due to interference. Interference can be radio frequency (RF)-related or non-RF-related, e.g., due to ionospheric scintillations, or station environment, e.g., attenuation due to tree leaves. The interference detection method presented in this article monitors and characterizes SNR residuals to classify whether signal interference is RF-related or not.

3.3 SNR residuals

When the direct signal (the effect described earlier) is removed, SNR residuals without interference follow a normal distribution. Figure 4 shows a histogram of GPS L1 SNR residuals averaged from all satellites for each station and stacked from all SWEPOS stations, and it shows that the SNR residuals follow a Gaussian distribution.

Figure 4

Histogram of GPS L1 SNR residuals from all SWEPOS stations. Cyan indicates the average SNR residuals from all satellites tracked at each station, stacked from all SWEPOS stations, and plotted together. The blue curve is the histogram fit.

In addition, a Shapiro–Wilk normality test confirms the normality of SNR residuals. The null hypothesis for the Shapiro–Wilk test is that the residuals are normally distributed. The null hypothesis is rejected at P -values less than 0.05, where P = 0.05 is the Shapiro–Wilk normality test. Figure 5 is a box-and-whisker plot of the Shapiro–Wilk normality test. P -values for GPS L1 SNR residuals averaged from all satellites for each station are calculated over different time lengths. The box-and-whisker plot is then made for all 500+ stations over different time lengths. The orange line on the box-and-whisker plot indicates the median, which corresponds to the P -value for the majority of stations. The red dashed line indicates a P -value of 0.05, which indicates a high probability that the distribution is not normal. A P -value greater than 0.05 indicates that the residuals tend to follow a normal distribution. This result indicates that SNR residuals tend to be normally distributed over short time periods. The p -values are greater than 0.05 for most of the stations over shorter periods. For longer periods ( > 6  h), the p -value is below 0.05 for most stations.

Figure 5

Boxplot of Shapiro–Wilk normality test of GPS L1 SNR residuals averaged from all tracked satellites. Shapiro–Wilk normalization test is performed using the averaged SNR residuals from all stations. The orange line in the boxplot is the median (where 50% of the stations are located), and the edges of the box are the upper and lower quartiles of 25%. The dashed red line in the plot indicates a P -value = 0.05, indicating that the distribution may not be normal.

Another characteristic of the SNR residuals is that they are uncorrelated between satellites tracked at the same time. Figure 6 shows the Pearson correlation of SNR residuals among tracked satellites for GPS signals L1, L2, and L5 for all stations during a single day. Cross-correlations are calculated between paired satellites, and an average is calculated from all pairs. The horizontal axis shows the list of stations, which are randomly distributed, and the vertical axis shows the average of the pairwise correlations. From this figure, it can be seen that the average correlation is about 0.1 and is less than 0.3 for all stations except for a green dot in the upper corner for the L5 band. It is not an outlier but for a station called 0GIS. Active L5-centered interference occurred at the 0GIS station on the specific day for which correlation analysis was performed (see Section 5.2 for details). This indicates that the SNR residuals are correlated between the tracked satellites in the presence of RFI. This is consistent with Roberts et al. (2022), which focused on interference detection from LEO satellites and noted that the signal degradation associated with RFI is a correlated change between signals tracked simultaneously.

Figure 6

Correlation coefficients of GPS L1, L2, and L5 SNR residuals for all SWEPOS stations. Cross-correlation is performed on a pair of simultaneously tracked satellites, and the average SNR for each pair is plotted.

4 SWEPOS RFI detection

An interference detection method that takes advantage of historical SNR measurements to predetermine the SNR characteristics of all GNSS signals from a given receiver is developed. It will be addressed through this study as the SWEPOS interference detection system. The process involves establishing a reference window using 2 days of interference-free data, while also considering the factors that can contribute to SNR drops. The inclusion of a 2-day data duration helps account for potential variabilities, diurnal cycles, and unknown factors associated with satellite orbit repeats. If there are equipment changes or firmware upgrades, a new reference window is defined. Furthermore, if long-term drifts are detected in the SNR time series, a new reference window is created. Once the reference data set is determined, the SNR model is developed for each receiver and GNSS frequency as described in Section 3.2.

Evaluation windows are established using incoming data to compare their distributions with the reference window. The coefficients obtained from the reference window model are employed to fit the evaluation windows, and subsequently, SNR residuals are computed for each satellite, following the methodology described in Section 3.3. The status of signal disturbance is reported by monitoring unexpected occurrences in the SNR residual values of each evaluation window. This is done by evaluating whether the evaluation window residual characteristics satisfy a predefined set of conditions and by comparing them with threshold values derived from the reference window.

Signal disturbances at all frequencies from all GNSS are monitored and reported based on the following characteristics. In the presence of RFI, e.g., when a jammer approaches a receiver:

• The noise increases and therefore causes unexpected SNR decreases. These unexpected drops in SNR values are monitored against the reference window.

• Variations in the SNR are correlated across tracked satellites.

The aforementioned key points are supported by comparing different frequency bands, monitoring the number of reachable satellites for a given receiver, and comparing between different receivers.

Once residuals are calculated for each tracked satellite, a pairwise Pearson correlation coefficient is calculated for simultaneously tracked satellites within the evaluation window time range ( T W ). T W is set to 10 s for 1 s sampling rate observations. The choice of T W depends on many factors. An important factor is that whether it is required to detect short duration pulses. As the SNR residuals are normally distributed in the absence of interference, the Pearson product–moment correlation ( r ) (Benesty et al. 2009) is used to calculate the cross-correlation between simultaneously tracked satellites.

For a given evaluation window length, T W , r for two simultaneously tracked satellites S i and S j ( r S i S j ) is calculated as:

For N simultaneously tracked satellites [ S 1 , S 2 , S 3 , … , S N ] , N p = N ( N − 1 ) ∕ 2 unique satellite pairs can be formed, and once the pairwise Pearson correlation coefficient is calculated for each pair, the average, r N p , is calculated as:

where ( i , j ) is the unique ( i ≠ j ) pair of satellites.

Moreover, the average of SNR residuals for simultaneously tracked satellites within the evaluation window is calculated as:

where S i SNR is the SNR residuals of satellite i .

For each GNSS, signal disturbances are reported according to predefined threshold values as follows:

• if E W SNR > − 2 dBHz , no signal disturbances are reported

• if E W SNR < = − 2 dBHz , signal disturbances are reported

  • – if r N p < = 0.3 , disturbances are recognized as RFI-related

  • – if r N p > 0.3 , disturbances are recognized as non-RFI-related

Non-RFI-related observational degradation, e.g., due to ionospheric scintillation, is highly unlikely to be correlated because GNSS signals may have different penetration points in the ionosphere (Roberts et al. 2022). This property is used to classify whether the signal interference is caused by RFI or other causes.

Consequently, a SWEPOS signal disturbance map is generated, which shows the status of all stations in the entire network in near real time, and if signal interferences are reported, e-mail alerts with status information are generated and sent to SWEPOS operations for further troubleshooting and risk management.

Strong RF interfering signals can cause complete signal loss, which will result in data gaps. However, gaps in data can also arise from power outages or data connection issues, which encompass problems or disruptions in the network or communication channels that hinder the reliable transmission of data. The online status of the stations, which is routinely checked every minute, is used to supplement the interference detection algorithm, which, in turn, helps to identify strong interference causing complete loss of signals from power outage-related data gaps. On-line status of SWEPOS stations is performed using network availability and performance monitoring WhatsUp Gold (WhatsUpGold 2022).

5 Results

This section presents a demonstration of the interference detection system with simulated interference waves and actual interferences detected by the algorithm.

5.1 Simulated interferences

As part of the effort to make an assessment on the potential of using SWEPOS for interference monitoring and detection, four different interference waveforms centered at GPS L1 (1575.42 MHz) were simulated in FOI GNSS-lab using GSS9000 Series GNSS simulator (Alexandersson et al. 2021). GNSS receivers were exposed to the waves in a controlled environment. The details of the simulation and the overall impact assessment have been demonstrated and reported in Alexandersson et al. (2021).

The four simulated interference waves were an additive white Gaussian noise (AWGN) with 20 MHz bandwidth, AWGN with 2 MHz bandwidth, a continuous wave (CW) unmodulated carrier, and a frequency-modulated (FM) wave. Figure 7 is the ramp, which shows the simulated waves plotted against time when the waves were simulated. The interference waves were simulated on 20 January 2021, which started at 09:00 and ended at 14:30 UTC.

Figure 7

Simulated interference waves plotted as noise (interference) power in RMS (dBm) as a function of time as in Alexandersson et al. (2021).

The interference waves were repeated on 21 January 2021 at a similar time, 09:00–14:30 UTC. As four different waveforms were simulated within the specified time interval, each interference waveform was initiated and terminated with a 5-min interval between each interference wave, leaving the receivers undisturbed and in steady state before the next interference was initiated and terminated.

The power of the simulated interference waveforms varied with time. From the beginning, the power was increased by 2 dB per minute, the maximum power was maintained for 5 min, and then, the power was decreased by 2 dB per minute. Three different receivers commonly used within the SWEPOS network were used for testing. Receivers tested were Trimble Alloy, Trimble NetR9, and Septentrio PolarX5. Alexandersson et al. (2021) described in detail how each receiver handles interference. Data from the Septentrio PolarX5 receiver are only shown in Figure 8.

Figure 8

Top graph shows the average raw SNR of all tracked GPS satellites for GPS L1, while the bottom graph shows the average SNR residuals of all tracked satellites. The dashed red line in the bottom plot is the SNR threshold at which interference is reported.

The top plot in Figure 8 shows raw SNR averaged from tracked satellites for GPS L1 plotted as SNR in dBHz against epoch for both days 20 and 21 January 2021. The bottom plot in the same figure shows SNR residuals, calculated as raw SNR data subtracted from the SNR model described in Section 3.2 plotted for the same duration as the top plot. The SNR model was created from the interference-free data. The red dashed line indicates the − 2 dBHz threshold. Four distinct spikes below the threshold line are observed on both days. These are linked to four different sources of interference in Figure 7.

The four spikes are linked in the order AWGN 20 MHz, AWGN 2 MHz, CW, and FM waves, respectively. It can be seen that the AWGN 2 MHz wave has the greatest effect in terms of reducing GPS L1 signal power and in turn the SNR. It is also noteworthy that the receiver succeeded in suppressing the influence of CW waves compared to other simulated waves. As can be seen from the figure, the detectability of all simulated waves is evident as they caused SNR drops below the threshold. Furthermore, the study investigates how the simulated interference waves affect individual satellites in comparison with other satellites tracked simultaneously.

Figure 9 shows a correlation matrix displaying the correlation coefficients between simultaneously tracked satellites. The correlation matrix is calculated from the interference-free data. Satellite pseudorandom noise (PRN) numbers are shown on the vertical and horizontal axes, and the legend illustrates the correlation values, with the color scale ranging from cooler tones (blue) to warmer tones (red), representing the strength of the correlations. The main diagonal is red because it is the covariance of each satellite with itself. As can be seen from the figure, the cross-correlation for each pair of satellites is almost zero. The average correlation for all pairs is − 0.01 , which indicates that interference-free SNR is not correlated between tracked satellites.

Figure 9

Correlation of SNR residuals for simultaneously tracked satellites when there is no interference.

Figure 10 shows the correlation matrix of the tracked satellites when the receiver is exposed to the simulated interferences. The upper-left, upper-right, lower-left, and lower-right plots show the correlation matrices when the receiver is exposed to AWGN 20 MHz, AWGN 2 MHz, CW, and FM waves, respectively. As can be seen from all plots, the SNR residuals are highly correlated across the tracked satellites. Averaged from all satellite pairs, the mean correlations are 0.94, 0.98, 0.86, and 0.92 for AWGN 20 MHz, AWGN 2 MHz, CW, and FM waves, respectively.

Figure 10

Correlation of SNR residuals for simultaneously tracked satellites in the presence of interference waves.

This means that different types of interference waves cause high correlations between the tracked satellites. Therefore, SNR thresholds and correlation analysis are very sensitive to different types of interference waves and can detect different interference sources, but cannot classify the type of RFI (e.g., identify whether the source is AWGN or CW). Overall, it can be seen that SNR residuals are highly correlated when interference is present.

5.2 Actual RFI

The SWEPOS interference detection system has detected hundreds of RF-related interferences within the SWEPOS network. Most SWEPOS stations are located in relatively quiet areas. However, some stations are located near cities or ports, where human activity somehow causes RFI. So far, all detected interferences have been unintentional, such as occasional human activity, nearby radio communications, or amateur radio.

Detected RFI range from weak, where there is no actual effect of the interfering object on the station, to very strong where it caused complete loss of signals, such as strong L5/E5 band interference causing receivers to track no Galileo satellites. Most of the detected disturbances were very short-lived, but there were also long-lasting disturbances that lasted for weeks or months.

An example is a strong L5/E5 band interference at one of the SWEPOS stations, Grisselhamn (0GIS), a station located in the marina of Grisslehamn, Sweden. Figure 11 shows a spectral diagram of the L5 band, which shows a strong peak of an interfering signal. The detected interference was a narrow band centered at 1182.6 MHz, which is almost 50 dB above the noise floor. Although the interfering signal is characterized by a narrow band, it affects a wider band of signals from 1176.45 MHz (GPS L5, GAL E5a, and BDS B2a signals) to 1207.14 MHz (GAL E5b and BDS B2b signals). The intensity and power of the disturbances used to vary, and it sometimes caused a complete loss of the L5 band and loss of tracking of the Galileo satellites.

Figure 11

Spectrum plot that depicts an interference detected in the L5 band at station 0GIS. The spectrum was obtained using FFT (fast Fourier transform) analysis. The raw signal data were windowed, optionally zero-padded, and transformed into the frequency domain using FFT. The resulting spectrum plot displays the power distribution across different frequencies, enabling the identification of signal characteristics and potential interference such as the peak at 1182.6 MHz in this plot. The vertical-colored lines indicate where the main GNSS signals are located in the spectrum.

Figure 12 shows the SNR residual correlation matrix of the tracked satellites in the presence of the narrowband interference at 0GIS. The upper-left and lower-left plots are for GPS L1 and Galileo E1 signals, while the upper-right and lower-right plots are for GPS L5 and Galileo E5a signals. It can be seen that SNR residuals are correlated among tracked satellites for GPS L5 and Galileo E5a during the interference. However, the SNR residuals are not correlated for GPS L1 and Galileo E1 because the interfering signal is in the L5 band and does not affect the L1 band.

Figure 12

Correlation of SNR residuals for simultaneously tracked satellites for GPS L1, Galileo E1, GPS L5, and Galileo E5a for station 0GIS during the presence of the narrow band interference in Figure 11.

The disruption at 0GIS continued for an extended period of time until the source was identified and located by the Swedish Post and Telecom Authority (PTS). Since the station is located in a marina, the source of the interference was found to be a boat with some kind of equipment that interferes with the GNSS L5 band. After the boat was moved to another location, the disturbances disappeared. However, there is not enough information about the type of equipment on the boat that caused the disturbance.

5.3 Non-RFIs

There are several factors that cause the SNR to degrade from its nominal value. Many of the factors are explained in Section 3.1. Most of the factors can be monitored and modeled well, but environmental changes around stations are very difficult to model. For example, tree foliages growing next to a station can cause signal degradation and may degrade the SNR. Depending on the type of leaf and the moisture level, foliage attenuation can be higher or lower. As a result, SNR drops from these causes can be a false signal for RFI-related SNR drops. Therefore, unmodeled SNR drops such as those due to ionospheric scintillations or tree foliage must be separated from RFI-related SNR degradation.

Statistical characterization of SNR among tracked satellites could be used to distinguish SNR drops of the former from RFI-related drops. Figure 13 shows the SNR residuals for Galileo E6 signal for a station from the SWEPOS network, Tived (0TIV). Signal interference was detected for the station. The red dots show the SNR residuals that fall below the threshold, and therefore, a disturbance was detected and reported for the station.

Figure 13

Galileo E6 SNR residuals averaged from tracked Galileo satellites over a day for station 0TIV.

An elevation-azimuth diagram of the raw SNR data is shown in Figure 14. The color code in the figure shows the SNR values scaled according to the legend. It can be seen that the SNR values falling below the threshold have a localized nature, located at 0–50 and 290–360 degrees azimuth. When inspecting the location of the station, it could be noted that there are large trees in the above-mentioned azimuth angles, which are likely the cause. Such types of disturbances are not correlated among simultaneously tracked satellites, as can be seen in Figure 15, which is a correlation matrix of the SNR residuals during the epochs where the residuals are below the threshold. It can be seen that the average correlation calculated from each station pair cross-correlation is − 0.01 . Consequently, the detection system identified the disturbance as a non-RF-related interference.

Figure 14

Diagram depicts the elevation-azimuth mapping of individual Galileo satellite orbits, showcasing the raw SNR of the E6 signal. Derived from a 20-day dataset, the diagram’s colors represent the varying SNR values.

Figure 15

SNR residuals correlation matrix for epochs 18:00 to 19:00 in Figure 13.

5.4 Equipment-related disturbances

Not all interference sources can be easily classified as RF- or non-RF-related as there may be other factors that can cause correlated interference between satellites. Faulty equipment, such as a faulty antenna, can lower SNR values, and they can be correlated between tracked satellites. An example is a detected interference at one of the SWEPOS stations called Rosvik (0ROS). The station was equipped with TRM59800.00 antenna and Septentrio PolarX5 receiver when the interference was detected. Figure 16 shows the spectral plot of the L5 band for the station. It is clear from the figure that there are strong peaks that reach up to 50 dB above the noise floor. Peaks could also be seen on the L1 band (image not shown).

Figure 16

Spectrum plot of an interference detected in the L5 band at station 0ROS. The vertical colored lines indicate where the main GNSS signals are located in the spectrum. The spectrum was obtained using FFT as described in Figure 11.

When the interference was detected in the station, a visit was made to the station site with an extra antenna and a receiver to troubleshoot the source of the interference. The new antenna was installed near the station. However, the interference that could have been detected at station 0ROS could not be detected with the new antenna that was installed nearby.

However, further analysis was able to conclude that interference occurred at the station when the temperature was low.

Figure 17 shows a monthly signal interference report for the station. Red markings indicate the days when interferences were reported, while gray days are interference-free days. The numbers in the middle of the boxes indicate the daily low temperatures in degrees Celsius, while the numbers in the corners of the boxes are the day numbers of the month. It can be seen that interferences were reported for the station when the temperature was relatively low, which is likely to be linked to the aging and temperature sensitivity of the antenna. An antenna replacement was performed, and the interference has not recurred since then.

Figure 17

Monthly interference report for February 2022 for station 0ROS. Red squares show the days where disturbances were reported, while gray squares show the disturbance-free days. Numbers in the middle of the squares are daily low temperatures in degrees Celsius.

6 Discussion and conclusion

GNSS interference poses a continuous threat to both GNSS systems and infrastructures that rely on GNSS technology. This threat is attributed to a growing prevalence of interference from both intentional and unintentional sources, impacting both GNSS systems and the infrastructures dependent on GNSS technology. In addition, critical applications such as self-driving vehicles and drones are emerging that cannot tolerate even a few seconds of GNSS interference or non-availability. Interference detection and situational awareness is the first critical step in managing the risk of interference in GNSS and GNSS-dependent infrastructures.

An interference detection system using GNSS RINEX data from the entire SWEPOS network is developed. The method makes use of SNR; it monitors for unexpected changes in the SNR values and characterizes how these changes correlate between satellites tracked at the same time. It has been shown both using simulated interference waves and actual interference cases that the method can detect and is sensitive to different types of interference and also capable of distinguishing unexpected SNR changes due to other factors, such as tree leaf attenuation, from RFI-related SNR changes. The disturbance detection system monitors the entire SWEPOS network containing more than 500 stations in near real time, and it produces a signal disturbance status map and sends e-mail alerts if disturbances are detected.

The additional goal of this study is to demonstrate the leverage of an existing infrastructure, such as the SWEPOS CORS network presented here, to control the quality of GNSS signals and to be able to detect disturbances in GNSS and neighboring frequency bands. Signal interference in satellite-based navigation systems such as GNSS is a particularly serious threat to society, and a detection system of the kind presented here covering an entire country plays its own part in the overall picture for emergency management use and providing clean GNSS spectrum.

Future work includes improving the interference detection system with better multipath filtering techniques and supplementing it with other data sources such as AGC for better detection and classification of interference sources.