Abstract
Radio frequency interference (RFI) in global navigation satellite system (GNSS) frequencies can endanger human life and safety by preventing the use of these signals for navigation and positioning, in addition to degrading measurements for science applications. Here, we use data from the Cyclone GNSS (CYGNSS) constellation to map GNSS RFI from 2017 to 2022, identify the location of several potential sources of RFI, and quantify the duration of transmission. Although our method of RFI detection can only provide a rough approximation of transmitter positions, it is possible that advanced data processing techniques could better pinpoint their locations, once guided by these observations. We find that, since the launch of CYGNSS, GNSS jammers have proliferated across the world and are often associated with the beginnings of geopolitical unrest. Our results agree well with previous studies that have also used satellite observations to map ground-based RFI transmission.
1 INTRODUCTION
Global navigation satellite system (GNSS) signals, such as those transmitted by global positioning system (GPS) satellites, are used by both military and civilian users for navigation, positioning, and scientific purposes. Although signal transmission in GNSS bands by unauthorized persons is restricted, radio frequency interference (RFI) in the GNSS wavelengths can occur, either intentionally or unintentionally, which can limit or prevent the use of GNSS signals for their intended purpose. Often seen in conflict zones, RFI can be caused intentionally by transmitters sending out noise at the carrier frequency (jamming) or false or modified pseudorandom noise (PRN) codes to trick a receiver into calculating a false position (spoofing). However, not all cases of GNSS RFI are nefarious in origin, and there have been reports of RFI caused by faulty instrumentation or civilian communication systems (Berglund et al., 2011; National Academies of Sciences, 2023).
One widely published source of GNSS RFI was first described by Murrian et al. (2021). Using an instrument on the International Space Station, the authors were able to pinpoint the location of a transmitter to a Syrian airbase, which has been a constant source of GNSS RFI since 2017. Murrian et al. (2021) indicated additional potential sources of RFI in both Libya and China, although they did not pinpoint their locations. Since the publication of Murrian et al. (2021), other studies have also mapped RFI from terrestrial sources using satellites in low earth orbit. For example, Roberts et al. (2021) used data from the GRACE and COSMIC satellites to create yearly average global maps of possible locations of RFI transmission. Roberts et al. (2021) also noted RFI over Syria and Libya, in addition to several hotspots over both Europe and Russia.
Here, we add to the body of work identifying GNSS RFI hotspots around the world and identify the approximate locations and durations of transmission of terrestrial GNSS jammers and spoofers, some of which have not been identified in previous studies. To achieve this aim, we use data from the National Aeronautics and Space Administration (NASA) Cyclone GNSS (CYGNSS) constellation, which is a set of eight modified GPS receivers in low earth orbit that were designed to receive Earth-surface-reflected signals for the retrieval of ocean surface wind speed. In the sections that follow, we describe the CYGNSS data used as an indicator for RFI, show that the data can indicate the location of previously mapped GNSS jammers, and then comment on other potential terrestrial RFI sources, their duration of transmission, and possible reasons for their existence.
2 BACKGROUND
CYGNSS was launched in December 2016 as NASA’s first Earth Venture Mission. Originally designed for the retrieval of hurricane force winds over the ocean, CYGNSS utilizes a remote sensing technique called GNSS-reflectometry (GNSS-R). The GNSS-R technique records GNSS signals that have reflected off the surface of the Earth for the purpose of remote sensing. Each CYGNSS satellite contains two downward-looking GPS antennas, each of which are inclined at an angle of 28° off nadir. These antennas passively record surface-reflected GPS signals and are thus susceptible to RFI. Because CYGNSS was designed to quickly sample winds during hurricanes and tropical storms, the satellites are in an equatorial orbit, and the data collected by CYGNSS are limited in latitude to approximately ±40°, which also limits the latitudinal extent of our analysis.
CYGNSS records delay Doppler maps (DDMs), which are two-dimensional cross correlations of the received signal and a locally generated replica that has been altered for a variety of time delays and frequency shifts. An example of a DDM is shown in Figure 1(a). The peak value typically occurs at the specular reflection point, which corresponds to the shortest-path distance between the transmitter, ground, and receiver on the surface. Power can exist in other bins as well, often creating a “horseshoe” pattern in the DDM, with rougher surfaces producing more pronounced horseshoes.
Four examples of DDMs recorded by CYGNSS
All pixels above the red line in each DDM are used to calculate the noise floor. (a) DDM recorded over the Bahamas. (b) DDM recorded over Saudi Arabia with RFI present (striped lines). (c) DDM recorded over land at a surface elevation of 1350 m. (d) DDM recorded over calm inland surface water. Green ellipses indicate sidelobes of the autocorrelation function often found in coherent reflections.
RFI in CYGNSS DDMs appears as stripes of high power in one or more Doppler bins and across all or most delay bins. An example of RFI stripes is shown in Figure 1(b). In some cases, the reflected signal can still be seen in the DDM; yet, in other cases, the DDM is entirely dominated by RFI and overwhelms the reflected signal, particularly when the RFI stripes are present in the same Doppler bins as the reflected signal.
CYGNSS currently flags DDMs with possible RFI by using the kurtosis of the noise portion of the DDM, which is defined as the pixels in the DDM in the first 45 rows of the delay bins (i.e., all pixels above the red lines shown in Figure 1). Also called the “forbidden zone,” this region of the DDM should not contain any information about the reflecting surface. Only summary statistics of the forbidden zone, such as the kurtosis and mean value, are normally downlinked from the satellites. The values at each pixel are only occasionally saved for special “full-DDM” collections, some of which comprise the DDMs in Figure 1. Kurtosis is a measure of the “tailed-ness” of the data distribution. Data with high kurtosis have more extreme values than data with low kurtosis. The RFI flag for CYGNSS is created by thresholding kurtosis values: any DDMs with a kurtosis greater than 4.0 in the forbidden zone are flagged as RFI. However, the CYGNSS RFI flag was developed for use over the ocean, and there are two reasons why the current RFI flag does not work well over land.
The first reason is that the CYGNSS tracking algorithm uses the mean sea surface for predicting where the surface reflection should occur in terms of delay. Over land, higher surface altitudes cause deviations from the predictions, and the surface reflection occurs in the forbidden zone when the elevation exceeds a few hundred meters above sea level (e.g., Figure 1(c)). In December 2017, CYGNSS changed its DDM compression algorithm such that the cropped DDM still contains the surface reflection for these higher-elevation areas; however, the first 45 rows of delay are still used for calculating the RFI flag. Thus, the kurtosis of the noise floor is always artificially high, triggering the RFI flag in these areas. The second reason that the RFI flag fails over land is the presence of sidelobes of GPS autocorrelation in DDMs recorded over surfaces that produce strongly coherent reflections (e.g., calm inland surface water). An example of these sidelobes is shown by the green ellipses in Figure 1(d). Sidelobes also produce high kurtosis that is not due to RFI. The effect of these two confounding factors over land on the kurtosis of the DDM (and thus the RFI flag) is shown in Figure 2, where high-altitude areas tend to produce kurtosis values >10 and inland water bodies, particularly over the Amazon or Mississippi Basin, often produce kurtosis values >6.
Mean kurtosis of the forbidden zone for the first two weeks of 2019, gridded to 36 km
3 METHODS
3.1 Processing the Mean Noise Value
Because of the difficulties involved in using the standard CYGNSS RFI flag over land, we will use a different method for characterizing RFI that is less sensitive to higher surface elevations and inland surface water. Instead of kurtosis, we will use the mean value of the forbidden zone, also called the mean noise value, which is a standard parameter provided in the Level 1 CYGNSS files. The mean noise value is the mean of all pixels in the forbidden zone of the DDM. Although the presence of surface reflections and sidelobes in the forbidden zone will also affect the mean value, we will show that the mean value is not affected nearly as strongly by these features as is the kurtosis.
In data recorded prior to July 2019, DDMs were incoherently integrated over 1 s; in July 2019, the incoherent integration time was shortened to 0.5 s to increase the spatial resolution of the reflection. This caused an immediate 3-dB drop in the noise value for all observations. To remove this effect, we add 3 dB to all observations that we show after the 1- to 2-Hz transition. In all of the following figures, we grid our mean noise values to a 9-km equal-area scalable Earth version 2 (EASE-2) grid, after first calculating the position of the antenna boresight on the ground (see the Appendix for a description of this calculation), which accounts for the inclination of the antennas.
3.2 Noise Spatial Characteristics
Figure 3 shows the average DDM noise value for the year 2018. Here, we split up the mean noise values both by antenna (starboard or port) and by whether the satellite was ascending or descending. Compared with the kurtosis map in Figure 2, high-altitude areas do not noticeably affect the mean noise of the DDM, and inland surface water only minimally affects the mean noise value. Land exhibits less noise overall than the ocean, as noted in other studies (Foti et al., 2017). Foti et al. (2017) hypothesized that higher noise over the ocean in general could stem from other L-band signals reflecting strongly over the water surface, although this has not been proven. Hotspots with much higher noise over the ocean are associated with the presence of satellite-based augmentation system (SBAS) signals (Gleason et al., 2020).
Mean noise in 2018 (gridded to 9 km)
Mean noise for ascending (a) and descending (b) tracks of the starboard antenna. (c) and (d) are the same as (a) and (b), except for the port antenna. Circular spots of elevated noise over the ocean have been attributed to SBAS signals. The wide swath of elevation noise in data collected by the port antenna is likely caused by a GNSS jammer in Syria.
Noise data recorded by the port antenna in Figure 3 show a notable wide swath of elevated noise levels over land north of the Arabian Peninsula, with a maximum value exceeding 45 dB. We will associate such hotspots with sources of terrestrial GNSS RFI. A similar noise hotspot in a different location was previously attributed to RFI by Foti et al. (2017), using data from TechDemoSat-1, which also carried a GNSS-R instrument. The swath of high noise in data collected by the port antenna in Figure 3 is only barely detectable in data from the starboard antenna. Further, depending on whether the port antenna is ascending or descending, the high-noise region has a tail that is oriented to the southeast for ascending tracks and southwest for descending tracks. These differences can be attributed to the direction in which the antennas are pointing—if the noise source is always to the north of the CYGNSS satellites, then the starboard antenna, which always faces south, will never be pointed in the direction of the noise. Similarly, the high-noise trail present in the port antenna data will always be in the direction in which the port antenna is facing.
Because the CYGNSS satellites have two antennas facing opposite directions, which change depending on whether the satellite is ascending or descending, noise data recorded over time provide several different viewpoints, which can map an approximation of the source location if all viewpoints are averaged over time. In all subsequent figures, we show noise data that have been averaged for both antennas and all orientations, which will better indicate the location of RFI transmitters.
3.3 Noise Hotspots and RFI: Known Transmitter Location
To provide evidence that spurious high noise in CYGNSS DDMs is very often indicative of RFI, we first start with an analysis of the high noise over the Middle East that was apparent in the data recorded by the port antenna in Figure 3.
Mean noise values for both antennas and all orientations for the year 2018 are shown in Figure 4(a). Note that, because of the antenna pointing corrections, data from the starboard antenna do not reach as far north as those from the port antenna; likewise, data from the port antenna do not reach as far south as those from the starboard antenna. For this reason, there is a small but noticeable shift in the mean noise levels for latitudes outside of ±33°. In contrast to the data for ascending and descending tracks from the port antenna alone in Figure 3, the high-noise region in Figure 4(a) is confined to a smaller, more circular area. This region is the same area in which RFI has been previously mapped by other sensors (Murrian et al., 2021), and the location of the GNSS jammer is already known (black dot in Figure 4(b)). This location coincides with the highest noise values recorded by CYGNSS, and thus, we are confident that the cause of the high noise levels is the RFI observed in that location. Note that, unlike the work reported by Murrian et al. (2021), we are unable to pinpoint the location of the jammer beyond the city level when using this simple method. However, it is possible that using more advanced data processing techniques, specifically using raw intermediate-frequency data, could enable one to precisely pinpoint the transmitter location after an RFI hotspot has been identified.
Mean noise for the year 2018
(a) Mean noise for the year 2018 for both antennas and all orientations. Black dots indicate the approximate locations of the RFI transmitters discussed in this study. The black outlined box shows the region in (b). The black dot in (b) is the location of a GNSS jammer that was previously determined by Murrian et al. (2021).
4 RESULTS
The Syrian transmitter has continued to operate until the time of writing (September 2023); however, since 2018, several other notable sources of RFI have appeared. We now describe some of the most significant and persistent examples and comment on potential reasons for their appearance. The approximate locations of the transmitters described in the following sections are indicated by the other black dots in Figure 4(a)—note that, in this figure, none of these locations had anomalously high noise in 2018.
Figure 5 summarizes notable RFI hotspots and their durations, some of which are described in the subsections below. We visually identified these hotspots while looking for regions in which noise varied over time and then confirmed that the regions did not coincide with inland water bodies. The mean noise and its temporal variation in each identified hotspot differ, and in nearly all cases, the mean noise is not as high as the noise observed over Syria. Thus, it is possible that RFI in these locations might not be strong enough to seriously degrade GPS positioning solutions or data collected for scientific purposes, at least from satellites in low earth orbit. While we did not develop a change-detection algorithm to identify RFI hotspots here, future research efforts could do so.
Time series of the appearance and disappearance of RFI hotspots in the CYGNSS data, along with approximate locations
Each dot represents a two-week period.
4.1 A Mobile RFI Transmitter in Libya
In the first half of June 2019, another hotspot appeared with a noise amplitude rivaling that emitted over Syria. This hotspot was centered east of Tripoli, Libya, near the city of Khoms (Figure 6(a), green star). This hotspot is coincident with that mentioned in Murrian et al. (2021), and reports of GNSS RFI in this general region corroborate our attribution of this hotspot to a terrestrial RFI source (United States Coast Guard, n.d.). Noise remained elevated in this region until the second half of June 2020, at which time the hotspot temporarily disappeared, only to reappear a few days later. When it reappeared, the hotspot had moved to the east, closer to the town of Sirte (green star in Figure 6(b)).
Mean noise for selected time periods for both antennas and all orientations
(a) June 5, 2019–December 31, 2019; (b) June 13, 2020–December 31, 2020; (c) April 1, 2022–April 30, 2022; (d) December 1, 2022–December 31, 2022. Black dots indicate the approximate locations of the RFI transmitters discussed in this study. Colored stars indicate the specific transmitters discussed in the text.
The RFI over the new location also had a different spatial pattern than that for the first location. Although the diameter of the main RFI hotspot was approximately the same as before, there was an additional quasi-bullseye pattern surrounding the main hotspot (Figure 6(b),(c)), the reason for which is unknown.
Although definitively attributing an increase in RFI to a particular event is speculation, the appearance of the RFI transmitter in June 2019 did coincide with an increase in fighting near Tripoli during the Second Libyan Civil War (Lacher, 2019). Moreover, the disappearance/reappearance of the RFI source to the city of Sirte exactly coincides with a shift in the fighting from Tripoli to Sirte in June 2020 (Al Jazeera, 2020; al-Warfalli, 2020). It is possible that the RFI source was a powerful KORAL electronic warfare system used by Turkish forces against the Libyan National Army (Bakir, 2021; Rondeaux et al., 2021); however, the reports of its destruction in July 2020 do not agree with our observations of continued RFI in the region for the next two years.
4.2 RFI in Caracas, Venezuela
Libya was not the only new source of high noise and likely GNSS jamming in 2019. On January 20, 2019, a small hotspot appeared over Caracas, Venezuela, which persisted until the end of the year (yellow star in Figure 6(a)). Since then, the hotspot has been intermittent, disappearing for months at a time only to reappear later. Although significantly lower in power than the hotspots observed over Syria and Libya, the circular region of high noise is still clearly present in the data. The hotspot appeared on January 20, ten days after the start of nationwide protests against Venezuela’s leader, Nicolas Maduro, and one day before a failed military coup (Daniels & Zuniga, 2019).
4.3 Hotspot over Yemen
Although not clearly centered over a major metropolitan area, a hotspot similar in size to that over Caracas also appeared over Yemen in 2019 (black star in Figure 6(a)). This hotspot first appeared in mid-February 2019 and disappeared in January 2020. Yemen has been engulfed in a civil war for the past several years that continues to this day.
4.4 RFI Activity in 2022
2022 saw the appearance of several potential RFI hotspots (Figure 6(c)), most of which are associated with increasing geopolitical unrest. For example, the Tigray War on the border of Ethiopia and Eritrea (green star in Figure 6(c)) was associated with high noise in April 2022, which first appeared in November 2020. A similar hotspot in the Kashmir region (yellow star in Figure 6(c)) first appeared in June 2021 and has since persisted, only occasionally disappearing. Kashmir has experienced conflict for the past several years, as both Pakistan and India vie for control of the region.
Small transient hotspots also appeared over northeastern China (black star in Figure 6(c)) and Somalia near Mogadishu (purple star in Figure 6(d)) in 2022; however, it is unclear what led to their appearances. The hotspots were both short-lived, although the Somalian hotspot notably reappeared toward the end of 2022. Somalia has been experiencing civil war, similar to many other countries mentioned here, although there were no notable reports of an increase in conflict that coincided with the appearance of the hotspot in June 2022.
A low-power but notable hotspot appeared over the Central African Republic from January to June 2022 and then reappeared in mid-November of that year (red star, Figure 6(c)). Several news reports (Salih & Burke, 2023; Serwat et al., 2022) have indicated that Russian mercenaries from the Wagner group are present in this country. While possibly coincidental, when researching potential reasons for the appearance of RFI in the regions described here, we found that the Wagner group was mentioned as being active in most of these regions (e.g., Libya (Barabanov & Ibrahim, 2021), Venezuela (Tsvetkova & Zverev, 2019), and Yemen (Fasanotti, 2022)). However, although Russia is known to use GPS jammers in the war in Ukraine (Seligman, 2023; Tammik, 2023), we were unable to find any evidence that the Wagner group also has access to these jammers.
A generalized RFI hotspot also appeared over Myanmar in mid-December of 2022 and has persisted until the time of writing (Figure 6(d), red star). Although Myanmar has been embroiled in conflict for the past several years, we were unable to find any significant reports that coincided with the appearance of the RFI.
Finally, the Libyan RFI hotspot disappeared at the end of October 2022, after persisting in the country for more than two years (Figure 6(d)), though at a lower power for the majority of 2022 (Figure 6(c)).
5 DISCUSSION
An open question is the extent to which sources of GNSS RFI can disrupt the use of GNSS for navigation, positioning, and scientific uses. Although several of the RFI hotspots we have identified in this paper have been associated with reports of GNSS outages by aviators and sailors (e.g., Syria and Libya (International Civil Aviation Organization, 2022)), we were unable to find notable reports of GNSS interference in the hotspots over Venezuela and Yemen, which did not exhibit CYGNSS noise as high as that in either Syria or Yemen. It is possible that the less powerful RFI hotspots do not significantly disrupt the use of GNSS for navigation in these areas or at least do not cause widespread outages.
The CYGNSS data themselves will also be variably affected by the RFI sources we have described. If the magnitude of the surface reflection exceeds the magnitude of the noise incurred as a result of RFI, then the data may likely still be used for geophysical retrievals of wind speed, soil moisture, or inundation extent without concern. Within smaller RFI hotspots, where the RFI magnitude may only be 1–2 dB above the expected background value, the signal magnitude of most surface reflections will greatly exceed the RFI magnitude and thus overshadow the RFI. However, RFI significantly impacts data collected by CYGNSS in the Middle East, as surface reflections rarely result in data with a reflectivity higher than the strength of the RFI. Users of CYGNSS data should be particularly cautious when interpreting data from this region, as the current RFI flag does not consistently remove impacted data. Ongoing research aimed at reducing the impacts of RFI on spaceborne GNSS-R data (e.g., (Wu et al., 2022)) will be important for future GNSS-R missions.
6 CONCLUSION
The appearances of elevated noise hotspots in CYGNSS data are associated with geopolitical conflict and correspond to the locations of previously mapped GNSS jammers. Since the launch of CYGNSS in 2016, RFI hotspots have proliferated, and mapping the location and duration of transmission is important in order to monitor their impact on positioning solutions and other scientific applications. Although the data we have presented here cannot pinpoint the exact location of these transmitters, it is possible that more advanced data analysis techniques could do so.
HOW TO CITE THIS ARTICLE
Chew, C., Roberts, T. M., & Lowe, S. (2023). RFI mapped by spaceborne GNSS-R data. NAVIGATION, 70(4). https://doi.org/10.33012/navi.618
ACKNOWLEDGMENTS
The authors thank the University of Michigan for helping fund this effort under the NASA CYGNSS Extended Mission contract 80LARC21DA003.
APPENDIX
Two downward-pointed antennas are present on each CYGNSS spacecraft, one on the port side and one on the starboard side. These antennas do not point straight down; they are inclined at an angle of 28° (see Figure A1 for an illustration). This inclination causes geolocation offsets when noise is plotted as a function of spacecraft position on the surface, as illustrated in Figure A2. Figure A2 shows the mean noise value for the port (Figure A2(a)) and starboard (Figure A2(b)) antennas. Relative to the coastlines, there is an obvious shift in the noise of the land and ocean surfaces, which is caused by the difference in spacecraft position relative to the location at which the antenna boresight is pointing.
We applied a simple geolocation correction for the port and starboard antennas to better represent the location on the surface to which the antenna boresight is pointed. Given the position of the spacecraft and the inclination of the antenna, we can approximate the location on the surface to which the antenna is pointing as follows:
Determine the offset (in km) between the spacecraft position and the location at which the antenna boresight is pointing (scalt = spacecraft altitude):
1Determine the angles of the port (p) and starboard (s) antennas:
2
FIGURE A1Conceptual illustration of how the antenna direction affects the geolocation of the received noise for the port (north-pointing) antenna for both side (a) and plan (b) views
If plotted as a function of spacecraft position, the noise values will be shifted south of the source for the port antenna and north of the source for the starboard antenna. This applies for both the effects of land surfaces on the received noise, as shown here, and point RFI sources, which are the subject of this investigation.
FIGURE A2Mean noise in 2018 for ascending tracks of the starboard (a) and port (b) antennas
Note that noise over Madagascar (black box) and other land surfaces is shifted north for the starboard antenna and south for the port antenna.
Quantify how many kilometers are in one degree of longitude, which depends on the latitude of the spacecraft (sclat):
3Determine the latitude (latp) and longitude (lonp) of the location at which the port antenna is pointing, based on the latitude and longitude of the spacecraft (sclat, sclon). Determine these values for the starboard antenna as well (lats and lons):
4
5
6
7
Once we apply the geolocation correction to both the port and starboard antennas, we find that the location of the noise has shifted to better align with the land/ocean interfaces (i.e., Figure 3 in the main text).
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
Jump to section
Related Articles
Cited By...
- No citing articles found.