Analysis and Characterization of an Unclassified RFI Affecting Ionospheric Amplitude Scintillation Index Over the Mediterranean Area

Radio frequency (RF) signals transmitted by Global Navigation Satellite Systems (GNSSs) are exploited as signals of opportunity in many scientific activities, ranging from sensing waterways and humidity of the terrain to the monitoring of the ionosphere. The latter can be pursued by processing the GNSS signals through dedicated ground-based monitoring equipment, such as the GNSS Ionospheric Scintillation and Total Electron Content Monitoring (GISTM) receivers. Nonetheless, GNSS signals are susceptible to intentional or unintentional RF interferences (RFIs), which may alter the calculation of the scintillation indices, thus compromising the quality of the scientific data and the reliability of the derived space weather monitoring products. Upon the observation of anomalous scintillation indices computed by a GISTM receiver in the Mediterranean area, the study presents the results of the analysis and characterization of a deliberate, unclassified interferer acting on the L1/E1 GNSS signal bands, observed and captured through an experimental, software-defined radio setup. This article also highlights the adverse impacts of the interferer on the amplitude scintillation indices employed in scientific investigations, and presents a methodology to discriminate among regular and corrupted scintillation data. To support further investigations, a dataset of baseband signals samples affected by the RFI is available at IEEE DataPort.

scale irregularities embedded in the Equatorial Plasma Bubbles (EPB) (see e.g.[10]- [14]).At mid latitude, ionospheric scintillations can be due to poleward expansion of the crests of the Equatorial Ionization Anomaly (EIA) [15] or equatorward expansion of the auroral oval during geomagnetic storms [16].Very few cases of mid latitude GNSS scintillations during quiet times are reported in the literature [17].By exploiting the GNSS signals transmitted by Medium-Earth Orbit (MEO) and Geostationary-Earth Orbit (GEO) satellites as signalsof-opportunity, it is possible to investigate the ionospheric irregularities for scientific purposes, as well as to monitor ionospheric scintillations in the framework of operational space weather services [18].This is achieved by means of groundbased passive instruments, such as the GNSS Ionospheric Scintillation and TEC Monitor (GISTM) receivers [19] which provide the estimation of the so-called amplitude and phase scintillation indices (S 4 and σ ϕ respectively), allowing to quantify ionospheric scintillations [20].Besides ionospheric irregularities, however, a numbers of different phenomena related to both space weather events (e.g.Solar Radio Burst [21], [22]) and environmental conditions, may impair the GNSS signals and the detection of ionospheric scintillations.A well-recognized source of error in the computation of the scintillation indices is the reception of GNSS signals from multiple paths due to the reflections caused by obstacles in the proximity of the receiving antenna, known as multipath [23].
To compensate for such phenomena, GISTM receiver antennas are typically deployed in multipath-free conditions, i.e., isolated areas with limited natural or anthropogenic obstacles, and elevation masks can be configured to neglect mulipathsusceptible signals received from low-elevation satellites [24].
Similarly to the multipath, misleading effects on navigation signals and the derived scintillation indices can also be observed due to intentional or unintentional in-band Radio Frequency Interference (RFI)s, captured by instruments' receiving antennas [23], [25]- [27].These interferences are typically attributed to malicious actions aiming at disrupting GNSS receivers' operational activities by forcing misleading Position, Velocity, Timing (PVT) estimation, degrading their estimation accuracy up to cause a denial of their Positioning, Navigation and Timing (PNT) capabilities (a.k.a.Denial-of-Service (DoS) attack) [28].These attacks are classified as spoofing, meaconing, and jamming, with the first aiming at fooling receivers' operations by transmitting plausible yet fake GNSS signals, and the latter aiming at transmitting structured or unstructured Radio Frequency (RF) signals to  1a) and detail of Lampedusa island (Italy) showing the position of the ENEA observatory and other areas of interest (Fig. 1b).
ditions.Moreover, the political and environmental situation of Lampedusa may favor deliberate RF transmissions against navigation and communication systems: the island hosts military settlements and NATO radar equipment, a civilian and military airport, and is a hotspot of irregular migratory flows from the coast of North Africa [41], [42].Furthermore, possible RFIs in the area were detected in the second semester of 2020 by Airbus aircrafts [43] and a recent paper has highlighted intense RFIs in the Mediterranean region by analyzing the data of the GNSS receivers carried by GRACE Follow-On (GRACE-FO) Low Earth Orbit (LEO) satellites [44].
Moving from the know-how gathered during previous, joint test campaigns and activities [34], [45], [46], a renewed, SDRbased hardware and software architecture was designed and implemented to perform long-term grabbing of GNSS RF signal samples in the attempt to identify and characterize the source of the disturbances.
The main contributions of the article are the following: • we prove the presence of an interferer affecting the GNSS signal in the Lampedusa area and present a characterization of the RFI through the analysis of the IF samples acquired by the dedicated SDR architecture.We discuss the impact of such interference on the estimation of the amplitude scintillation index and propose an analytic model of the interferer, which may allow for further theoretical analyses and the development of mitigation techniques.
• we assess the adverse impact of the RFIs on the scintillation data computed through the GISTM receiver, which may impair both near real-time monitoring applications as well as scientific investigations of ionospheric scintillation.At the time of writing, on-field proofs of such a vulnerability are still undocumented in the literature.We also propose a preliminary methodology to automatically detect and filter the interfered observation from the collected data.
The article is organized as follows: Section II provides background information about the computation of scintillation indices through GNSS signals in GISTM receivers.Section III presents a preliminary analysis of the anomalies detected in the where x GNSS,fc is the sum of the received GNSS signals from the visible satellites at the receiver location for a given bandwidth and center frequency f c [47], and where Wide-Band Power (WBP) and Narrow-Band Power 218 (NBP) are respectively defined as and and the I and Q terms in (3) where C/N 0 is the estimated carrier-to-noise ratio [48], and

A. Lampedusa GISTM station
The ionospheric observatory of Lampedusa hosts, since 2018, a Septentrio PolaRx5S GISTM receiver.The Po-laRx5S is a multi-frequency, multi-constellation GNSS receiver equipped with a low-noise Oven Controlled Crystal (Xtal) Oscillator (OCXO).It acquires, for every satellite in view and for every available frequency, the raw phase (in cycles) and post-correlation I p and Q p samples with a sampling rate of 50 Hz, as per the generalized architecture presented in Section II-A.It is able to provide, with a 1-minute resolution, the S 4 and σ ϕ indices together with the Total Electron Content (TEC) and its Rate of Change (ROT).The data acquired by the station are transmitted in near-real time to the INGV-SWIT (Space Weather Information Technology) system and collected into a database publicly accessible to the scientific community through the eSWua (electronic Space Weather upper atmosphere: eswua.ingv.it)website [50].These data are also provided to the PECASUS consortium (www.pecasus.eu)for the provision of Space Weather services to the International Civil Aviation Organization (ICAO) [18].

B. Investigation about the S 4 anomalies
The following analysis focuses on the scintillation indices recorded by the GISTM receiver during August 2021 wherein several anomalies were observed in the collected data.In order to avoid misleading contributions possibly caused by multipath-effects, only satellites with elevation above 30°are considered; indeed, the Lampedusa observatory is located nearby a lighthouse, whose building was proven as a nonnegligible source of multipath for those signals acquired at lower elevations, as it will be shown in the results of Section V-C.The area observed by the receiver, considering this elevation mask, cover the mid-latitudes between 30°N and 40°N and a longitudinal sector between 7°E and 19°E.The signals taken into consideration are the one belonging to the Global Positioning System (GPS), Galileo, BeiDou Navigation Satellite System (BDS) and GLONASS constellations.The reported S 4 and σ ϕ indices are the slant values calculated at 1-minute resolution from the L1/E1 frequency band for each satellites in view in the considered timespan.Fig. 3a and Fig. 3b reports the maximum hourly values of the S 4 and σ ϕ respectively, recorded during August 2021.As it is possible to see from Fig. 3a, several occurrences of the S 4 above the threshold of moderate scintillation (lower dotted red line in Fig. 3a and Fig. 3b) recurred during the month; the same behavior was not registered for the σ ϕ (Fig. 3b).
The observed values of the S 4 are definitely unexpected considering i) the latitudes covered by this analysis and ii) the overall space weather conditions registered during the month of August 2021.Indeed, as mentioned in Section I, ionospheric scintillation at the Mediterranean latitudes are not common and are generally caused by disturbed space weather conditions [15], [16], [51], [52] originating the socalled super fountain effect [53].However, as Fig. 3c shows, no relevant geomagnetic storms capable to induce a poleward expansion of the crests of the EIA were detected during August 2021 according to the local K-index recorded at the INGV Geomagnetic Observatory of Lampedusa [54], [55].It is worth recalling that the K-index quantifies the disturbances in the horizontal component of the magnetic field with respect to the quite conditions and can be employed as an indicator of the intensity of geomagnetic storms measured at a given geomagnetic observatory [56].Usually, K-index values below 4 are representative of quiet/low-disturbed conditions, while values from 5 to 9 indicate minor to extreme storm conditions, respectively.Moreover, the diffractive effects induced by ionospheric irregularities on the GNSS signals passing through them will produce fluctuations of both the phase and amplitude of the signals, thus increasing the value of both the S 4 and σ ϕ indices [5], [57], contrary to what shown by Fig. 3a and  (c) Fig. 3: Maximum hourly values of the S 4 (Fig. 3a) and σ ϕ (Fig. 3b) indices during August 2021 (satellites elevation above 30°) and local K-index (Fig. 3c) recorded during the same period.Thresholds (dashed horizontal lines) of Fig. 3a and 3b are defined according to Table I.
butions of the indices with those recorded during the event 375 of the 10th of March 2022, it is possible to observe the 376 expected behavior in the case of a real ionospheric scintillation 377 event (images of Fig. 5 and Fig. 6b) and eventually conclude 378 that the anomalies were not induced by natural ionospheric 379 phenomena.Indeed, given the small scale (a few hundreds of 380 meters) of the irregularities leading to L-band scintillations, 381 and considering the latitudes under investigation, not all the 382 satellites in the FoV of the receiver are expected to be affected 383 by scintillations; as a consequence, the mean and maximum 384 values of the S 4 will exhibit different patterns, as shown by 385       with a duration of 1 s, and on the full capture of 10 minutes, according to where M is the amount of signal samples and N is the amount of evaluation point of the Fast Fourier Transform (FFT).The Welch PSD is hence given by averaging the periodogram as where K is the amount of frames over which the power spectrum is averaged and W identifies the Welch formulation [58].The analysis provided a preliminary feedback on possible spectral anomalies with respect to GNSS signals observed in nominal conditions.
2) Persistence Spectrum: was adopted to investigate the RFI spectral signature and the stability of an intelligible PSD over short time periods [60].This analysis is based on the accumulation of Welch spectrograms (9) on a grided PSD plot.The longer a particular PSD envelope persists in a signal as the signal evolves, the higher its time percentage and thus the brighter is the heatmap in the plot.The tool is also helpful to identify hidden coherent signals in noisy patterns as well as sporadic or fast pulsed signals with unknown duty cycles.
3) Time-Decimated Time-Frequency Analysis (TD-TFA): was performed through the estimation of partially-overlapping Short Time Fourier Transform (STFT).A signal chunk composed by N samples is filtered through a shaped window of length K, and a Discrete Fourier Transform (DFT) is computed over N DF T points.The window slides over the next N samples with an overlap of the previous L samples, and a DFT is performed for each window.By sliding the window along the samples vector, a Time-Frequency Analysis (TFA) provides a time-frequency view showing the evolution of the frequency content of a signal along the time [61].The technique was exploited to describe the evolution of the signal by measuring its PSD profile over the whole acquisition time-span.To reduce the size of the output data, a time decimation (TD) was performed by skipping a predefined timespan in between subsequent signal chunks, with an acceptable reduction of the time resolution.Shorter signal time spans are preferable in terms of time consumption since they allow faster STFT computation, by dealing with smaller amounts of samples.In terms of readability of the TD-TFA output figures, the following options provided similar results A. t i = 20 ms and t s = 100 ms → 285 MB B. t i = 1 s and t s = 1 s → 28.5 MB where t i is the integration interval and corresponds to the overall duration of the signal samples processed through STFT, and t s is the skip interval included between two subsequent integration intervals.While the first corresponds to the actual amount of input data, the latter indicates the duration of unprocessed signal chunks, thus representing the decimation factor of the proposed TD-TFA.TFA analysis contains more information in configuration A, however, this appeared not relevant as it does not significantly impact the visual detection where B eq = 1/T c with T c stands for the coherent integration 527 time, and The C/N 0 is hence computed over a window of length M that where j refers to the j-th GNSS signal, W = 60 s indicates

541
C. Analysis of the GISTM scintillation data 1) Ground Based Scintillation Climatology (GBSC): It consists in building maps of the percentage occurrences of the scintillation indices above a predefined threshold and evaluated over a certain time period [2].The climatological maps report the percentage occurrences on a bi-dimensional time-grid having the hour of the day in the horizontal axis and the day of the year in the vertical one or as geographic maps, showing the percentage occurrences evaluated over geographic cells with a given spatial resolution.The technique is used to perform climatological analysis of scintillation events, but it can also be adopted to highlight the spatial and temporal features of scintillations over shorter time-periods (e.g.few months).With regards to the S 4 index, the S 4 percentage occurrences in a given time-interval (S 4P Ot ) is evaluated as: where S 4thr (∆t) is the total number of the S 4 occurrences above the chosen threshold in the given time-interval ∆t and S 4tot (∆t) is the overall number of S 4 measurements available in the same time-interval.The S 4 percentage occurrences over a specific geographic cell (S 4P Os ) is evaluated as: where S 4thr (∆t, ∆lat, ∆lon) is the total number of the S 4 occurrences above the chosen threshold in the given time-interval ∆t and limited to the specific geographic cell (range of latitudes ∆lat and longitudes ∆lon), while S 4tot (∆t, ∆lat, ∆lon) is the overall number of S 4 measurements available in the same time-interval and pertaining the same geographic cell.
2) RFI filtering: In order to remove the RFI-induced anomalies from the S 4 data, all the epochs in which the mean values of the S 4 (calculated on all the available signals at that epoch) are above a certain threshold have to be filtered out from the dataset; indeed, as follows from the considerations reported in Section III-B, the RFI has the effect of increasing the S 4 values of the majority of the satellites in view at the same epoch, differently from actual ionospheric scintillation events.In the case of Lampedusa, given that the average number of satellites simultaneously in the FoV above 10°578 of elevation is 30, and assuming that 20 percent of the signals could be at most simultaneously affected by actual ionospheric scintillations at these latitudes, a threshold of 0.15 for the mean values of the S 4 has been chosen as a good   2) Spectral persistency and RFI spectral signature: The set of plots in Fig. 9 shows examples of persistent spectrum analysis performed on 1 ms signal chunks every 10 s for an overall observation time of 60 s.As we can observe through the subplots, the spectral signature of the interferer considerably changes along the time.A nearly-symmetrical spectral signature is visible in Fig. 9d that may suggest a 2-Frequency Shift Keying (FSK) modulation.However such a signature slightly recurs only in Fig. 9b with a lower intensity, thus weakening the hypothesis.Similar asymmetrical signatures can be observed in Fig. 9a and 9e.A flattened spectral shape is instead visible in Fig. 9c and 9f where RFI intensity dramatically drops.Such a time varying behaviour makes the signal particularly difficult to be automatically identified, or tracked.Additionally, autocorrelation of time series along the observed datasets did not show any relevant similarity of the RFI bandwidth, being possibly detrimental for Galileo E1 signals.Fig. 10d shows a fragmentation of the RFI power spectral density with an unusual behaviour and mid to low intensity sporadic peaks were observed in the second half of the dataset.Fig. 10e shows an increasing RFI intensity with time that reaches its maximum (−5 to 0 dB-Hz) by the end of the dataset.The dataset presents a unique example of regular intensity growth.Fig. 10f shows a sharp drop in the received RFI power density at about 05:45:30.The phenomenon suggests a sudden interruption of the RFI transmission.In the first quarter of the plot the PSD shows moderate to strong intensity in the range −10 to −5 dB-Hz.Additional Continuous wave (CW) interferences were sporadically observed, such as in Figs.10d, 10e, and 10f with a non-negligible intensity at ±0.5 MHz and ±1.5 MHz.However, their presence cannot be directly related to the RFI target in this study.It is worth remarking that power variations highlighted by TFA appear slower than the changes observed in the spectral signature, thus we cannot assume they are related.show a more accurate match among such abrupt variations of the estimated C/N 0 and the anomalous increments of the corresponding amplitude scintillation index S 4 computed by the GISTM receiver.Noisy data series are obtained through (13) and they are plotted along with their 95 % confidence interval (shaded grey areas).The plots presented in Fig. 11, show the variation of the C/N o, namely δC/N 0 , with respect to to its mean estimated over non-overlapping windows of 60 s for the selected datasets.By comparing the results with the TFA analysis of Fig. 10, it can be seen that in correspondence of intense RFI occurrences, rapid fluctuations of the C/N o are present, thus they have not been properly compensated in the computation of S 4,n through (7) .Despite this effect is more evident for GPS L1/CA records, intense RFI occurrences also lead to remarkable fluctuations in Galileo E1c data 3 .More in detail: Fig. 11a shows the strongest fluctuations both in GPS and Galileo E1c signals.Peaks overcome a range of ±5 dB up to severe drops of −8 dB for GPS L1/CA and confidence interval appears larger in correspondence of the

B. RFI Numerical Emulation
Relying on the TD-TFA it can be inferred that no patterns can be recognized both in the temporal evolution of the signal and in its spectral content.Furthermore, RFI received power shows slow variations and a generous intensity range.TD-TFA was fundamental to observe that the RFIs occurrences may show a sharp starting and ending time that can be easily attributed to artificial, deliberate transmissions.Relying on these observations, the most relevant information that justify the modeling we propose hereafter comes from the persistence spectral analysis and from background literature on communication systems and GNSS threats and mitigation.A basic model for a Multiple FSK (MFSK)/Frequency-Hopped (FH) signal was implemented to be compared with the identified RFI and foster the design of new countermeasures to mitigate its action.Despite of being a conventional modulation scheme for communication channels, MFSK has been employed in radar applications for its capacity of measuring and resolving targets in range and Doppler frequency simultaneously and unambiguously even in multitarget situations [62].A MATLAB script was exploited to numerically evaluate the expression where f m (nT s ) is a function that randomizes the generation of a set of m sub-tones included in a predefined frequency range,  (e.g., adaptive notch filters), which may be unable to track the 799 jamming signal.The designed MFSK signal shows frequent 800 and remarkable changes in its spectral signature as shown in 801 Fig. 13, where the numerical RFI shows a similar behaviour 802 to the one observed in persistence spectra analysis of Fig. 9, 803 in Section V.

804
C. Impact of the RFI on scintillation data and filtering algo-805 rithm 806 1) Effects of the RFI on Low-latitudes ionospheric scin-807 tillations investigation: As mentioned in Section I and III, 808 mid-latitudes scintillation may occur as a consequence of 809 disturbed space weather conditions; on the contrary, low-810 latitude scintillations are also possible during quiet time, 811 especially for the geomagnetic latitudes close to the northern 812 and southern EIA crests, due to the formation of small scale 813 irregularities embedded in the EPBs.Considering the position 814 of the Lampedusa observatory, an investigation addressed to 815 the observation of low-latitude scintillations would require to 816 also include the signals coming from low-elevation satellites 817 with respect to the receiver FoV; this will introduce additional 818 outliers in the data due to the effects of the multipath, as 819 Being not possible to exclude the low-elevation satellites (due to the necessity of observing low-latitudes), a possible way to remove the outliers produced by the multipath is by increasing the threshold of the S 4 occurrences above the level of severe scintillation (S 4 > 0.7); this operation has also the beneficial effect of removing the less intense S 4 anomalies caused by the RFI, but will prevent the capability to detect possible real ionospheric scintillations events of moderate intensity.The result of this operation is shown in the images of Fig. 15: the background feature due to the multipath visible in Fig. 14a are removed (see Fig. 15a) and the overall spatial and temporal extent of the anomalies induced by the RFI is minimized as expected (see Fig. 15a and Fig. 15b in comparison to Fig. 14a and Fig. 14b).
2) RFI filtering and detection of ionospheric scintillation events: The S 4 percentage occurrences reported in Fig. 15 are due to both RFI-affected observations and possibly actual ionospheric scintillation events.To finally detect and remove the remaining S 4 anomalies due to the severe effect induced by the RFI, it is possible to reprocess the original data according to the methodology reported in Section IV-C2.The result of this filtering operation is shown by the images of Fig. 16.By detecting and removing the occurrences attriubuted to the RFI, the timeline of the S 4P Ot reported in Fig. 16a allows to detect, without ambiguities, severe scintillation events (highlighted by the white dotted box) occurred in the post-sunset hours during the period of the autumn equinox 2021.Similarly, the map of Fig. 16b reports the S 4P Os , showing the actual geographic area affected by scintillations (highlighted by the white dotted box) which cover the lowest latitudes in the FoV.The scintillation events highlighted in Fig. 16 reflect the typical features of ionospheric scintillations induced on GNSS signals by small scale irregularities embedded in EPBs reaching the north crest of the EIA.Even though an accurate

VI. DISCUSSION
No natural events or human, licit or illicit activities being known to the authors seem related to the anomalous occurrences and the features of the disturbance.Additionally, no other instruments were expected operating in GNSS L1-band at the ENEA station or can interfere by emitting spurious harmonics in such a frequency range.The RFI may be generated in the proximity of the GISTM station (jamming or selfjamming) through a fixed or moving transmitter but the slow, yet remarkable power variations may indicate variable distance or heading of the transmitting antenna.This feature may be attributed to a moving transmitter carried on board of a plane, ground vehicle, or ship (mobile transmitter with fixed/moving antenna).Independently on the dynamics of the emitter, the RFI transmitting antenna may change its orientation along the time (e.g., fixed emitter with a spinning antenna as per radar applications).However, nor the regularity of the power fluctuation nor evident duty cycles in the received power suggest the possibility of a regularly spinning antenna.In light of this, the hypothesis of a moving emitter appears more reasonable.We cannot exclude the presence of jamming activities in the area of interest, as well as the possibility of experimental tests for MFSK radar systems or undocumented applications such as steganography in GNSS band for stealth data transmission.In fact, the characterization of the RFI detected in Lampedusa reflects the features of a deliberate MFSK transmission that may occasionally turn into a jamming interference on the L1/E1 frequency band in case of intense received signals.It mainly affects and severely degrades GPS L1/CA and Galileo E1c signals, but it seems poorly effective as a jammer against Galileo E1b, GLONASS and Beidou signals; in light of this, the gathered clues suggests the observed RFI may constitute a rough attempt of RF steganography covered by GNSS signals or a modern FH jammer.As a general remark, similar transmissions over GNSS L1/E1 center frequency are generally forbidden.However, while the United States (U.S.) prohibits unauthorized transmission on the GNSS frequency bands by federal laws [64], European regulations are more fragmented and may differ among member and nonmember states.Specifically, the Italian legislation, with articles 340, 617, and 617 bis of the Penal Code, punishes the use and installation of jamming devices.In Italy, the deliberate use of interferers is allowed only to law enforcement and military forces, but the limitations at the continental border between Europe and Africa, such as in the area of Lampedusa, may not be exhaustively disciplined by regulations.Nonetheless, their occurrences are growing worldwide and at the European borders they might be due to the intensification of war actions and the presence of military enforcement.Therefore, an increasing attention is nowadays placed on their effects on several civil GNSS-related activities, such as flight operations, maritime navigation, critical infrastructures.A remarkable effort is indeed being placed towards RFI monitoring and localization by means of LEO satellites [44], [65].From a terrestrial perspective, the deployment of multiple synchronous stations would allow as well for TDOA/FDOA-based interfer localization [66]- [68].At the time of writing, RFI localization falls outside the scope of this article.Despite the interferer detected in Lampedusa is, at the moment, of unknown origin, its appearances during summer periods and the geopolitical conditions of the area make it possibly related to the migratory flows phenomena involving the surrounding seas, from the African coast to the east Mediterranean.
With regards to the scientific activities, recent discussions in the ionospheric community have raised the attention about the possible disruptive effects of RFIs on the data collected for scientific investigations of the ionosphere as well as for space weather monitoring applications.This paper provided an onfield proof of such vulnerabilities, showing the adverse impact of RFIs for both near-real time GNSS scintillation events detection as well as in case of climatological investigations of ionospheric scintillations.In the case of Lampedusa, the intensity and repetition over time of the S 4 anomalies allowed to promptly acknowledge the presence of a possible source of interference; however, similar but less impacting RFIs may not be easily recognizable and yet affecting the quality of the collected data.At the same time, deploying capturing systems 994 to detect and characterize RFIs, like the one presented in 995 this study, is not a sustainable solution for both economical

Fig. 1 :
Fig.1: INGV ionospheric scintillation monitoring network in the European area (Fig.1a) and detail of Lampedusa island(Italy)  showing the position of the ENEA observatory and other areas of interest (Fig.1b).

Fig. 2 :
Fig. 2: Block diagram of a conventional, single-channel tracking loop architecture for GNSS receivers.I p and Q p outputs from the prompt correlator (P) are employed in the estimation of amplitude scintillation indices, i.e. S 4 , while σ ϕ is estimated through the output of the loop filter in charge of tracking the IF or the residual carrier frequency.

Fig. 3b .
Fig. 3b.Further considerations on the observed temporal and spatial distribution of the scintillation indices, when compared to the case of a real ionospheric scintillation event, allow to eventually exclude ionospheric phenomena as the source of the observed anomalies.The following analysis focuses on the data of the 7th August 2021, when several anomalies were recorded, compared to the data of the 10th March 2022, when a real ionospheric scintillation event was detected over the area under investigation.With regards to the data of the 7th of August 2021, Fig. 4a reports a daily view of the time profiles of the S 4 index, where different colors are attributed to the different satellites in view (Space Vehicle ID are reported in the legend).As Fig. 4a shows, the occurrences above the threshold of moderate scintillation seems to affect the signals from the majority of the satellites in view during the day; on the contrary, the time profile of the σ ϕ does not exhibit similar patterns, as shown by Fig. 4b.Fig. 4c reports a daily view of the maximum (blue line) and mean (green line) values of the S 4 index calculated on all the signals in view.As Fig. 4c suggests, most of the satellites in the FoV exhibit similar patterns; as a consequence, the S 4 mean and maximum values appears to be very close each other.Fig. 4d shows a daily view of the time profiles of the maximum S 4 values calculated among all the signals pertaining the same satellites constellation.From Fig. 4d, it is possible to spot similar patterns among the GPS (blue line), Galileo (red line) and BDS (yellow line) satellites, while GLONASS satellites (purple line) seems to be not affected by scintillations most of the time.Finally, Fig. 6a reports on a geographic map the S 4 occurrences above the threshold of moderate scintillation (S 4 > 0.25) during the same day (7th August).The points on the map represent the Ionospheric Pierce Points (IPP)s at 350 km for all the satellites in view and their color represents the values of the S 4 .As Fig. 6a shows, moderate to severe scintillations are visible across the entire FoV of the receiver, while ionospheric scintillations in quiet geomagnetic conditions are more likely to occur in the proximity of the EIA crests, respectively at ca. ±20°from the magnetic equator.Similar features of the spatial and temporal distributions of the scintillation indices reported for the 7th of August were eventually observed in each day of August 2021 affected by the anomalies.When comparing the previous temporal and spatial distri-

Fig. 5c (
contrary to Fig. 4c, when the RFI was present), and 386 (a) Time profile of the S4.Different colors are attributed to the different satellites in view (Space Vehicle ID in the legend).(b) Time profile of the σ ϕ .Different colors are attributed to the different satellites in view (Space Vehicle ID in the legend).Time profile of the S4 by considering maximum and mean values among all the available satellites.Time profile of the S4 by considering the maximum values among all the satellites pertaining the same GNSS constellation.
(a) Time profile of the S4.Different colors are attributed to the different satellites in view (Space Vehicle ID in the legend).(b) Time profile of the σ ϕ .Different colors are attributed to the different satellites in view (Space Vehicle ID in the legend).Time profile of the S4 by considering maximum and mean values among all the available satellites.Time profile of the S4 by considering the maximum values among all the satellites pertaining the same GNSS constellation.

Fig. 5 :
Fig. 5: (10th of March 2022) Scintillation indices in case of real ionospheric scintillation event.Thresholds (dashed horizontal lines) are defined according to Table I

Fig. 6 :Fig. 7 :
Fig.6: Map of the S 4 occurrences above the threshold of moderate scintillation (S 4 > 0.25) for the 7th of August 2021 (Fig.6a) and for the 10th of March 2022 (Fig.6b) and for satellites elevation above 30°.Geographic coordinates are labeled at the border of the maps and represented by the dotted lines inside the map; geomagnetic latitudes are labeled inside the maps and represented with the continuous lines.
394IV.METHODOLOGY395 A. Experimental Setup and data collection 396 In September 2021, new investigations were carried-out to 397 assess the nature of the anomalies presented in Section III.398 In order to acquire possibly-interfered GNSS signals, a dedicated experimental setup was deployed alongside the GISTM receiver, based on a SDR architecture.A high-level block scheme of the setup is provided in Fig. 7a while a picture of the operational hardware deployment is shown in Fig. 7b.General-purpose SDR front-ends are typically employed for research and development activities in radio-communication systems as they facilitate the acquisition of RF signals through configurable and flexible hardware and software architectures.By exploiting such flexibility, the setup aims at collecting IF signals samples of the received GNSS L1-band (center frequency 1575.42MHz) to perform investigations on possible intentional or unintentional interferences affecting the GNSS signals (and the derived scientific data) recorded on the island.

508 4 )
GNSS signal tracking: it was performed on the acquired 509 datasets to quantify the impact of the RFI on GNSS receivers 510 tracking stage, thus assessing the induced jamming effect 511 on navigation signals in terms of C/N 0 .The signal track-512 ing leverages the cross-correlation of Direct-Sequence Spread 513 Spectrum (DSSS) Code Division Multiple Access (CDMA) 514 signals transmitted by GPS and Galileo satellites.The software 515 receiver architecture imitates the conventional channel tracking 516 already described in Fig. 2. For the scope of these analysis, 517 the tracking was performed on the acquired GNSS signals 518 with a coherent integration time T c = 0.020 s.A key metric 519 for the conditioning of S 4 is the C/N 0 measured at each 520 channel.According to the analysis presented in Section III, 521 common effects are expected to be concurrently observed on 522 different satellites signals.Therefore, we propose an aggregate 523 estimation of the variation of C/N o, namely δC/N o, with 524 respect to the mean value used in (7).Formally, an estimate 525 of the C/N o is given by 526 C/N 0 = 10 log 10 (SNRB eq )

529 is typically set to 1 /
T c .To be consistent with the definition of 530 the indices provided in Section II-B, its aggregated variation 531 for all the tracked signals has to be measured by averaging 532 the 60 s de-trended series of the respective C/N 0(11)

534
the observation window, and S refers to the overall number 535 of available signals.536 5) RFI signal emulation and model: provided the features 537 observed through the above-mentioned analysis tools and 538 the recent literature on GNSS interferences and threats, a 539 signal with similar features was numerically simulated and 540 reproduced by means of a MATLAB routine.
Fig. 8: Single and multiple datasets data probing performed on 1 s signal chunks by means of a GNSS signal analysis tool embedded in the GNSS software receiver.

4 )
C/N 0 estimation in GNSS receiver open-loop tracking stage: According to the theoretical definitions of corrected amplitude ionospheric indices provided in Section II-B, the impact of rapid C/N 0 fluctuations induced by the RFI may cause misleading output values at GISTM.The following results

Fig. 10 :
Fig. 10: TD-TFA of the datasets in Table III showing different RFI behaviours in terms of PSD time evolution, compared to maximum and mean S 4 time series (top panels).Filled and blank markers indicate mean and maximum S 4 values, respectively (top panels).Spectrograms and S 4 data series are obtained from independent devices.

Fig. 11 :
Fig. 11: Mean variation of the estimated C/N 0 (13) for GPS L1/CA, Galileo E1b and E1c during the observation timespans of the selected datasets (limited to 9 minutes).Filled and blank markers indicate mean and maximum S 4 values, respectively (magnitude on the right y-axis).Background, grey-shaded areas show the 95 % confidence interval (left y-axis).

TFig. 12 :
Fig.12: PSDs of a simulated MFSK transmission observed over different snapshots duration and acting as an FH jamming interference.The spectral signature shows remarkable similarities with respect to the RFI's counterpart in Fig.8and Fig.9.Lower noise floor is considered with respect to the collected data.

Fig. 13 :S 4
Fig. 13: Example of the evolution of the signal PSD of the emulated RFI.Different frequency resolutions are achieved by spectral estimation performed on different durations of the signal chunk under analysis, i.e., 10 ms (light-grey lines) and 10 µs (black lines).

Fig. 14 :Fig. 15 :
Fig.14: S 4P Ot (Fig.14a) and S 4P Os (Fig.14b) above the threshold of moderate scintillation (S 4 > 0.25) between July and October 2021.The white lines of Fig.14arepresents the solar terminator at 350 km.In Fig.14bgeographic coordinates are labeled at the border of the maps and represented by the dotted lines inside the map; geomagnetic latitudes are labeled inside the maps and represented with the continuous lines.

Fig. 16 :Fig. 17 :
Fig.16: S 4P Ot (Fig.16a) and S 4P Os (Fig.16b) above the threshold of severe scintillation (S 4 > 0.7) between July and October 2021 after applying the filter for the RFI removal.The white lines of Fig.16arepresents the solar terminator at 350 km.In Fig.16bgeographic coordinates are labeled at the border of the maps and represented by the dotted lines inside the map; geomagnetic latitudes are labeled inside the maps and represented with the continuous lines.The white dotted boxes highlights ionospheric scintillation events due to EPBs.

996
and technical aspects.At the time of writing, no real-time 997 mitigation techniques for such elaborate interferers are known 998 to the authors, and only a-posteriori processing may allow 999 to detect interfered observations and provide quality metrics 1000 for the collected data.In this regard, this work proposed a 1001 preliminary post-processing methodology to detect and remove 1002 the RFI-induced anomalies from the scintillation data acquired 1003 by the GISTM receiver.The filter is not based on the specific 1004 characteristics of the RFI under investigation and, in principle, 1005 it can be also effective for different types of RFIs acting 1006 within the GNSS bandwidths; however, it has the bottleneck 1007 of being based on a threshold which is defined through a-priori 1008 assumptions and which is location-dependent.The design of 1009 more robust post-processing algorithms falls outside the scope 1010 of this paper and deserve dedicated investigations.1011 Summarizing, the lack of an accurate RFI model constitutes 1012 the main concern for a systematic analysis of its impact on 1013 the scintillation index.Besides, it is worth pointing out that 1014 a methodology to evaluate the RFI impact on the scintillation 1015 index is also lacking in the literature, and it deserves dedicated 1016 investigations in future works.

1018
This paper presented an investigation of a real scenario1019 where an unclassified RFI affecting the GNSS signals jeopar-1020 dize scientific activities like those carried-out by the INGV in 1021 the Mediterranean area of Lampedusa.It was shown that the 1022 computation of the ionospheric scintillation indices through 1023 modern commercial GISTM receivers may be misleading in 1024 those circumstances, thus triggering false ionospheric scintil-1025 lation events and compromising the reliability of real-time 1026 monitoring applications as well as the quality of the data 1027 collected for scientific investigations.The analysis presented 1028 on the recorded GNSS signals specifically demonstrated that 1029 altered scintillation indices may be due to the non-stationarity 1030 of the estimated C/N 0 caused by the observed RFI.Further 1031 on-site campaigns are expected in the future by refining 1032 the experimental setup with a complete decoupling of the 1033 GISTM/SDR acquisition chain (e.g., antenna) and by imple-1034 menting a multi-frequency acquisition unit (including L2/L5 1035 GNSS bands).Moreover, by deploying multiple synchronous 1036 stations would allow to implement Time Difference of Arrival 1037 (TDOA)/Frequency Difference of Arrival (FDOA) interferer 1038 localization [66].
214 quantify ionospheric scintillations based upon received GNSS 215 signals features.S 4 measures the variability of the signal 216 intensity (SI), that is estimated as Eventually, w RX models the additive thermal noise introduced by the receiving chain and the quantization noise injected by the Analog-to-Digital Conversion (ADC) operated at the RF front-end.Within this study, GNSS signals are considered continuously available at the receiver while RFI terms may occasionally occur.The RF front-end downconverts the input signal to a pre-defined IF prior to its sampling and quantization at the ADC.As shown in Fig. 2 the baseband numerical 204 samples from In-Phase (I) and Quadrature (Q) branches are 205 correlated with early (E), prompt (L) and late (L) replicas of 206 the locally-generated spreading code.Eventually, the Integrate 207 & Dump block provides prompt In-Phase (I p ) and Quadrature 208 (Q p ) samples which are used to estimate the S 4 index, while 209 the σ ϕ index is derived through the output of the loop filter 210 in charge of tracking the IF carrier, as depicted by the bottom 211 branch of the diagram in Fig. 2. 212 B. Amplitude and phase scintillation indices 213 The S 4 and σ ϕ are the statistical indices typically adopted to and (4) are the I p and Q p 221 components of the received signal after the integrate and dump 222 operation performed by the receiver tracking stage and M 223 is the total number of accumulated periods.The S 4 index is 224 defined as the normalized standard deviation of the detrended 225 50 Hz raw signal intensity over a given interval of time, 226 typically 60 s 227 S ′ 4 = ⟨SI 2 ⟩ − ⟨SI⟩ 2 ⟨SI⟩ 2 (5)

TABLE I :
[49]entional thresholds for the classification of ionospheric scintillation events based upon amplitude and phase indices[49].

TABLE II :
Configuration parameters of the front-end and the acquisition software.
419 10 MHz reference signal to the ER USRP, and a Network-420 Attached Storage (NAS) for the storage of large data volume.421A2-way splitter is exploited to feed both the GISTM receiver 422 and the front-end with the RF signals received at the GNSS 423 PolaNt Choke Ring B3/E6 antenna.The acquisition routine, 424 continuously executed on the host PC, is being part of a 425 proprietary GNSS fully-software receiver designed to emulate 426 the processing chain of commercial receivers in a more flexible 427 and controllable environment.The configuration parameters 428 of the front-end and of the aforementioned acquisition routine 429 are reported in Table II.To partially overcome the well-known 430 issue of storing TBs of binary files produced by such systems, 431 the Lampedusa setup took advantage of a NAS unit which 432 directly stores the IF signal samples during the acquisition.433Moreover,a fully-automated procedure continuously acquires 434 24/7 the IF samples and daily freed the space on the NAS 435 from the non-useful datasets.436Thefirst collection campaign provides 171 datasets of 10 437 minutes each (28.5 hours), affected by the RFI with different 438 intensity and time behavior.The collected datasets is in-439 cluded in an open data collection, i.e., Lampedusa Scintillation 440 Monitoring Interfered Data (LAMP_SMID_2109) 1 , and an 441 overview of their time distribution over the test campaign is 442 shown in TableIII.443 B. Post-processing Signal Analysis (SDR data) 449 of its effect on the estimation of the S 4 .450 1) Spectral analysis through Power Spectral Density (PSD) 451 estimation: the analysis was performed through a PSD esti-452 mator, i.e., Welch spectrogram [58], [59], on signal snapshots 453 1 http://ieee-dataport.org/10996

TABLE III :
Amount of datasets collected during the September test campaign in Lampedusa, and available in the LAMP_SMID_2109 open data collection.Reference dataset not automatically retrieved by the system but still affected by low-intensity RFI. a

TABLE IV :
Datasets selected as representative samples of the observed anomalous GNSS signals for the presentation of the analysis results in Section V-A.
b Datasets not kept by the automated grabbing system.compromise to detect most of the RFI-induced anomalies, 583 avoiding at the same time to filtering-out possible actual 584 ionospheric scintillation events.It has to be noted, however, 585 that the proposed filtering technique potentially removes from 586 the dataset the actual ionospheric scintillation events occurring 587 contemporary the interferences.

TABLE V :
Simulation parameters for the emulation of a MFSK/FH jamming signal.