Corresponding author: O. Isoz, Division of Space Technology, Department of Computer Science, Electrical and Space Engineering, Luleå University of Technology, Box 812, SE 981-28 Kiruna, Sweden. (firstname.lastname@example.org)
 It is well known that terrestrial GPS/Global Navigation Satellite Systems (GNSS) receivers are vulnerable and have suffered from intentional and unintentional interference sources. Unfortunately, space-based GPS/GNSS receivers are not exempt from radio frequency interference originating from the Earth. This paper explores data recorded by the GNSS Receiver for Atmospheric Sounding (GRAS) instrument onboard MetOp-A in September 2007, which is assumed to be representative of the typical environment for GPS/GNSS instrumentation in LEO orbit. Within these data it is possible to detect both pulsed interference and variations in the background noise. One plausible source of the pulsed interference is identified. We also show that neither the pulsed interference nor the variations in the background noise degrades the performance of the higher level products from GRAS.
 It is well known that ground-based Global Navigation Satellite Systems (GNSS) receivers are vulnerable to both intentional and unintentional interference [Navigation Team AJP-652, 2011; Carroll, 2001]. Severe ground interference has also been observed in data from the GNSS radio occulation (RO) receiver onboard the MetOp-A satellite, [Bonnedal et al., 2010]. It is also known that other space-based instruments operating in the L-band such as the ESA's Soil Moisture and Ocean Salinity (SMOS) mission have had issues with ground-based interference [Skou et al., 2008, 2010]. The SMOS mission, which operates in the frequency range of 1403–1423 MHz, initially experienced a significant amount of interference from ground-based sources [Skou et al., 2010], but after a collaboration between European Space Agency and local authorities, the interference was significantly reduced [European Space Agency, 2010]. SMOS is particularly sensitive to ground-based interference since it measures thermal radiation from the Earth. This shows that in some cases, ground-based transmitters can be a problem for space-based instruments even if the frequencies used by the particular instrument are supposed to be interference free.
 Another problem for high accuracy space-based receivers is that the L2 frequency band is shared between GNSS and radio location systems.
 These radio location systems are often used to detect targets at long range and is therefore using a high power transmitters (e.g., the U.S. ARSR-4 aviation radar has a peak power of 65 kW and can detect target at a range of 460 km [Istok et al., 2009]).
 This paper uses data from the GRAS (GNSS Receiver for Atmospheric Sounding) radio occultation receiver on the polar orbiting MetOp satellite series [Bonnedal et al., 2010]. We have used two methods to characterize the interference received by GRAS. Both methods focus on the received raw signal instead of the derived high level data products in order to extract as much information about the source as possible. The first method is based on the receiver's measurement of the background noise. Since the received GNSS signals usually are below the thermal noise floor for an Earth-based receiver with hemispherical antenna, most receivers adjust the gain/attenuation in the analog parts of the receiver to the variations of background noise [Kaplan and Hegarty, 2006; Bastide et al., 2003]. In Borowski et al. , it was proposed to use this kind of measurement for the detection of false GNSS transmitters. This type of measurement can be used to localize where the receiver detects signals significantly stronger then the average background noise and further assesses any impact on the radio occultation data products. The main drawback with this method is that it only measures the received signal power, and therefore, it is not possible to determine if the signal is narrow or wide band.
 The other method is focused on determining the type of interference from the sampled 1 kHz open loop data. It has previously been shown that these data sometimes contain pulsed interference [Bonnedal, 2010].
 However, the amount of interference or the impact on the final products has not been assessed in detail.
 This paper is based on GRAS data from 1 to 19 October 2007 and is assumed to be representative for the nominal environment. The data was retrieved from European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT)/Radio Occultation Meteorology Satellite Application Facility (ROM SAF) during the summer of 2012. The data was processed from level L0r to L1a and L1b using the GRAS processing software prototype, YAROS version 0.8.4.dev. Preprocessed (O-B)/B values for the same time period were received from the Radio Occultation Meteorology Satellite Application Facility (ROM SAF).
2 GRAS Radio Occulting Receiver
 GRAS is a 12 channel L1/L2 GPS receiver designed to work both for orbit determination and tracking of occulting GPS satellites. The receiver's 12 channels are divided into three receiver chains. One receiver chain is used for navigation and has eight channels, this receiver chain is connected to a two element antenna mounted on the backside of the satellite that faces away from Earth. The two other receiver chains have two channels each and they are used to track occulting GPS satellites. One of the two chains used for receiving occulting satellites is connected to an 18 element antenna with a fixed radiation pattern that looks forward in the flight direction of the satellite whereas the other occulting receiver chain is connected to a second 18 element antenna that is directed aft, in the antiflight direction.
 Both antennas for occulting satellites have beams that are focused toward the horizon of the Earth with an azimuthal beam width of ±55°. The antenna for navigation is a two-element antenna with an antenna pattern that is directed away from the Earth with a gain of 7 dBi in zenith and with less then −10 dBi gain toward the Earth's horizon. Radio signals from Earth are also partly obscured by the satellite body. The receiver is capable of both L1/L2 semicodeless and L1 C/A tracking. The decorrelated GPS signal is sampled at 50 Hz when the receiver tracks satellites at high altitude using closed loop tracking. When the tangent altitude of the straight line path between the occulting GPS satellite and GRAS decreases, the signal will become weaker, and eventually, the closed loop tracking will fail. When the L2 tracking fails in setting or before the L2 signal acquisition starts in rising, one channel is used for open loop L1 tracking at 1 kHz, often in parallel with the closed loop tracking [Bonnedal et al., 2010].
3 Background Noise Measurements
 The background noise measurement is essentially the integrated signal power received by the antenna from all directions at all frequencies that passes through the receiver's bandpass filter over some time interval. In the GRAS receiver this measurement is taken over a 20 MHz bandwidth. Due to the antenna pattern, the signal will have a direction-dependent gain/attenuation. The occulting antenna has a maximum gain of 11 dBi; in nadir, the gain is about −25 dBi; and 90° azimuth, it is about 0 dBi.
 A map with approximate interference locations was presented in Bonnedal et al. , where GRAS had experienced an increase of the noise power density of at least 2 dB. The map indicated that there are fairly large areas in the Northern Hemisphere where the background noise power density is increased.
 GRAS continuously saves information about the state of the receiver in a number of observables [Saab Ericsson Space, 2004]. The rate at which these are saved varies between 1 Hz for some of the gain stages to 1 kHz for the variables associated with the raw sampling mode. Figure 1 shows a map similar to the one presented in Bonnedal et al. , where data from the same month was analyzed.
 The main difference is that each dot shows the mean value of all measurements during 1 s from all receiver channels connected to the specific antenna. The exact number of values averaged for each dot varies but is approximately 3 samples per active navigation channel and 50 samples per active occulting channel (gives usually about 20–50 samples per dot), whereas in Bonnedal et al. , one dot was made for each occultation with at least 2 dB increased noise. Each value has been been plotted at a location that is 1700 km from the subsatellite point in the direction of the antenna.
 This point should represent the center of the wide occultation antenna beam. The 1700 km shift was determined manually by comparing the ground patterns from north and south bound tracks given by data from the forward and aft RO antennas.
 Figure 1 shows how the estimated noise power density received by the L2 raising antenna varies over the world. Only L2 has been plotted here but similar plots can be made for the data from L1+L2 and L1 as well as for the other antennas. Figure 1 indicates that the background noise increases by up to 6–8 dB in some populated areas. For the navigation data, shown in Figure 2 where the data values has been plotted with zero offset from the satellite position, there is only a slight increase in the noise of about 2 dB between the Kamchatka peninsula and eastern Russia.
 Figure 3 shows the standard deviation of the noise power density of the L2 rising channel.
 Both the mean and standard deviation are based on 1 s of data for all channels connected to the specific antenna.
 It is significantly higher for the measurements located in eastern Asia compared to the measurements located in the Southern Hemisphere. The increase in the standard deviation in the navigation channel (Figure 4) is fairly consistent with where the standard deviation of the rising channels were highest (Figure 3). It seems that whatever causes the increased background noise in eastern Asia impacts both the RO and navigational channels of the GRAS receiver. Since the mean of the noise power density for the navigational channels only increases by about 1 dB but the standard deviation of the noise power density increases by as much as 20 dB, it seems that this interference source has low average power but strong pulses, such as a long range radar. Since GRAS uses a common analog gain stage for both L1 and L2, it is not possible to determine if the transmitter is L1 and/or L2, although L2 is the most likely since it is shared between radio-based navigation and radio-based localization [Electronic Communications Committee, 2013].
4 Pulsed Interference
 When the occulting satellite becomes too weak or too dynamic for the closed loop tracking to work, GRAS will track the L1 signal with an open loop architecture with a 1 kHz sampling of the resulting I/Q data. If the sampled I/Q data is plotted in a spectrogram, it is sometimes possible to see many different features in the received signal.
 These features can be ground reflections of the tracked GPS satellite, other GPS satellites (caused by imperfect orthogonality of in the GPS C/A code), and/or pulsed transmitters [Bonnedal, 2010]. After some initial manual analysis of the spectrograms, a script was written that could automatically determine if the I/Q data contained pulsed interference. It was determined that the data only contained two types of pulsed interference, one with a low pulse repetition frequency (PRF) of about 30 Hz and one with a higher PRF of about 210 Hz. All available occultations were analyzed and it was found that 561 of the 12479 or about 4.5% of the occultations contained pulsed interference.
 By plotting the positions of the occultations with pulsed interference and further separating them into rising/setting, it can be seen that the occultations with the higher PRF were recorded when MetOp-A was over eastern China (Figure 5) whereas the occultations with the lower PRF were recorded over a larger area and slightly further east (Figure 6).
 Since the typical radars operating on the L-band are long range, it is likely that their antenna beams are focused at a fairly low elevation angle. Therefore, it is reasonable to assume that the source(s) of the transmission with low PRF seen in Figure 6 and the high PRF seen in Figure 5 will be outside the area where the interference was received. One possible source of one of the radar signatures is the long-range radar located at Shemya Island, Alaska, USA that has a beam directed toward North West [Chorman, 2008]. When the locations of the raised noise in Figure 4 is compared to the antenna pattern of the radar, it becomes even more likely that it is one of the sources for the pulsed interference.
 For many of the occultations with pulsed interference, the noise power density varied rapidly about 1–2 dB during the recording of the occultation. The same amount of variation could also be seen in some of the occultations where only the occulting satellite could be seen, although the variations were at a much lower rate in these cases.
5 Impact on the GRAS Radio Occulation Products
 One method often used to validate radio occultation data is to compare the retrieved profiles with global circulation data from weather centers such as the European Center for Medium-Range Weather Forecasts (ECMWF). These centers assimilate the data from a wide range of sensors.
 By retrieving the forecasts where the GRAS observations have not yet been assimilated and then interpolating and forward modeling them to generate predicted bending angles, it is possible to do a statistical quality measurement of the retrieved bending angles. These comparisons are called (Observation - Background)/Background or (O-B)/B values, where observation refers to the RO measurement and background is the profile predicted by the model. Given all the variations in the background noise and the pulsed interference, one might suspect that there could be areas where the accuracy of the measurements is decreased.
 ROM SAF employs two quality scores in order to identify and remove low-quality occultations. The two quality scores are called L2 quality score (L2) and statistical optimization quality score (SO). These were originally introduced by in Gorbunov et al. [2006, 2011] and run from zero to infinity with high values representing problematic data or data of low quality. A histogram of all occultations that passed the ROM SAF quality control can be seen in Figure 7.
 The thresholds employed here to distinguish between “good” and “bad” occultations are 40 for the L2 score and 30 for the SO score. These threshold values retain the majority of occultations in the main bulge of the distribution while removing the long tail of problematic cases.
 The black lines show the mean (O-B)/B and ±1 standard deviation of the (O-B)/B values for the profiles that passed the quality control. The sharp change in standard deviation at 25 km is a known artifact introduced in the processing by a change in smoothing parameters at this altitude.
 The 462 of the 546 or about 85% of occultations with pulsed interference passed the quality control. This can be compared to the ratio for all occultations in the data set used in this paper, where about 83% of the occultations passed the ROM SAF quality control. The histogram for the occultations with pulsed interference is shown in Figure 8.
 The black lines show the mean (O-B)/B and ±1 standard deviation of the (O-B)/B values for the profiles with pulsed interference that passed the quality control. The histogram is very similar to the histogram calculated for all occultations Figure 7. Although the number of occultations plotted here are only 462, based on these numbers, there seems to be no significant difference between the occultations that contain pulsed interference and those that do not contain any significant pulsed interference.
 Since the ratio of occultations that pass the QC is similar for both occultations that contain pulsed interference and for those that does not contain any significant pulsed interference, it indicates that pulsed interference does not degrade the performance of the GRAS instrument.
 Figure 9 shows a map with the SO quality score for those occultations that had a SO score over 30 and/or L2 quality score over 40. As can be seen, the bulk of occultations with a large SO and/or L2 quality score is fairly evenly spread around the world mainly between 50°N and 50°S.
 The tracked signal becomes weaker and harder to track due to an increase in the variations of the refractivity index of the atmosphere at lower altitude. These variations in refractivity can cause both atmospheric multipath and attenuation of the signal. It is also sometimes possible to receive GPS signals that have been reflected off the ground.
 Given the existence of raised noise floor and pulsed interference, one might assume that there are areas where the mean L2 quality score is consistently higher relative to other areas. In Figure 10, the gridded mean of the L2 quality score for all occultations has been plotted. The mean is based on all occultations that occurred in an area with the size of 10° in east/west and 10° in north/south. For clarity, the values above 2000 is shown as being 2000.
 Figure 10 has similar pattern as the previous Figure 9, where the mean L2 quality score is slightly higher close to the equator compared to L2 quality score for occultations further north/south. Therefore, based on the data presented here, it does not seem that ground-based interference causes any degradation in either L2 performance or how far down in the atmosphere GRAS can track the GPS satellites.
6 Impact on the GRAS Navigation Receiver
 Even if the pulsed interference is easily detected by the increased noise in the navigational receiver chain (Figures 2 and 4), no impact on the C/N0 can be detected when the satellite flies over those areas.
 We assume that this is primarily due to the architecture of the GRAS signal tracking, its ability to recover quickly after pulsed interference has saturated the first stage amplifier and the fact that the retrieval signal processing is very frequency selective. Also, the GPS signal is known to be robust against pulsed interference.
 But it raises an important issue, powerful transmitters might have the potential to degrade the performance of orbit determination receivers using the L2 frequency on satellites in low Earth orbit.
 Although as a result of the high velocity of the satellite and the curvature of the Earth, the impact would be limited to certain parts of the satellite orbit.
 In this case the combination of the short geographic proximity, the pulsed nature of the interference, and the GRAS processing did not result in degraded performance in the navigation measurements.
 GRAS data from October 2007 have been analyzed to determine the amount and types of ground-based interference. Due to the wealth of low level information in the GRAS data products, it was possible to do a detailed analysis of the interference. Two types of pulsed interference were detected and the approximate location of one source was estimated. It was shown that there is a raised noise floor when the MetOp satellite is passing certain areas of the world and the GRAS occultation antenna is pointing toward these locations. Nevertheless, it was not possible to see any spatial variations in the degradation of the higher level GRAS products that correlated with the location of either raised noise floor or ground-based transmitters. Therefore, we conclude that, based on the analyzed data, the ground-based interference observed in this data does not seem to degrade the performance of GRAS.
 Part of this work was conducted as a Visiting Scientist activity of the Radio Occultation Meteorology Satellite Applications Facility (ROM SAF), which is a decentralized operational Radio Occultation processing center under EUMETSAT.
 The authors gratefully acknowledge EUMETSAT for data and support.