Alexandru Pandele, Antonia Croitoru, Institute of Space Science, Romania; Andrei Hulea, Romanian InSpace Engineering, Romania; Costi Cherciu, Institute of Space Science, Romania; Alina Radutu, Irina Stefanescu, Romanian Space Agency, Romania; Katarzyna Urbanska, ESA; Dumitru Andrescu, Romanian Maritime Hydrographic Directorate, Romania; Claudiu Dragasanu, Marius Trusculescu, Mugurel Balan, Institute of Space Science, Romania

View Abstract Sign in for premium content

Abstract:

Introduction The “Maritime GNSS navigation local error sources characterization” (MARGOT) is a project developed under ESA NAVISP Element I, in response to the call for “Multipath & Interference Error Mitigation Techniques for Future Maritime e-NAV Services.” The aim of this activity is to assess the levels of multipath and interference impacting Position, Navigation and Timing (PNT) information in maritime environments for different operational scenarios. The main objectives of the project are the determination of the over-bounding multipath and interference error models for the maritime environment, the determination of applicable mitigation methodologies for multipath and interference impact minimization and the determination of L-band multipath channel models for different navigation environments. The following sections present the results of the activity with regard to the multipath issues. From this point of view, the maritime environment is more challenging than aviation, as multipath levels may be impacted by the type of maritime operations (port navigation, coastal navigation, fluvial navigation, open sea navigation), the type of vessel, the sea state, the antenna location and its height above the sea, etc. Aiming to support the expansion of GNSS augmentation services to this environment, MARGOT assesses the possibility to develop and integrate one or several over-bounding multipath models, along with recommendations regarding the GNSS equipment set-up. Methodology Multiple data collection campaigns were performed on board of three vessels, with different dimensions and structural configurations. The campaigns started on the smaller vessel, which navigated on inland waterways, and then the equipment was moved on the larger vessels, which navigated mainly near the Romanian border of the Black Sea, with one larger campaign from Romania to the Crete Island in Greece, crossing the Bosphorus and Dardanelles straits between the Black Sea and the Mediterranean. Starting in May 2018, the data collection activity includes more than 100 days of maritime navigation and hundreds of hours of measurements collected at the docking position. The equipment configuration was composed of three GNSS receivers (2x Septentrio AsteRx-U and 1x AsteRx-m2), recording all available GNSS constellations, including L1/E1 and L5/E5 bands for GPS and Galileo systems. Three antennas are being used: two Septentrio PolaNT-X MF maritime antennas and one Septentrio Choke Ring antenna. Each of the two identical receivers are connected to two GNSS antennas. A Septentrio OEM receiver board (AsteRx-m2) is connected to one PolaNT-X MF. This was chosen as it was a good trade-off between a representative maritime GNSS receiver (the OEM board could be integrated in such a receiver) and the access to a wide range of measurements that is necessary in order to properly assess multipath and interference impact. This choice of the setup allowed the collection of multipath data on two types of antennas (a regular antenna with performances similar to existing Marine GNSS antennas and a low-multipath antenna (choke-ring), not regularly in use in the marine domain, but rather in geodetic surveying). Additionally dedicated equipment was used in the collection of ship pitch and yaw data for determining the effect of these phenomena on multipath and the collection of data from a GNSS receiver similar in performance with the ones in use in the Maritime domain, but allowing logging of necessary data. The multipath error characterization methodology started with the determination of multipath error terms using the pseudo range values estimated from code and carrier measurements, for GPS L1-L2 and L1-L5 and Galileo E1-E5a, E1-E5b and E1-E5. Besides the “raw” pseudo range measurements, 100 seconds smoothing was also applied. Having determined the multipath error for each satellite and each epoch, the resulted values were grouped in elevation “bins” of 1 degree. For each bin, the average value, the Root Mean Square (RMS) value and the standard deviation were calculated. The use of the standard deviation as a measure of multipath characterization, was confirmed by the Gaussian bell shape of the multipath distribution. The gaussian shape of the multipath distribution was tested for the elevation degrees corresponding to each 5th bin (i.e. 5, 10, … 85, 90 degrees). In order to assess the impact of ship structure on the recorded multipath levels, a polar plot was created using the difference between the azimuth of the satellite and the heading of the vessel. An important step in the analysis was the data classification, considering the following influencing factors: navigation type, vessel type, receiver type, antenna type, antenna location and height, and sea state. Also, the available data was divided into modelling data (80%) and validation data (20%). Following the objectives of the project, over-bounding models for the multipath errors have been developed. As the multipath error samples were grouped in elevation bins, each of these data sets were over-bounded by a Gaussian distribution. The corresponding standard deviations were used to fit an exponential function that provides the multipath error model. The Gaussian over-bound of the expected distribution was produced by following the method described by Juan Blanch, Todd Walter and Per Enge in "A MATLAB Toolset to Determine Strict Gaussian Bounding Distributions of a Sample Distribution" (ION 2017). First, the data was placed in equally sized bins. Then, the distribution defined by the binned data was modified by forcing unimodality for the bins above the median. This was done by modifying the bins one by one from the right-hand side by making sure that each was equal or larger than the previous one. The process was stopped when the top half of the distribution was defined. Each side of the distribution was defined by forcing the symmetry around the mean. The width of the middle bin was chosen to make sure that the distribution was unimodal. Then, a uniform distribution is assumed within each bin. The gaussian distribution is an over-bound for any interval of the form [x,+?] where x is above the mean. For each elevation bin such an over-bounding distribution was determined. The corresponding standard deviation provided a data point that was used to fit a function of elevation. An exponential function is used to determine the over-bounding fit. The over-bounding model is determined by using the same function, while adapting the free term such that the largest over-bounded value is considered. The trust of the model is computed by measuring how far above the ideal fit are the over-bounding values. Results The results of the data analysis can be discussed in relation with the aforementioned data classification. In terms of navigation type, the collected data was analyzed considering coastal navigation, open sea navigation, inland waterways navigation, port operations and port approach navigation. The largest amount of data was collected during coastal navigation (more than 1000 hours) while the smallest data set corresponds to port approach navigation (less than 100 hours). The inland waterways environment is the one experiencing the smallest multipath overall, while port operations experience the highest. The analysis covers both GPS and Galileo. More details of how each type of navigation is affected by multipath shall be available in the final presentation. Regarding the vessel type, the differences between the three vessels (“Istros”, “Mare Nigrum” and “Catuneanu”) were in dimensions and in the configuration of the on-board structures located in the proximity of the GNSS antennas. For GPS, “Catuneanu” has better results than “Mare Nigrum”, except on the L2 frequency. “Istros”, the smallest ship, has multipath over-bounding models that are generally lower than those for the other two ships. Also, the normalized azimuth analysis revealed higher multipath values concentrated in the directions in which high metal masts or chimneys were placed on the vessel. In port, the frequency of the multipath spikes intensifies, due to the presence of port structures. The comparison between the two receiver types (Septentrio AsteRx-U and AsteRx-m2 OEM board) was made considering a short campaign with one of the ships navigating in coastal waters. It could be observed that the behaviour of the two receivers is similar for all frequencies, both smoothed and un-smoothed data. This was expected, as multipath affects all GNSS signals at the receiver antenna level. The impact of the receiver on the multipath error can be seen only when internal, usually proprietary techniques, are employed by it in order to reduce multipath. However, for the purpose of the project, these features were switched off for both receivers. This assessment confirms that the impact of the used receivers is minimal with respect to multipath. In terms of antenna type and location, two different types were used: one Choke Ring antenna and two PolaNt-X MF antennas. They were placed along the longitudinal axis of the ship on “Istros” and perpendicular to the longitudinal axis on “Mare Nigrum” and “Catuneanu”. Looking at the impact on multipath of the antenna placing, it was observed that two similar antennas, placed at 4 m one from another, experience very similar multipath, both for smoothed and un-smoothed pseudoranges. Small differences, up to 10 cm, can be seen for small ranges of elevations that vary with the frequency. The assessment is not influenced by other possible factors, such as antenna height (they were placed at the same height), type of navigation, ship type, sea state, receiver type (data was recorded simultaneously from the two antennas by the same receiver). Regarding the antenna type, it is clear that the choke ring antenna has a better performance than the PolaNT-X MF, with improvements of almost 50 cm in low elevations and 20 cm at high elevations. However, the performance benefits are reduced when using smoothing, especially for GPS. Moreover, the improved performance may also be due to the central placement of the choke ring antenna. One of the two PolaNT-X MF antennas was placed at different heights on board of the vessel “Mare Nigrum”. The antenna was initially placed at a height of 13.85 m above the sea level, then the height was changed to 14.45 m above the sea surface. By placing the antenna 0.6 m higher, the multipath errors are reduced with approximately 0.2 m for coastal navigation. The higher antenna gives slightly better results when smoothing is applied for both GPS and Galileo. On the other hand, in the port operations case, the height of the antenna seems to have almost no effect on the multipath error. This behaviour can be explained by the presence of the port structures, which are significantly higher than both antennas. The analysis of the over-bounding fits reveals that the type of navigation does not add significantly to the impact of the antenna height. The port and coastal navigation models are very similar for both heights. For coastal navigation, the higher antenna has slightly better results for Galileo. For GPS, the higher antenna has a slightly worse behavior in port. In maritime navigation, sea state is traditionally recorded in the ship log using the Beaufort scale. This scale, assigning numbers from 0 – Calm to 12 – Hurricane, is essentially subjective, basing itself on visual observation of the sea and of the ship. In time, ranges for wind speed and wave height have been assigned to each Beaufort number. Regarding the influence of the sea state on the multipath error recorded by a GNSS receiver onboard a ship, it is to be noted that this is only one factor in the ship’s roll, pitch and yaw movements. Other factors are ship’s velocity with respect to the wind and waves, ship’s shape, mass and loading, sea currents, etc. In order to avoid an arbitrarily defined ship movement scale, we propose a re-interpretation of the Beaufort scale. Thus, we compute the average ship movement associated to a given sea state Beaufort index, including the relative wind direction. The corresponding values are collected in a ship movement, Beaufort equivalent scale. In order to compute the average ship movement, first we compute the average absolute movement angle between heading and wind direction and for each Beaufort index. All the average movement angles are averaged once more for each Beaufort index, resulting in 7 values. The Beaufort equivalent scale is an objective measure of the ship movement that takes into consideration all factors (sea state, sea currents, ship mass, ship velocity, waves, etc.). This is possible by continuously recording the ship’s angles. In our analysis, this equivalent scale was used to classify each 10 minutes data sample, irrespective of the actual Beaufort index recorded for the sea state. The models for Beaufort levels 7-9 present lowest values for multipath. This is unexpected, as it was assumed that higher ship movement shall induce higher multipath. It may be explained by the fact that, instead of an increase in multipath, an increase in signal losses may appear. Moreover, when more movement is experienced, the satellite sighting is much more fragmented, only high elevation satellites presenting long continuous arches. Moreover, the smoothing method was also analyzed for a subset of data, by comparing the “raw” multipath results with 100 seconds and 300 seconds smoothing. It was revealed that smoothing has a positive impact on multipath. In most of the cases, a small improvement is observed between 100 s and 300 s smoothing. Conclusions Considering multipath, there is practically no difference between coastal navigation, open sea navigation and port approach. As expected, the antenna environment has the highest impact on multipath. While the three vessels are quite different in shape, size and function, relatively similar levels of multipath are observed between Mare Nigrum and Catuneanu, while Istros is presenting roughly half the level of multipath compared to the others. The GNSS antennas’ environment is rather similar on Mare Nigrum and Catuneanu, with multiple possible multipath sources, whereas almost no multipath sources are present on Istros. Port environment, including berth area, has very similar multipath effects on all three types of ship. Traditional methods for reducing the impact of multipath, such as choke ring antennas and smoothing, may be used with very good results. Ship movement has a small impact on multipath.