Previous Abstract Return to Session D3 Next Abstract

Session D3: GNSS Augmentation and Robustness for Autonomous Navigation

Modelling of EGNOS Navigation System Errors for Cat I Autoland
C. Milner, C. Macabiau, A. Gantelet, S. Prakasen, Ecole Nationale de l’Aviation Civile; F. Tranchet, L. Azoulai Airbus
Date/Time: Thursday, Sep. 22, 11:03 a.m.

Peer Reviewed

Implementations of the Satellite Based Augmentation System (SBAS) concept are providing the capability to perform LPV operations in continental U.S (CONUS) and European airspace down to 200ft minima, equivalent to Cat I minima. Aircraft equipped with receivers meeting RTCA DO229D/E standards have been shown in flight trials to potentially be capable of performing Autoland under Cat I conditions. Automatic landing (Autoland) starts "from the beginning of the landing flare until aircraft exits the landing runway, comes to a stop on the runway, or when power is applied for takeoff in the case of a touch- and-go landing" [ICAO-CAST, 2010]. Airbus and ENAC as members of ICAO’s All Weather Operations Harmonization Working Group (AWOHWG) helped to develop a Navigation System Error (NSE) model to be used for demonstration of Cat I Autoland certification of the Ground Based Augmentation System (GBAS).
Such a demonstration must be statistical as requested by the regulator, and as such Monte Carlo simulation has been employed accounting for both situations of average risk (nominal case) and limit case risk (faulty case) [CS AWO 131]. The results of this simulation lead to statistical metrics regarding landing within the touchdown box that is defined for a safe landing. Limiting the probability of a short landing is the most critical need, short landings resulting from negative errors in the vertical direction (true aircraft position below the glidepath).
Since GBAS Autoland certification, Airbus and ENAC have continued to develop an NSE model with the goal of achieving Autoland with SBAS. It must be noted that whilst ranging models are available, simply from combining SBAS messages with the processing described in the MOPS, the residual error standard deviations contain significant margin, with respect to the nominal case, since they are designed to support the SBAS range bounding concept to ensure integrity at the 10-7 level. Therefore, analysis is needed to assess nominal performance of SBAS, described by tighter bounds of the residual error variation that those broadcast by the system. This work began with a concept paper [Azoulai et al, 2012] which outlines the basic approach to NSE modelling and has continued in the frame of the COCOTIER project. COCOTIER (COncept de COckpit et Technologies Intégrées En Rupture, 2019-2022) is a collaborative project on new technologies for future intelligent cockpit in Single Pilot Operations (SPO, 2030+). The project is supported by the French Directorate General of Civil Aviation (DGAC) and coordinated by Airbus. This paper presents the end-to-end developments of an NSE model in COCOTIER. The methodology has been applied to data from 2014 and 2020. The year 2014 was chosen initially, due to it being the solar maximum of the current solar cycle. However, due to possible evolutions and improvements in SBAS processing and as a performance cross-check, the more recent 2020 data is analyzed.
The NSE model is developed in three phases. In the first phase, the residual ranging errors following SBAS (EGNOS in the case studied) corrections are modelled. This relates to the corrected Signal-In-Space errors (orbit, clock, nominal signal and antenna biases) as well as the corrected ionospheric errors. The tropospheric and airborne (multipath and noise) errors are not addressed, and the MOPS models are used in the following phases.
Estimates of the residual errors are obtained by the difference of the EGNOS corrections (processed for both 2014 and 2020) and a respective truth model. The fast and long-term corrections are compared to an effective IGS correction obtained from using the precise orbit and clocks to correct the navigation message based solutions. In the case of the ionosphere, accurate truth references from UPC in terms of Slant TEC are used. These estimates are collected over a global network of stations and determined employing a carrier phase-based network solution. Difficulties arise in each of these processes regarding both the choice of parameterization, normalization of the errors, statistical characterization, and the presence of differential biases (DCBs). The paper presents in detail how each of these issues is addressed. Ultimately, residual fast and long-term corrections are normalized with the respective fast and long-term standard deviation (from the UDRE following processing by PEGASUS [ECTL PEGASUS]) and collected into bins parameterized by the satellite position latitude and longitude and the SVN. It is critical to estimate and remove the clock offset between EGNOS system time (the basis for the corrections) and IGS time (the basis for the truth). Furthermore, differential biases between EGNOS and IGS per satellite are removed on a monthly basis. These biases were observed as being stable over many days and may be partly explained by the presence of components of the EGNOS corrections correcting nominal biases which are not included in the purely orbit and clock components of the IGS truth. Biases of the fast and long-term correction residual errors were of the order of +/- 50cm with standard deviations in the range of 20-60cm. Higher values are obtained for satellite locations at the edge of the EGNOS reference receiver network, although this effect is lessened by normalization.
Parameterization of the ionospheric errors is made with respect to the Ionosphere Pierce Point (IPP) [RTCA, 2020]. Decimeter level biases and standard deviation were also observed in line with those for the FLT residuals. The output from both processes were therefore tables of cells containing the mean (normalized) error, the standard deviation of the (normalized) error, the number of samples and the mean of the normalizing standard deviation. The normalization process is critical in order to have data that is nominally from the same distribution when determining the mean and standard deviation of each bin. The mean of the normalizing standard deviation (either the sigma FLT or sigma UIVE after cancelling the effect of elevation with the obliquity function) is also of importance, firstly for use in the weight matrix employed by the SBAS airborne algorithm and secondly to obtain errors back in the range domain (un-normalized). Finally, the number of samples is critical to ensure that only data which has a significant quantity of data points is used in the position domain processing of phase two.
The second phase involves utilizing the residual range error models obtained, with knowledge of the satellite geometry (almanac), taking certain assumptions regarding the constellation (24 baseline, 23 degraded) in order to determine the position domain performance. Once again simple mean and standard deviation statistics in both the horizontal and vertical domains are computed for a selection of users (over the LPV-200 compliant sub-region of ECAC) and at thirty second intervals. Moreover, SBAS protection levels are calculated and compared to LPV-200 requirements to ascertain the availability of performing an approach with a 200ft minima, a prerequisite for Autoland.
An intermediate crosscheck using real receiver measurements is conducted to validate the distributions of the SBAS range residuals obtained. That validation step consists in computing the distribution of real range observation residuals at European IGS stations. These residuals distributions from IGS stations measurements reflect some errors which are also considered in the above derivation (ephemeris and clock errors, iono errors), but reflect also some error which are different from airborne observations (different environment, different receiver and antenna constraints). Indeed, multipath, noise, tropospheric errors, receiver antenna, SV antenna and nominal signal distortions impact these real residuals, and this is an interesting complement to the residuals computed above as broadcast minus reference data. These real residuals are computed for years 2014 and 2020. This final step serves as a crosscheck in the sense that if the distribution of the residuals obtained in the steps above using EGNOS and GPS broadcast data exceeds the distribution of the real observed residuals, then the results obtained above are sufficiently conservative.
In the final phase, Gaussian error generation is used to obtain a single model for each user averaging over time the outputs from phase two. The output of phase three is then a distribution of navigation system errors (NSE) both horizontally and vertically, to be employed in Airbus Autoland simulations. It is observed for the 2014 processing that in the central ECAC region the Autoland nominal case requirements are met whilst performance is borderline insufficient at the corners (most significantly the South-East). Processing for 2020 is currently on-going and the 2020 results will form part of the completed paper.
The proposed paper will therefore start with the presentation of the Autoland requirement framework, then will address the modelling of the residual ranging errors following SBAS (EGNOS in the case studied) corrections using EGNOS and GPS broadcast data for 2014 and 2020. In a third section, the paper will present the statistics of the position error performance derived from the range error model. Then, the statistics of the range residuals obtained from IGS stations range observations for 2014 and 2020 will be presented, and it will be checked that the distribution of the residuals obtained in the steps above using EGNOS and GPS broadcast data exceeds the distribution of the real observed residuals, then the results obtained above are sufficiently conservative. The paper will conclude with a distribution of navigation system errors (NSE) both horizontally and vertically, to be employed in Airbus Autoland simulations.

[Azoulai et al, 2012] SBAS Error Modelling for Category I Autoland, L. Azoulai, P. Neri, C. Milner, C. Macabiau, ION GNSS 2012
[AWO] EASA, CS AWO 1, “Joint Aviation Requirements – All Weather Operations”, Subpart 1, “Automatic Landing Systems”
[AWO36, 2010] Minutes of AWO HWG 36th Meeting, Brussels, 2010.
[ICAO-CAST, 2010]: "Phase of flight definitions and usage notes", Commercial Aviation Safety Team and International Civil Aviation Organization, June 2010

Previous Abstract Return to Session D3 Next Abstract