Centimeter-Level GNSS Positioning Using C-ITS for Correction Data Delivery: An Experimental Study

Accurate and reliable positioning in all environments is important for highly automated driving. For reaching the required positioning accuracy and reliability in automated driving (cf. [1]), different sensor technologies are typically used (see e.g., [2‐4]). In this context, Real-Time Kinematic (RTK) Global Navigation Satellite System (GNSS) can provide valuable global position information with an accuracy that can reach the cm-level (cf. [5]). RTK positioning requires the rover to use different types of measurements and information (typically denoted as corrections) from a positioning service provider, and typically these corrections are sent via Internet/fourth-generation (4G) in the Radio Technical Commission for Maritime Services (RTCM) format. This implies that RTK is not available in regions and areas where 4G is not well deployed.

In the future, systems with an increased level of autonomy will intensively utilize infrastructure support to handle the requirements of the automated driving task. Cooperative Intelligent Transport Systems and Services (C-ITS) play a vital role in this regard for the exchange of information between the vehicles and the road infrastructure. C-ITS, in general, provides benefits in various areas, including improving road safety, reducing congestion, optimizing transport efficiency, enhancing mobility, increasing service reliability, reducing energy use and environmental impacts, and supporting economic development (cf. [6]). C-ITS could also improve the management of urban surface traffic and make it more efficient and safer (cf. [7]). Hence, due to the strong expected development of C-ITS (cf. [8]), it may be an attractive way for RTCM correction data delivery for GNSS, which could be used as an alternative or supplement to 4G to expand the RTK coverage. This would significantly increase the availability of high-accuracy localization. In this paper, we experimentally investigate the feasibility and performance of using C-ITS to deliver RTCM messages and compare it with 4G. Therefore, we collected data in a real scenario with a car traversing the highway A2 in Austria. The work has been done in the H2020-EUSPA funded project ESRIUM (cf. [9‐11]).

Bild vergrößern

Figure 1 shows a simplified scheme of the main components of the localization process in our experiments and the flow of information. The equipment installed in the vehicle is represented inside a light-orange square on the right-hand side of the figure. Two Septentrio Mosaic X5 GNSS receivers were connected to the same NavXperience 3G+C geodetic antenna via an active splitter and configured to compute RTK positions from GPS L1/L2 and Galileo E1/E5a satellite signals with fallback to so-called single point positioning (SPP) at 20 Hz. Thus, the receivers produced so-called fixed or floating positions when RTK was possible depending on whether they could or not solve for the cycle ambiguities (cf. [5]), and SPP when not, where only data from the satellites were used in the position, velocity and time (PVT) computation. Each receiver independently computed RTK positions using the incoming RTCM messages. The upper one received the messages via 4G, and the lower one via C-ITS. The position solutions from both receivers were stored together with additional relevant information for offline analysis. The RTCM messages were generated by the Austrian positioning service provider EPOSA using the virtual reference station (VRS) method (cf. [12]). When 4G was used, the GNSS receiver shared its approximate position with EPOSA using the networked transport of RTCM via Internet protocol (NTRIP). EPOSA then generated RTCM messages with observations corresponding to a hypothetical, not existing (and therefore virtual) receiver located near the user’s receiver and sent these messages back to the user’s receiver.

Bild vergrößern

When C-ITS was used to deliver the RTCM messages, an NTRIP client forwarded a mean position of the testing area to EPOSA’s NTRIP caster. EPOSA again generated and sent dedicated RTCM messages to the NTRIP client. The received RTCM messages were forwarded to the road side units (RSUs). Once in the RSUs, the messages were encapsulated in RTCM extended messages (RTCMEMs) according to the C-ITS standard (cf. [13]) and broadcast, received by the on-board-unit (OBU)¹ of cars in the vicinity of the RSUs and then forwarded to the car personal computer (PC), where they were finally converted back to RTCM and handed over to the GNSS receiver. For our test drives, Yunex Traffic prepared two RSUs for sending RTCM messages, which were separated by 900 m along the A2 motorway in Austria, (cf. Fig. 2b).

The motorway in the test region has noise barriers on both sides of the carriageway and in the middle, as well as hills and trees behind the noise barriers.

The used test vehicle is shown in Fig. 2a. In addition to the equipment represented in Fig. 1, a tactical-grade iMAR iNAT-FSLG-01 GNSS/IMU system was installed to generate the ground truth solution.

When comparing the RTCM message delivery via 4G and C-ITS, the mean message age for the C-ITS-based delivery method was around one second larger on average. The minimum observed RTCM data age for both methods was 0.55 s. For 4G, the typical age was 0.55 s rising to 1.5 s until the next RTCM message arrives (RTCM message rate is 1 Hz). For C-ITS, the typical age was 1.55 rising to 2.5 s until the next RTCM message arrives. The analyses showed that the architectural design of the C-ITS demonstrator was leading to this difference. This architecture can be easily optimized for deployment at the production stage.

Figure 3 shows the cumulative distribution function (CDF) of the horizontal (2D) position errors, Table 1a their 95 and 99.5% percentiles together with the availability of positions with errors below 10 and 20 cm, and Table 1b the percentages of position types obtained. From the tables and figures, we see that the percentage of RTK-fixed solutions is very high in both approaches (98.78% and 98.90% when using 4G and C-ITS, respectively) and that their accuracy is comparable, with errors below 10 cm for more than 99% of the time. This implies that the additional latency when using C-ITS did not significantly impact the accuracy. For PVT solutions with errors above 10 cm, the reliability of the receiver is very good always reporting a correct (overbounding) error estimate when applying a factor of 2.5 to the standard deviation.

Bild vergrößern

Some issues in our setup allowed us to investigate the impact of RTCM data gaps on the position accuracy. Such gaps typically occur in urban and suburban situations. In theory, a GNSS RTK solution requires continuous RTCM data and a little data delay for optimal performance. The test data shows that as long as an integer fix was possible, the Septentrio Mosaic X5 GNSS receiver was able to keep the 2D position errors below 20 cm up to 50 s. For float solutions, the errors stayed below 50 cm if the initial float solution was stable, otherwise, the errors can be as large as 22 m (maximum observed horizontal error of a float solution) (for details, cf. [11]).

This robustness against small coverage gaps increases the usability and suitability of RTK for high-accuracy positioning in automotive applications using both, 4G and C-ITS data links.

In summary, the observed horizontal position errors in our tests were below 10 cm for more than 99% of the time using either 4G or C-ITS as long as a data link was present and the transmission delays remained reasonably small. Gaps of a few seconds can be bridged by modern mass-market GNSS receivers like the used Septentrio Mosaic x5 while still maintaining a significant accuracy (better than 10 cm in our tests). For gaps of up to 50 s, the error could be kept \(\le \)20 cm under specific circumstances. In addition to confirming the feasibility of using C-ITS for RTCM message delivery, this relatively large tolerance to communication delays enables the use of different and complementary data links for the delivery of the necessary RTCM corrections, and lowers the barrier to the spreading of RTK positioning for high accuracy in automotive applications.