TU-TESL

• 22

Today, the advanced single band GNSS receiver can meet the high precision requirements of v2x, ADAS and in open sky conditions. In order to service reliably in various environments, GNSS receivers need to overcome the limitations in cities and other challenging environments. This paper demonstrates how to achieve this goal by using a multi band RTK inertial navigation system based on GNSS calibration service and vehicle dynamic model. Satellite based positioning plays a unique role in both v2x applications and advanced driving assistance systems (ADAS), including autonomous driving. It is the only technology that can determine the absolute position of vehicle in real time. It is independent of maps, cameras, and landmarks. Because its basic working principle is completely independent of other sensing technologies (such as lidar, camera, ultrasound) used in autonomous vehicles, satellite based positioning can provide important foundation and support for multi-sensor networks that no other technology can provide. Nowadays, GNSS receiver technology is constantly overcoming its limitations. The accuracy is increased to tens of centimeters, and the convergence time (the time required for the receiver to reach a predetermined accuracy level after signal interruption and subsequent re acquisition) is increased to a few seconds. Latency (the time from the measurement location to the time the device reports this location to the network) is approximately 10 milliseconds. The position update frequency can also be more than 10Hz. In addition, with more technological improvements, it can also be positioned in urban canyons, multi-storey roads and other challenging scenes. In short, in the era of v2x and ADAS applications, GNSS has finally achieved technical maturity. However, not all progress has occurred in GNSS receivers. Under the influence of Moore's law, the size of hardware is gradually reduced to a microchip suitable for portable low-power devices in the mass market. The ubiquitous wireless Internet connection enables GNSS correction service to minimize the impact of the ionosphere on GNSS accuracy, which is the main source of GNSS error. In addition, national and international investments in space provide us with new satellite systems tailored for innovative applications. This allows the receiver to use more (visible) satellites, thus gaining key advantages. These advances will enable us to equip our vehicles with the latest generation of multi band, multi constellation GNSS receivers, providing sub meter accuracy (up to tens of centimeters), depending on the application requirements. But what we need is not only the improvement of positioning accuracy. Low latency is another key requirement for emerging applications, such as vehicle to everything (v2x) communication. In v2x, vehicles use wireless messages to "talk" with each other or with roadside infrastructure, and transmit warnings and information about mobile locations when merging and overtaking, and negotiate priorities at intersections. Image 2: the impact of delay in v2x use cases when the impact is minimal, longer delays can cause problems, lead to unnecessary braking and acceleration, reduce traffic efficiency and passenger comfort. In the worst case, delays can be fatal. Especially on expressways, the length of a car is passed every 100 milliseconds. In most cases, the ETSI (European Telecommunications Standards Institute) standard used in v2x communication requires a system level delay of less than 100 milliseconds. The following table summarizes the requirements for different applications in the automotive market. Note: all applications require inertial navigation technology, including wheel speed information. The cep50 value corresponds to the radius of the circle covering 50% of the measured values at all positions. For ADAS, v2x, and ultimately for autonomous driving, GNSS receivers must be able to provide robust Lane positioning, even in challenging environments. When the satellite signal is temporarily blocked, they need to restore high-precision positioning in a few seconds. This can be achieved by fusing and filtering the following multiple complementary sensors. Figure 2: single fusion filter for multi constellation, multi band GNSS receivers for high-precision positioning solutions: the number of global GNSS constellations has increased from one (GPS) to four (GPS, GLONASS, Beidou, Galileo), which means that the receivers can "see" more satellites at any given location. In this way, we can solve the problem that the receiver needs more satellites to locate accurately: only a single constellation needs to use four satellites; but when there are three constellations, about seven satellites need to be used (in order to calculate the time difference between constellations, these constellations themselves have different time reference systems). In addition to more satellites, multiband GNSS receivers can also combine signals of different frequencies, each of which can play an advantage in specific applications. For example, processing two signals from different frequencies simultaneously can effectively eliminate up to 99.9% of ionospheric errors. Another technique called "geometry free combination" helps to detect cycle slips in the carrier phase. All of these technologies can only be implemented by multi band receivers. Real time dynamic (RTK) algorithm: the standard precision GNSS receiver tracks the GNSS signal number phase of at least four GNSS satellites to realize triangulation, while the high-precision GNSS receiver tracks the phase of high-frequency carrier. In order to solve the problem of carrier phase ambiguity, high-precision GNSS receiver uses real-time dynamic (RTK) algorithm. These algorithms have been integrated into some GNSS receiver modules. RTK algorithm widely uses the correction data provided by wireless connection. For the automobile market, the communication based on cellular network and satellite L-band is very suitable. In addition to saving data transmission costs, even in rural areas where cellular network signals are poor or not available at all, L-band receivers can receive RTK correction data via satellite. Broadcast GNSS correction service: the GNSS correction service provider continuously estimates the GNSS signal error by monitoring the GNSS signal from the base station network. For example, ppp-rtk services can compensate for satellite clock, orbit, signal bias, global ionosphere and regional ionosphere and tropospheric effects. Ideally, such a school would be effective in large areas such as the U.S. mainland, with the lowest bandwidth requirements. Traditional services are based on rough location estimation and send customized correction flow to individual users, while service providers use a more scalable method to broadcast the same dynamic GNSS error model to all users. In addition to improving the accuracy of GNSS receiver, high quality correction data can also shorten the time required for the receiver to converge to the exact position. For environments with overhead obstacles (such as flyovers, road signs, trees, and bridges), this feature is essential for normal driving, as these obstacles may temporarily interrupt the GNSS signal. Inertial sensor and sensor fusion: for many years, inertial sensors have been used to enhance GNSS positioning services. Through the implementation of inertial navigation (DR), they enable the vehicle positioning system to make up for the lack of GNSS signals encountered in tunnels, parking lots and other challenging common environments. By fusing the data collected by the components of IMU, the location module can continue to provide the estimated location in the environment of GNSS signal blocking. When the reception of GNSS signal is temporarily interrupted, the fusion of inertial sensor and sensor helps the positioning solution to keep the relevant information of position and speed. Compared with the pure GNSS solution, the fusion solution can shorten the re convergence time when the satellite signal is available again, that is, the time needed to solve the carrier phase ambiguity. On board sensor: combine the data of on-board sensor (such as wheel speed sensor) to further improve the performance of inertial navigation solution. If the algorithm finds that the wheels are not moving, the position change reported by GNSS system (due to signal error) can be ignored. The speed estimation based on wheel speed sensor weighting is more accurate than that only depending on the accelerometer with noise. In addition, the continuous calibration of the moving distance of the wheel speed sensor can correct the errors caused by the changes in winter and summer. Dynamic model: the dynamic model of vehicle can limit the influence of measurement error on position estimation. The model assumes that the vehicle will not slide laterally, jump vertically or accelerate in any unreasonable way. All GNSS measurement data will be checked by the dynamic model before being used in the navigation filter. It is a challenging task to quantify the performance of the above schemes in tunnels. First of all, the main error source is sensor error, and when they are integrated to get the vehicle speed (accelerometer) and attitude (gyroscope), the error tends to accumulate. This is mainly because the error comes from random rather than systematic phenomenon. In order to correctly characterize its impact, a considerable amount of tunnel data need to be collected and analyzed. Secondly, the exact "real" position cannot be obtained to compare with the measurement results. Ideally, positioning based on completely different technologies should be used as a reference in these tunnels to eliminate the influence of GNSS signals being occluded. Finally, even the expensive reference system based on inertial sensor will have drift error to some extent. Instead of testing real system settings in real tunnels, we first create virtual tunnels using data collected in open sky conditions. Therefore, we "disconnect" GNSS signal to simulate GNSS signal interruption, forcing the system to navigate in inertial navigation mode. In this way, we can compare the performance of the IMU with that of the high-end truth system. Recording the position output of the ins solution and the high-end reference GNSS receiver provides us with the necessary data to compare the performance in tunnels of different lengths. With this simple technique, we can run a set of tests large enough to analyze the performance quantitatively and get statistically significant results. Figure 3: in the inertial navigation mode without GNSS, the positioning error on the travel distance is shown in the figure above. By analyzing the data of 1758 signal interruptions generated by 31 tests, we determine that the positioning error on the travel distance is about 2% in the inertial navigation mode. In other words, the error of horizontal positioning increases by an average of 20 meters per kilometer. It is worth noting that the performance of IMU has a significant impact on the tunnel test results. In our configuration, we use a standard IMU with average performance rather than high-end performance. Tunnel simulation is only part of a broader set of equipment tests. In order to verify the above technical combination, i.e. by combining the multi band, multi constellation GNSS receiver with the built-in RTK algorithm, broadcasting GNSS correction data, IMU for inertial navigation, external wheel speed sensor and dynamic vehicle model, we reliably provide Lane accurate positioning, and we also tested it in various situations with different complexity. Due to the randomness of GNSS and IMU errors, the results of individual tests may exceed or be lower than those shown below. In the recent freeway driving, mainly in open sky conditions (the most challenging scenario), our solution can provide 100% availability and accuracy of 5.8 cm in 50% of the time. The accuracy of horizontal velocity component in 68% time is 0.02 km / h. In our test, we used RTK fixed solution (integer ambiguity of carrier phase is fixed), RTK floating-point solution (integer ambiguity of carrier phase is not fixed) and inertial navigation