Road Talk: Interoperability Considerations for V2X GNSS Receivers

March 1, 2011  - By 0 Comments

By Chaminda Basnayake

The presence of different types of devices, spanning multiple GNSS receiver types, configurations, hardware, software, and consequent widely varying capabilites, among a user mix of vehicles, cyclists, and pedestrians, poses several engineering challenges for a V2X scheme in which all road users share data with each other and with the road infrastructure.

The use of location awareness for transportation safety, efficiency, and security — a major area of research and development for academics, automotive manufacturers, and organizations such as the U.S. Department of Transportation — has focused attention on enabling communication between vehicles and other road user entities in a concept know as V2X, a term encompassing both vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems, so that they can share location and other status information. As a result, any road user entity may see all others around it. This capability is almost always built on GNSS technology.

Future V2X systems will be able to include all road user entities, ranging from vehicles to cyclists to pedestrians, in this information-sharing system. While it sounds natural for everyone to talk to each other and share data for collective benefit, the presence of different types of devices among this user mix poses several engineering challenges. As an example, a V2X device in a vehicle may have a built-in GNSS receiver with a roof-mounted antenna and another vehicle may have a retrofitted V2X device with a passive antenna and relatively limited accuracy capabilities. As the GNSS technology further develops, some vehicles may have multiple-frequency GNSS capability compared to legacy single-frequency devices. In essence, all compatible V2X devices will have to be carefully designed to ensure their interoperability with the rest of the system.

This article investigates positioning challenges arising from multiple GNSS receiver types, configurations, hardware, and software in a V2X operational environment. This produces a clear need to have minimum performance standards for V2X-capable GNSS receivers. The article further investigates the implications of land-based visibility obstructions on relative positioning, and implications on standalone position accuracy both as a result of limited GNSS satellite visibility and WAAS satellite visibility.

streetScene-W

Figure 1. V2X system users.

V2X Background

V2X systems rely on two critical enabling technologies: communications and positioning. Organizations and industry collaborations have developed and demonstrated various V2X systems over the last few years. These efforts have produced interoperable prototype V2V and V2I systems and over-the-air (OTA) messaging standards.

Figure 1 illustrates the general concept of combined V2V and V2I, or V2X. In a fully operational system, all vehicles and other road users carry short-range communication and positioning technology. At present, these technologies are expected to be based on dedicated short-range communication (DSRC) and GNSS, respectively. This enables each user to be location-aware and capable of sharing their location with others. Vehicles may use built-in systems, retrofitted devices, or those based on the occupant’s personal mobile device. Infrastructure elements and other road users such as pedestrians also form part of the V2X user community.

V2X Relative Positioning. Relative positioning of all communicating entities with respect to a given user is a required functional capability of a V2X system. To enable this functionality, positioning information from all communicating entities must be exchanged. For automotive V2X applications, Society of Automotive Engineers (SAE) J2735 DSRC Message Set Dictionary serves as the primary standard for message definitions. Current version of the messages consists of a basic safety message (BSM) , an optional variable rate message (VRM), and an option for including proprietary messages.

With BSM and VRM, vehicle position, speed, heading, and GNSS measurements can be communicated to others. GNSS relative positioning techniques such as real-time kinematic (RTK), code-based differential, or individual position differing (that is, distance between the positions reported by individual vehicles) can be used for relative positioning. The latter method, also known as DPOS, is a particular focus of this article.

Given the above, a system developer may develop a V2X relative positioning system that can operate based on techniques that can be broadly classified as position-based techniques, which include DPOS, and measurement-based differential techniques, including RTK and others.

The Simpler Approach. The SAE J2735 BSM accommodates the simpler approach of using the DPOS method, as it enables the sharing of critical state parameters. This approach is very attractive as it requires minimal OTA data volume compared to sending GNSS measurements. Secondly, DPOS relative position estimation process requires only a fraction of the processing resources required compared to measurement-based differential processing. Thirdly, any GNSS receiver in the market today is capable of outputting a position solution and most of the critical GNSS state parameters required for the V2X BSM. In contrast, most low-cost devices do not output measurements required for other methods.

However, there are quite a few challenges associated with DPOS. A vehicle or any other road-user entity, such as a location-enabled handheld device, will share its location information via BSM only. A relative positioning engine in each entity will use this information to provide lane-level and road-level data (relative distance, speed, and orientation) for its V2X applications. The challenges associated with DPOS method arise from multiple stages in this process.

The presence of many road-user types brings in the possibility of thousands of GNSS receiver types, models, hardware, and software in the user group. Thus the system must be interoperable with devices with a wide range of performance characteristics.

Secondly, each entity will transmit BSM only. This OTA information offers no information about the constellation the GNSS device sees or how the solution was derived in terms of filtering or applied constraints.

Thirdly, the position accuracy reported by each entity is a GNSS device-dependent variable, an estimate of the actual error as derived by a user device.

Finally and most importantly, V2X applications expect relative positioning information for each communicating entity classified in one of three possible accuracy categories: Which Road, Which Lane, or Where-in-Lane (see “Is GNSS up to the V2X Challenge?” GPS World, October 2010). The V2X system must be able to reliably identify this accuracy classification for each communicating entity with the limited information provided via the BSM.

Study Goals. To illustrate the impact of these challenges, several GNSS receiver types, configurations, and operational scenarios were investigated.

  • Between multiple receiver types: In a V2X environment, vehicles and other road user entities may have different GNSS receiver types and makes: dual-frequency, single-frequency, and so on.
  • Same receivers using different parts of visible constellation: In an urban canyon, it is possible for two adjacent vehicles to see two different parts of the GNSS constellation, due to obstructions.
  • WAAS-enabled and non-WAAS receivers.

Data Collection

This data is a combination of field-data collections and a series of RF record playbacks. The field vehicle-mounted test setup included two GPS receivers, a GNSS L1 RF data recording device, and a high quality GPS/INS reference system (Figure 2). Type A receiver is a hi
gh-sensitivity enabled, automotive-grade GPS L1 receiver using a patch antenna, WAAS-capable although WAAS usage was disabled in the real-time data collection. Type B receiver is a high-quality L1/L2 receiver using a geodetic-grade antenna, used with WAAS enabled. The GPS/INS system was connected to the geodetic-grade antenna. The RF recording system was also connected to the automotive-grade GPS L1 antenna.

Figure 2. Vehicle test set-up.

Figure 2. Vehicle test set-up.

The data was collected on a test route in Detroit, Michigan, that included durations of urban and deep urban canyon (40 miles per hour (mph) or less), freeway (55–70 mph), and suburban/local (30 mph) driving.

The RF data were subsequently replayed to GNSS receivers that were not a part of the field set-up. RF data was also replayed to receivers with forced sky-visibility obstructions and various WAAS settings. For limited sky-visibility tests, certain satellites were removed from each receiver’s view by receiver-specific configuration software. The satellite selection and restriction was done to mimic typical sky-view obstructions encountered in normal driving.

Type A receiver was chosen to illustrate the impact of visibility differences. A total of 13 satellites were visible in the entire data set (Figure 3). To create obstructed sky-view scenarios, two Type A receivers were configured to not use certain satellites in their position solutions. These configurations were:

  • Configuration 1 (C1): PRNs 7, 10, and 13 blocked
  • Configuration 2 (C2): PRNs 6, 16, 21, and 31 blocked

C1 mimics a vehicle/receiver with no visibility in the Northwestern part of the sky, whereas C2 mimics a receiver without visibility in the East/Northeastern part of the sky. Sky visibility restrictions do not vary with the heading changes of the vehicle. For example, for C1 receiver, Northwestern sky is always obstructed regardless of the vehicle orientation.

B-4

Figure 3. Sky view during the test.

Figure 4 shows an example RF data replay setup. The record and replay system was controlled through a PC and the recorded data was also stored in the controller PC. The output RF signal was split into multiple outputs such that multiple receivers can be tested at the same time. For each replay of the RF data, a benchmark receiver was also used to verify that there is no run-to-run difference as a result of the RF replay.

Outputs from each GPS receiver from field and replay runs were logged to PCs using receiver specific binary formats. The recorded output from each receiver included its position, position error estimate, velocity, satellite-specific measurements and indicators such as pseudorange, carrier phase, and signals-to-noise ratio.

Figure 4. RF data replay set-up.

Figure 4. RF data replay set-up.

Data Processing and Analysis

The data was first decoded from the receiver-specific formats to a common format, then corrected for antenna offsets. To simplify the process, the reference system position solution was translated to the position of the test antenna using the known between-antenna distance and orientation of the vehicle as measured by the reference system. As a result, all the receivers and the reference system are reporting the location of the test antenna. Then, data fields such as position and velocity for each receiver were time-matched with the reference solutions, and the actual error was calculated.

For a limited dataset, additional measurement-level differential processing was done to show the difference between a DPOS and an RTK or a code-based differential relative position solution.

Figure 5 shows a plot of the 2D position error observed from each receiver during the test as a function of driving environment. Overall, Type B receiver shows better accuracy as expected from a dual frequency high quality receiver. However, it shows spikes of large error increases at times, mostly observed in the freeway scenario with large error excursions. With Type A receivers, relatively larger errors are observed with the limited-constellation receivers.

Figure 5. Position error (2D) of each receiver as a function of driving environment.

Figure 5. Position error (2D) of each receiver as a function of driving environment.

Figure 6 shows the number of satellites used by each receiver in the same environments as in Figure 5. Overall, Type A receiver tracks and uses on average 2–3 satellites more compared to the Type B receiver, likely due to its high-sensitivity capability. Type A C1 and C2 receivers also track and use 2–3 satellites fewer compared to the all-in-view Type A receiver.

Figure 6. GPS satellites used by receivers.

Figure 6. GPS satellites used by receivers.

Freeway Data. The vehicle heading in this segment was predominantly north or northwest. The sky view can be considered a combination of urban and open sky conditions. As highlighted in Figure 6, all-in-view Type A receiver was able to use up to 11 GPS satellites with an average of around 9 satellites. Type A C1 and C2 receivers used, on average, about 3 satellites fewer than the all-in-view receiver. All three receivers show satellite count drops down to 4 at certain times in this segment.

The satellite count of the Type B receiver shows the limitations of not using the high-sensitive tracking capability. The satellite count shows frequent drops below 4 satellites and on occasion down to no satellites used.

Although the satellite count difference between all-in-view Type A and C1/C2 receivers was forced by means of receiver configuration, short-term sky visibility restrictions that resemble these conditions are in fact possible. Examples include a passenger car driving next to a semi truck or the side wall of the freeway in below-ground road sections.

Figure 7 shows the position outputs of all four receivers on a satellite image of a short segment of the freeway. The true location (reference) is shown in green. Type A, Type B, Type A C1, and Type A C2 are shown in red, black, purple, and blue, respectively. These colors identify the four receiver types in all figures for the rest of this paper. While biases can be seen in the outputs of all four receivers with respect to the reference, the Type A C1 shows the largest offset with the magnitude of more than a lane width.

Figure 7. Freeway positioning accuracy.

Figure 7. Freeway positioning accuracy.

Figure 8 illustrates a time series of the positioning error components of all four receivers. It clearly shows error ramp-ups from the Type B receiver at frequent intervals. These coincide with the satellite count drops of Type B shown in Figure 6. No such error ramp-ups are observed for any of the Type A receivers, although relatively large biases of the order of few meters can be seen. As anticipated, larger errors are observed in the height direction.

Figure 8. Freeway positioning accuracy time series.

Figure 8. Freeway positioning accuracy time series.

Local Road, Eastbound. This segment includes data gathered on an eastbound multi-lane local road with 40 mph posted speeds. As shown in Figure 6, a relatively larger number of satellites were continuously tracked in this segment as compared to the freeway. Therefore, this segment is considered to be an open-sky scenario with very limited number of obstructions. As shown in Figure 6, Type B receiver has used about 6 satellites on average, whereas the Type A has used around 3 more satellites most of the time. Type A C1 and C2 have also used around 3 satellites less compared to the all-in-view Type A receiver.

Figure 9 shows the vehicle position as reported by all three receivers and the reference system output for a short road segment in this drive. It clearly illustrates the lateral offsets of both C1 and C2 solutions. The C2 receiver (Blue) generated about a lane width offset towards north whereas the C1 receiver output is biased by around two lane widths to the south. Figure 10 presents a time series look of the positioning biases evident in Figure 9. It clearly shows large (more than 5 meter) biases in North and East position error components for C1 and C2 receivers.

Figure 9. Local (east) positioning accuracy.

Figure 9. Local (east) positioning accuracy.

Figure 10. Local (east) positioning accuracy time series.

Figure 10. Local (east) positioning accuracy time series.

Local Road, Northbound. In roadway characteristics, this resembles Local Eastbound. Figure 6 shows the sky view remained almost unchanged for Type A receivers. For Type A C1, the count remained at 6 throughout. C1 and C2 receivers tracked 2–3 satellites fewer compared to all-in-view Type A. Interestingly, Type B experienced two dropouts of 4 or fewer satellites during the run. Figure 11 shows the position output of all receivers on a short road segment. As in the case of Local (East), significant biases can be readily observed in the output of C1 and C2.

Figure 11. Local (North) accuracy.

Figure 11. Local (North) accuracy.

Figure 11. Local (North) accuracy.

Figure 12 shows the time series view of the positioning error in this segment, confirming that the biases observed in Figure 11 are not short-term biases, but are in fact vehicle heading-dependent biases. The short-term biases seen in the Type B receiver output coincide with the change in the number of satellites used (shown in Figure 6). This illustrates the implications of different estimation methods used in the two receiver types. For instance, Type B receiver allows stepwise changes in its position estimate whereas Type A output tends to gradually converge to different states.

Figure 12. Local (North) positioning accuracy time series.

Figure 12. Local (North) positioning accuracy time series.

Urban Canyon. Results of the urban canyon segment of the drive are shown in Figures 13 and 14. A statistical analysis is not presented for this segment, as receiver type and configuration dependent biases and errors are difficult to isolate from the errors that are the result of multipath and measurement noise. In Figure 14, much larger biases in the order of 10 meters or more can be seen for all three Type A receivers. In comparison, Type B receiver tends to output a relatively accurate position solution whenever it has sufficient satellites visible. In the case of less than optimal satellites availability, Type B receivers tend to show rapidly degrading positioning accuracy, which may be reliably detected using its quality indicators.

Figure 13. Urban canyon accuracy.

Figure 13. Urban canyon accuracy.

 

Figure 14. Urban canyon positioning accuracy time series.

Figure 14. Urban canyon positioning accuracy time series.

Position Error Distributions

Position error probability distribution functions were generated for the first three road segments using the time series data above. Figures 15-17 show these functions for Freeway, Local (East), and Local (North) segments, respectively. They lead to these general conclusions:

  • Based on the mean and the spread of the distributions, Type B receiver has consistently provided the most unbiased and accurate positioning performance out of all the receivers considered. Overall, the output appears to be unbiased, as should be the case for a high quality dual frequency receiver with WAAS capability.
  • Type A all-in-view receiver shows the next best overall accuracy statistics with some biases in certain cases. These biases can be time-of-day-dependent and may differ for different times of the day or if observed over a longer time.
  • Type A C1 and C2 receivers show very significant vehicle-heading-dependent biases/errors. This is with very limited sky view obstructions (that is, C1 only restricts less than 1/8 of the entire sky view whereas C2 covers around 1/4) and with the same type of the receiver.
  • Although enabling WAAS should theoretically help minimize the biases observed in these tests, the availability (open line-of-sight) of WAAS satellites for automotive applications in these environments must be taken into consideration for WAAS accuracy benefits to be applicable. For these datasets, WAAS signals availabilities for a Type B receiver were 58 percent of total driving time in urban canyon, 60 percent in the freeway scenario, 95 percent and 99 percent in the local road scenarios.
Figure 15. Freeway position error distribution.

Figure 15. Freeway position error distribution.

Figure 16. Local road (east) position error distribution.

Figure 16. Local road (east) position error distribution.

Figure 17. Local road (north) position error distribution.

Figure 17. Local road (north) position error distribution.

Velocity Domain Performance. For each test segment, velocity estimates from each receiver were logged at the default data rate of 4 Hz. For analysis purposes, North and East velocity readings from each receiver were converted to 2D speed estimates. These were used with reference system speed estimates to generate 2D speed error statistics (Table 1).

Based on Table 1, no significant biases or errors were observed from any particular receiver or configuration. The only exception was the increased errors in the Urban Canyon segment, particular for C1 and C2. This is expected .to be a result of limited satellite availability in a challenging environment with additional forced satellite eliminations.

Table1

Table 1. Receiver Velocity Error Statistics (meters/second).

Virtual Two-Vehicle Analysis. Assume that Type A and Type A C1 receivers were located in two vehicles. Ideally, both receivers should report the same location, as they were both connected to the same antenna on a single vehicle, creating a zero-baseline scenario. However, as shown in the previous section, a meter-level separation was observed between the two solutions.

In this virtual two-vehicle scenario, relative position of one receiver (Type A) with respect to the other (Type A C2) was estimated by three methods, using GNSS data processing software in post-mission. The methods were:

  • Differenced Positions (DPOS). Latitude and longitude reported by each vehicle were time-matched; distance between the two points was calculated.
  • Code and Carrier. Single frequency (L1) GPS RTK positioning with float ambiguity estimation.
  • Code Only. GPS code measurements generated a relative position solution.

The 2D receiver separation results of this processing are shown in Figure 19 as three subplots for freeway (top), local/east (middle), and local/north (bottom) scenarios. The 2D separation results for local scenarios show clear performance benefits for the GNSS measurements-based methods. In both east and north local scenarios, around a 5-meter bias is observed in the DPOS solution whereas this is reduced to around a meter in both code-only and code and carrier methods. The freeway scenario shows relatively smaller difference potentially due to measurement noise, multipath, and frequent interruption of sky view. Table 2 shows mean values of these results.

Figure 18. Position separation for processing methods.

Figure 18. Position separation for processing methods.

Table 2. Mean Accuracy (meters) using processing methods.

Table 2. Mean Accuracy (meters) using processing methods.

Discussion

OTA transfer of certain GNSS measurement data elements appears to be a critical requirement for reliable lane-level positioning capability. However, the method must be capable of supporting a certain level of performance even in challenging environments for GNSS. The solution for such challenging environments is likely to be GNSS integration methods with vehicle-based sensors (that is, GNSS/INS) for the foreseeable future.

Given these facts, a reliable and accurate V2X relative position method will require the OTA transfer of a combination of critical vehicle states which include the vehicle location, a confidence measure, and certain GNSS measurement data elements. With its ability to support all of these needs, the SAE J2735 provides a basic framework for further refinement of relative positioning technologies for automotive applications.

A reliable position confidence measure broadcast over-the-air is also a critical need, particularly if GNSS measurement data is not broadcasted on a regular basis. This also holds true for conditions under which a vehicle may be operating in a GNSS and vehicle sensor integrated mode or with less than optimal number of satellites in view. However, estimating such a parameter that can be trusted with high degree of confidence can be challenging given the presence of various biases that can depend on the environment, vehicle, GNSS receiver, and sensors and methods used. Potential examples are time-of-day, vehicle heading, vehicle speed, GNSS receiver/sensor type, model, and configuration. However, developing a parameter similar to the RTCA Horizontal Uncertainty Level (HUL) for automotive applications is an important consideration.

While there are many other candidate receivers to be considered for a study of this nature, only two receiver types were used in this analysis. Analysis of more receiver types can be beneficial to identify the desired characteristics for a certain applications. A consideration could be achieving a desirable balance between accuracy and the sensitivity of the GNSS receivers, as increased sensitivity often produces higher solution availability at the cost of degrading accuracy.

Another area to investigate in related work is the benefits of using WAAS under the test conditions given in this paper. The general expectation is to see less bias in the position solution with WAAS as the ranging errors are likely to be smaller as a result of WAAS corrections. However, for automotive applications in particular, availability of WAAS signals to land vehicles need to be investigated.


CHAMINDA BASNYAKE is a senior research engineer at General Motors Global Research and Development and GNSS technology expert for GM OnStar. He leads GNSS-based vehicle navigation technology R&D efforts at GM and holds a Ph.D. in geomatics engineering from the University of Calgary.

This article is tagged with , , and posted in Road
GPS World staff

About the Author:

Post a Comment