Network RTK for Intelligent Vehicles

Accurate, Reliable, Available, Continuous Positioning for Cooperative Driving

By Scott Stephenson, Xiaolin Meng, Terry Moore, Anthony Baxendale, and Tim Edwards

Adoption of network real-time kinematic GNSS positioning can lead to major improvements in vehicle localization, although implementation must overcome some real-world challenges. This article assesses the extent of GNSS signal outage in a motorway environment. The average total GNSS outage period and the average time to resolve ambiguity for the network RTK solution can help assess complimentary sensors for a ubiquitous positioning system.

Real-time vehicle localization is one of three key enabling technologies for the concepts of vehicle-to-vehicle and vehicle-to-infrastructure (V2V and V2I, collectively termed V2X, see opening graphic), a classification of intelligent transport systems (ITS). The further enabling technologies are ad-hoc dynamic networking of agents, and accurate dynamic local traffic maps. Jointly, these require that positioning be accurate, reliable, available, and continuous.

A natural evolution in road transport, V2X promises to deliver the next major safety breakthrough. The concept moves away from vehicles making individual decisions about road safety, as in advanced driver assistance systems, and towards a cooperative driving approach that shifts the emphasis from collision protection to collision prevention. The U.S. National Highway Traffic Safety Administration  estimates that V2X technology can avoid or minimize up to 80 percent of collisions of unimpaired drivers, and that even a small number of deployed vehicles will provide tangible safety benefits.

Network RTK GNSS positioning, like V2X applications, requires a communication system; and by its nature V2X has a positioning solution requirement. Thus it is envisioned that network RTK will play an essential role in the implementation of V2X systems. The consensus between car manufacturers and research organizations is that the future of V2X communication lies with Dedicated Short Range Communication (DSRC) devices, and a large pilot study is currently under way. However, in the short term many V2X applications could be achieved using existing technology, such as cellular communication, offering a legacy solution, and initiating early uptake of V2X applications.

Previous research by the Nottingham Geospatial Institute (NGI) at the University of Nottingham showed that network RTK positioning can provide a high-accuracy positioning solution during real-world trials, but also revealed two areas of concern: the loss of the fixed-integer ambiguity during satellite line-of-sight outages; and the fragility of the data communications service that delivers the real-time correction information. During road tests, a fixed-ambiguity network RTK solution was available for less than 50 percent of the time on United Kingdom (UK) roads.

Network RTK Vehicle Positioning

Figure 1. OS Net reference station network in Britain, owned by Ordnance Survey.

Networks of continuously operating reference stations (CORS) extend across Europe, North America, Australia, and East Asia. Networks vary in size from five or six reference stations for agriculture to systems of hundreds of CORSs providing national or regional service. Figure 1 shows the location of the OS Net CORS run by Ordnance Survey in Great Britain.

Figure 2 shows the main advantage of network RTK as compared to traditional RTK. The individual reference stations on the left suffer from the spatial decorrelation of errors as distance between reference and rover receivers increases. Adequate vehicle positioning would require individually operating reference stations to be placed approximately 20–30 kilometers apart. However, a CORS network can be used to develop a model of differential corrections, as shown at right, from which a rover receiver can interpret RTK correction information and use this during the computation of its position. The geometry of a CORS network allows two adjacent reference stations to be located up to 80–100 kilometers apart without degrading the accuracy, although in practice most systems tend to locate them closer together than this. This is essentially a reduction from 30 reference stations per 10,000km² for conventional RTK, to 5–10 reference stations for network RTK, delivering high-precision services to virtually unlimited users.

Figure 2. The improved navigation performance from RTK (left) to network RTK (right).

It is expected that the CORS networks will become a critical part of a country’s spatial infrastructure, and countries like the UK are leading the way. This makes network RTK one of the most promising positioning technologies for road vehicles and ITS applications.

As shown in previous research, network RTK can deliver a vehicle positioning accuracy of better than 5 centimeters, and in real-world tests this level of accuracy had an availability of 41–45 percent, depending on the environment. It was also found that the correction information was available via the GSM network for more than 80 percent of the time. In these same tests, the total time without any GNSS position solution (network RTK, DGNSS, or stand-alone) was up to 16 percent in a motorway environment. Network RTK was able to provide lane-level positioning accuracy, but the sensitivity of the technique to GNSS signal loss and coverage of the communication network had a significant effect on availability. GNSS outages could be caused simply by passing under a road bridge, and the network RTK solution would be lost, although there would continue to be a DGNSS solution for a short period. Finding effective solutions to these current barriers, which prevent wide adoption of network RTK, is a key enabling step for ITS.

Accuracy Assessment

In much more controlled tests to assess the accuracy of network RTK on a dynamic vehicle, the network RTK GNSS receiver was compared to an inertial navigation system (INS). This test was carried out using the NGI roof laboratory, which houses a 120-meter rail track running an electric locomotive.

Both the network RTK receiver and the INS used the same antenna, fed separately through a signal splitter. The network RTK solution was recorded in real time onto an SD card in NMEA GGA format. The INS data was recorded and post-processed in a tightly coupled solution using a continuously operating dual-frequency GNSS receiver base station located inside the rail track circuit. There were no recorded GNSS outages as there is a clear-sky view from the roof laboratory.

The antenna point was also tracked using a total station, recording observations at 10 Hz stamped with GPS time. Although the accuracy of the tracking mode of the total station is not high enough to assess the accuracy of the network RTK solution (because of time synchronization issues), it ensures that any gross errors in GNSS observations that could affect both the network RTK and INS solutions did not occur.

The results in Table 1 show that the network RTK solution consistently performs to a high accuracy, giving a low standard deviation from the mean in all directions. Listed are three laps of the rail circuit recorded at different times. There are a small number of epochs that encounter large differences of more than 200 millimeters, such as during laps 2 and 3, although these appear to be very short-term anomalies, possibly caused by dynamic GNSS signal multipath or delays and message loss in the communication system.

Table 1. Comparison of the tightly coupled (GPS+IMU) solution with the N-RTK solution.

The worst absolute accuracy is shown during lap 3, although even in this case, with a mean of 21 millimeters and 99 percent of the observations lying within 15 millimeters, this solution still delivers a solution within 36 millimeters of the ground truth. 50 percent of the network RTK observations are within 1 millimeter of the mean difference between the two solutions, showing remarkable consistency and precision.

Challenge: Comm Signal Strength

A fundamental aspect of network RTK is the delivery of reference station data used in the processing of the receiver’s position. Although there are various methods used to deliver this data, the most secure and reliable method involves transmitting raw reference station observations, so that the receiver may perform the calculation of the position with all possible data. This provides the highest integrity. The vulnerability here is not the algorithmic method used to transmit the data, but the communication system, in three ways:

• There is no connection between reference and rover receivers.
• There is data loss from the connection.
• There is an unacceptable delay in the transmission of the data.

Lack of Coverage. The preferable communication system is to use mobile Internet over the GSM/GPRS cell network, which is already well established. The major network operators claim over 99 percent coverage of the population in the UK, but this does not take into account physical and local conditions such as land and building obstructions, atmospheric conditions, and inter-ference from vegetation and other
radio signals.

A 2011 BBC survey in the UK found that when users had a cell-phone data connection it was 3G for 75 percent of the time (2G otherwise), but significant “notspots” include major rail and road networks. An ongoing study by OpenSignalMaps has found that a 3G service is only available 58 percent of the time. A 2011 government report detailed the extent of 2G and 3G services, shown in Figure 3. Areas with poor data communication coverage (below 50 percent) pose a significant problem for network RTK in vehicles.

Figure 3. 2G (left) and 3G (right) coverage by geographic area in the UK: green, >90 percent; yellow, 70–90 percent; blue, 50–70 percent; purple, 25–50 percent; red,

Data Loss. Continuity tests show that when using GSM/GPRS mobile communications to transfer the network  RTK corrections, the availability was approximately 88 percent, and the connection could be lost after a few hours of continuous use. This can be caused either by SIM cards that use dynamic IP addresses, creating interruptions when renewing the addresses, or where voice data was prioritized on the network. Other research has shown that a typical mobile Internet connection (a combination of wired public Internet and GPRS) suffers from approximately 20 percent data loss.

Message Delay. A network RTK receiver  imposes a transmission time limit on the correction messages that are used to fix the common integer ambiguity (in this case, the Leica GS10 limit is 10 seconds), although messages younger than 60 seconds can be used to give an accurate DGNSS solution. Messages older than 60 seconds result in the receiver only being able to output a standalone position, by which time the accuracy will decay beyond vehicle positioning requirements. Earlier research found the typical mobile Internet connection suffers from an average delay of 0.85 seconds.

Challenge: GNSS Outages

The majority of the transport infrastructure is outside and has a clear view of the sky, particularly away from heavily urbanised areas. However, the receiver gets no warning of impending signal obstruction, so that even momentary obstructions such as an overhead gantry on the motorway can cause significant loss of positioning accuracy, and often causes a receiver to output no solution at all, as shown in Figure 4. Here the vehicle is traveling in a northern direction in lane 1 of the left-hand carriageway and passes underneath a series of bridges at a motorway junction. This causes both GNSS outages and deteriorated positional accuracy, so much so that the vehicle is positioned in the southern carriageway (note that the underlying map image is of unknown accuracy).

Figure 4. The typical effect of overhead obstructions on vehicle GNSS positioning.

GNSS outages can occur in several ways: the obstruction of the GNSS signals can lead to a loss of signal lock; a momentary obstruction or partial obstruction can cause cycle slips to occur (during carrier-phase positioning); if the visible satellites at the rover receiver are not the same as at the reference receiver, then the ambiguity cannot be resolved; there may be intentional or unintentional signal jamming or interference; and if the receiver assessed the integrity or accuracy to be poor then it may not provide a solution.

NGI test vehicle.

Experiment Set-Up

The test vehicle was equipped with a GNSS receiver and antenna, receiving real-time corrections using a GSM/GPRS connection. The signal strength was measured simultaneously using the Android application RF Signal Tracker on an Android-based mobile phone.

The data recorded includes: GNSS raw data, RINEX format; network RTK real time output, NMEA format; GSM signal strength, CSV format. As the experiments were not intended for the analysis of the accuracy of the GNSS receiver, there was no need to utilize the ground truth system onboard the NGI test vehicle.

RF SIGNAL TRACKER. Android application and mobile phone used to record the GSM signal strength (left), and GNSS receiver (right).

Test Environment. Two test scenarios were chosen for the experiments. To assess the GNSS signal outages, the test vehicle was driven along the M1 motorway, a length of approximately 100 kilometers. The M1 is a major road transport artery linking London in the South to Leeds in the North of England, typically with three or four lanes in each direction. This route passes under 214 overhead obstructions (northbound and southbound directions), of known classification (gantry, footbridge, road bridge). This scenario was chosen as the environment is quite rigid, allowing repeatable tests, and it is the area in which future ITS technology is most likely to be adopted first.

To test the variation of GSM signal strength in real-world conditions, a small circuit was chosen close to the Nottingham Geospatial Institute (shown in Figure 5), which incorporates a variety of environments from open sky to bridge underpasses, and dense tree coverage. Using a repeatable path allows the identification of issues that are attributable to problems with the communications link as opposed to other issues (such as hardware problems and GNSS signal outages), and despite the short distance, the loop also provides a wide range of GSM signal strengths. During the experiments to follow, the data was measured during three consecutive laps of the circuit.

Experiment Results

GSM Signal Strength. The variation in color along the NGI test route is an indication of the RSSI (Received Signal Strength Indicator). In this area, the RSSI varies between –50 dBm and –105 dBm, which are the typical maximum and minimum strengths of a cellular network. This is despite the assessment from the network provider that this entire area delivers high-speed Internet and email. Figure 5 also shows the subjective rating and expected performance of the RSSI.

Figure 5. The GSM signal strength around the NGI circuit in Nottingham, with the subjective RSSI ratings.

Table 2. The spread of RSSI observations recorded during the trials around the NGI circuit.

Table 2 details the RSSI observations measured during the signal strength trials around the NGI circuit. The range of values shows the typical maximum and minimum RSSI values experienced by a cell-phone user (other than no signal being received). The signal strength is recorded every 5 meters, in order to achieve a good geographic spread across the area (as opposed to biasing the results with observations recorded whilst the vehicle is stationary). The RSSI observations do not correspond to a typical Gaussian distribution, suggesting that there are external influences on the strength of the signal and the handover between one cell tower and the next.

Figure 6 shows an increase in the age of correction (AoC) of the messages following a drop in signal strength (RSSI) to approximately –100 dBm. This is visible from the peaks in the age of correction message to over 8 seconds. The graph shows three laps of the NGI circuit, noticeable by the repeated pattern of signal strength. The increase in the AoC occurs at approximately the same geographic location on each lap ­— an area in the northwest of the circuit that suffers from weak signal strength, as seen in Figure 5. The received signal strength is the sum of the direct and indirect (or reflected) waves, varying with distance between a series of maximum and minimum values. On a moving vehicle, the RSSI will vary with time as it moves between these maximum and minimum values, and is especially complicated in urban areas where there may be no direct waves at all, and waves are propagated by a series of reflections. A moving receiver also suffers from a Doppler shift in the received signal’s frequency.

Figure 6. The effect of GSM RSSI on the age of correction messages.

During network RTK positioning, the receiver considers messages older than 10 seconds unusable for a fixed network  RTK solution, although messages younger than 60 seconds can be used to give an accurate DGNSS solution. This scenario has a brief occasion during the loop in which loss of the network RTK solution is attributable to weak GSM signal strength.

A close inspection of Figure 6 highlights a slight delay between the drop in RSSI to –100 dBm and the increase in the AoC. This delay needs further analysis, but is assumed to relate to the slower update rate of the ionospheric and tropospheric corrections (10 seconds and 60 seconds respectively). There are also periods of increased AoC that are uncorrelated with a drop in RSSI, for which there is no clear explanation, although none of these occasions results in a loss of the fixed ambiguity network RTK solution.

Eighty cell handovers were recorded during the trials, which is higher than average as this area is liable to carry a large volume of cellular traffic (there is a university, a large hospital, and major roads, as well as general housing and business properties). The cell handovers showed an average improvement of +1.2 dBm from just before the handover until just after. The maximum improvement is +22 dBM, although there are occasions when the RSSI gets worse, the biggest fall in received signal strength being –12 dBM. Figure 7 displays the frequency distribution of the change in RSSI during a cell handover. The resolution of the RSSI measurements is 2 dBm.

Figure 7. Frequency histogram of the RSSI change during a cell handover (2 dBm bins).

Cell handovers occur at a range of RSSI, not just low signal strength. This suggests that cell handovers are managed by the network operator in a way that does not disrupt the data connection. There appears to be no correlation between a cell handover and a problem with the correction message delivery.
Although this part of the experiment was not a test of receiver performance, during the NGI circuit trial 63.1 percent of the receiver observations were network RTK fixed, and 33.0 percent of the observations were DGNSS observations. Therefore, 3.9 percent of the possible epochs had no observations, partly due to passing under bridges. The largest GNSS outage during circuit trials was 4.85 seconds. These values show an improvement over previous research, particularly as this is considered a difficult GNSS positioning environment.

GNSS Outages. During the GNSS outages tests, the vehicle traveled at a constant speed of 60 mph, mostly in lane 1 of the motorway. Table 3 shows statistical breakdown of the GNSS outages and the resulting reacquisition of the fixed ambiguity in network RTK positioning.

Table 3. Statistical breakdown of GNSS outages caused by overhead objects.

The longest total GNSS outage caused by an overhead obstruction was 4.65 seconds, when passing under a road bridge. At 60 mph this translates into a distance of almost 130 meters without any GNSS solution, which is much further than the width of the overhead object. Once the GNSS signal is reacquired, there is a short period during which the fixed integer ambiguity is resolved, in order to achieve the centimeter-level accuracy. The longest duration between start of a GNSS outage and reacquisition of the fixed ambiguity for the network  RTK solution is 52.10 seconds, or 1,450 meters. Although during this period, a DGNSS solution is available as soon as the satellites are reacquired.

Discussion

Nationwide adoption of cellular Internet services by cell phone users has provided a useful communication system for positioning systems. But network providers do not guarantee the type of communication service demanded by advanced ITS and V2X applications. The quality of service is too easily disrupted by passing into an area with weak signal strength, or when many users congest the bandwidth.

Future generations of cell networks, such as 4G, will significantly increase the available bandwidth and increase download speeds, but there is an unknown increase in the demand on the system from non-critical cell-phone users. The issues in the existing system can be minimized slightly through improvements at the user end, such as using stronger gain antennae or accessing multiple networks with different SIM registrations. The nature of cell networks also leads to a decrease in signal strength occurring prior to the cell handover, which can cause delays in the message delivery, so the management of this process could be improved. Future testing of the GSM network can be carried out at the new innovITS ADVANCE test facility at MIRA in the UK, where the private network can be controlled and manipulated as desired.

An alternative communication method, that has the same wide area coverage of a cell network, is satellite communication. In tests, observation of static positions showed 98 percent of messages were received correctly at a latency of less than 10s. This compares with the High-Speed Download Packet Access (HSDPA) cell network figures of 99.8 percent and 1.2s. When in a kinematic mode, the satellite communications fared less well. Testing three separate satellite communication systems, problems were encountered with reacquisition, long latency, and static initialization. At best, 70 percent of correct messages were received, with a latency of 4.2s, although often over 20s.

Digital Audio Broadcasting (DAB) is capable of being used as a future communication method for network  RTK positioning. Compared to traditional VHF and UHF radio communication, it uses the frequency more efficiently and is more robust to degradation.

The design of the GNSS receiver used in testing is aimed at delivering a very reliable and highly accurate solution. It was not intended for use on vehicles and in dynamic environments. The receiver deals well with multipath, rejecting low-strength GNSS signals, allowing the resolution of the integer ambiguity. However, this means that in city environments it may provide fewer solutions than a modern smartphone, albeit with a much higher accuracy when it does. Recent research shows it is possible to increase the speed of ambiguity resolution, and customize integrity controls, making the resolution process close to instantaneous in certain circumstances.

Conclusions

As cellular communications networks evolve in the UK and other countries, the performance of the network  RTK receiver also improves. We found that once the RSSI drops to approximately –100dBm, the correction messages suffer from either message loss or message delay that causes the receiver to underperform. The performance of the communication link during a cell tower handover has shown that there is no deterioration in the performance linked to the handover, although cell tower handovers generally occur at the limits of a cell tower’s coverage, and hence at low signal strengths.

The resolution of the fixed integer ambiguity is crucial for the high-accuracy solution available with a network RTK receiver. The resolution is relatively fast, typically within two minutes from a cold start, or fewer than 20 seconds from a hot start. During tests on the M1 motorway, passing under an overhead obstruction caused a maximum total GNSS outage of 4.65 seconds, and a maximum time until the ambiguity was resolved of 52.10 seconds. On average, the GNSS outage was 1.14 seconds with an average re-fix time of 13.13 seconds. Until the ambiguity is resolved, the receiver can continue with a DGNSS solution delivering lane-level accuracy.

Manufacturers

NGI’s inertial nav system is an Applanix POS/RS, which consists of a NovAtel OEM4 dual-frequency GPS receiver combined with a navigation-grade Honeywell consumer IMU. The network RTK position was provided by a Leica GS10 receiver and Leica SmartNet correction service over the Vodafone network. Both receivers used a Leica AS10 antenna.

Scott Stephenson is a Ph.D. student at the Nottingham Geospatial Institute within the University of Nottingham.

Xiaolin Meng is associate professor, theme leader for positioning and navigation technologies, and MSc course director for GNSST and PNT at the Nottingham Geospatial Institute of the University of Nottingham. 

Terry Moore is director of the Nottingham Geospatial Institute (NGI) at the University of Nottingham, where he is the professor of satellite navigation and also an associate dean within the Faculty of Engineering.

Anthony Baxendale is head of Advanced Technologies & Research at MIRA Ltd.

Tim Edwards is the lead engineer of the Intelligent Transportation Systems (ITS) research group at MIRA Ltd