Centimeter-Level Positioning: Options for Outdoor Testing

Summary

Ten years ago low-cost but high-precision GNSS options did not exist. Any cm-level positioning system would cost $20k or greater. During this time it made sense for some engineering organizations to acquiring and maintain custom in-house GNSS systems for testing their intelligent systems. But this has changed and the trend will continue. Outlined in table below are 3 options for satisfying the reference navigation system requirement using GNSS receivers. The remainder of this post details the options and associated findings.

Reference System Specifications

A navigation truth or reference is essential to the development and testing of intelligent systems. This we discussed in our related post: Truth System - What Really Happened?. Listing the specifications required or desired assists in selecting a suitable system. A listing of commonly useful reference system specifications are given in Table I as are six categories of reference system solutions.

Each system is rated according to the utility of the system in meeting the specifications. A key observation from Table I is that the primary advantage of a custom in-house GNSS system is the flexibility to customize. However, if budget and time are not available to customize and maintain the system, this becomes a disadvantage.

Testing Expenses

Consider a data collection and analysis campaign to collect data from a device under test (DUT) along 2 test paths and analyze heading accuracy. The estimated effort required to fulfill this campaign is captured in Table II for four categories of reference systems. The key observations from this table are:

1. Maintenance: The choice of reference system has implications on the amount of time required for yearly maintenance (e.g. firmware updates, software updates, calibrating device) as well as reference system testing (static tests, dynamic tests).
2. Acquisition: The presence of low-cost commercial systems and/or rent options makes it difficult to justify the cost of acquiring and maintaining a custom in-house GNSS system solely for testing purposes. Ten years ago such low-cost options did not exist. Any cm-level positioning system was easily $20k. But this has changed.
3. Post-Processing: Indirect measurements requiring additional post-processing takes time and can be a distraction from the core testing objective. For example, extracting body-axis heading when only measuring position and velocity-based heading is subject to error and introduces uncertainty.

Next we go into detail on each of the three options for outdoor testing outlined in the start of this article.

I - Low-Cost RTK Solutions

The performance of low-cost RTK receivers can be quantified on metrics like accuracy, continuity, availability, and time to first fix (TTFF). The performance with respect to these metrics is environment dependent and therefore manufacturer information is scarce. The information that is shared does not facilitate decision making or planning. This is part of the price paid when dealing with low-cost sensors.

A set of candidate low-cost RTK commercial solutions are outlined in Table III. The manufacturer descriptions do not shed light on the performance of these devices. This means we must rely on our own testing experience or the evaluation and reviews of others. We draw on a 2018 technical report summarizing an evaluation of low-cost RTK GNSS receivers done for the Minnesota Department of Transportation and authored by researchers at the University of Minnesota [1]. From the candidate list presented in Table IV, the Swift Piksi Multi and the u-Box NEO-MP8 were included in the research evaluation. An extract from the concluding remarks in Chapter 5 clearly conveys the trend.

Extract from Concluding Remarks [1]:

“The results of this evaluation showed that the low-cost receivers, in general, do not perform at a level of existing high-end receivers. [...] [T]he following observations can be made about all of the L1-only, low-cost receivers evaluated in this work:
1. They can achieve centimeter-level accuracy in static applications [and] in rural environments.
2. They perform better when using a high-quality antenna versus a low-quality antenna.
3. They cannot hold an RTK fixed-integer solution for any significant time in dynamic applications.
4. They spend most of their time in an RTK floating-point solution.
The multifrequency Swift Piksi performed better than the L1-only, low-cost receivers but displayed the same shortcomings (degraded accuracy, worse availability) in the suburban and urban environments during static testing.”

Tables 5.1 and 5.2 from the same report summarize the static and dynamic performance measured in rural (i.e. ideal) environments from the devices under test. The Piksi Multi was a strong performer achieving 2.4 cm accuracy 1 with 95.5% fixed-integer availability under static conditions. However, even this top performer had significant degradation in performance under dynamic conditions. The accuracy degraded by a factor of 2 as did the fixed-integer availability. Loss of lock occurred an average of 2.53 times per minute.

The u-blox ZED-F9P is a newer product and looks like a strong multi-frequency candidate. Customer reviews from the electronics retailer SparkFun indicate a positive RTK user experience. One such review is included below. While RTK is inherently brittle, the multiple frequencies and continuous product improvement trend appears to be slowly overcoming the performance limitations observed in the 2018 evaluation. We can expect this trend to continue.

Review of ZED-F9P purchased from SparkFun[2]:

“I’ve been running Sparkfun GPS boards for years for both personal and government projects. Analysis continues to get easier. When the previous GPS-RTK board came out I jumped for it... antenna on roof... local CORS station for a reference... RTKLIB post-processing. By averaging weeks of data I got to a sub-cm solution for my roof antenna. Still... it was a differential measurement with a 6 km baseline. Now, however, with the F9P and the NTRIP stream of SSR corrections I see a real time solution of roughly 2-3 cm. Your Hookup Guide was a great help to getting it up and running quickly. Now its time to dive back into RTKLIB (demo5 from rtkexplorer.com) to reach a PPP absolute position in post-processing and see what happens.”

II - Low-Cost GNSS-Aided INS Solutions

Commercial GNSS-aided INS solutions are becoming increasingly capable and popular. The advantages over a standalone GNSS receiver include:

• Higher data rates (e.g. 100 or 200 Hz)
• Senses position, velocity, and orientation
• Includes inertial sensors useful for control and other autonomy applications
• Adds robustness to brief loss of GNSS signals

Products in this category include the VectorNav VN-200 and the CTi Sensors CS-GN300. The former has many years of maturity. The latter is a newer entrant which means it can take advantage of the latest underlying sensor technology. Generally in this category support for RTK or cm-level GNSS aiding will depend on the product configuration.

III - Rent Reference System

Renting a navigation reference system involves giving up ownership of the device but has the advantage of transferring the burden of both acquisition and on-going maintenance to the supplier. Dropping the cost of acquisition means higher-grade sensors can be rented for testing. Removing the burden of maintenance frees resources to focus on the project at hand. The ratings in Table I showed the rental option had strong performance in nearly all specifications considered.

References

1. J. Jackson, R. Saborio, S. A. Ghazanfar, D. Gebre-Egziabher, and B. Davis, “Evaluation of Low-Cost, Centimeter-Level Accuracy OEM GNSS Receivers,”
   Minnesota Department of Transportation, Tech. Rep., February 2018. [Online]. Available: http://www.dot.state.mn.us/research/reports/2018/201810.pdf
2. “Member 894355”, Customer Reviews - SparkFun GPS-RTK2 Board - ZED-F9P (Qwiic), 2020 (accessed September 16, 2020). [Online]. Available:
   https://www.sparkfun.com/products/15136#reviews