NovAtel Wingsuit Jump
In 2011, a unique set of conditions provided NovAtel with the opportunity to test our new OEM615 GNSS receiver under exciting, high-dynamic conditions. As NovAtel engineers were finalizing the development of the OEM615 - the smallest dual-constellation/dual-frequency survey-grade receiver ever developed by NovAtel - NovAtel Applications Engineer Andrew Levson was embarking on a passion of his own - wingsuit skydiving. With designs on defining and setting new Canadian records for both distance and formation flying, Andrew thought about ways to train for and confirm these potential records. Naturally, he gravitated towards the technical solution he was most familiar with: GNSS!
After months of preparation and planning, and backed by a dedicated NovAtel support team, Andrew took to the skies on August 2, 2011 for a day-long series of wingsuit jumps. The goal: to verify that NovAtel's OEM615 receiver and ALIGN® heading technology provides extremely accurate position, heading and velocity measurements in the most demanding of environments.
We are happy to share the details of our wingsuit project with you.
Attitude Determination and ALIGN™
GNSS technology is used in various ways to find attitude or trajectory. The simplest method relies on measuring the velocity of a single receiver and interpreting the direction of that vector as the vehicle's heading. This works for applications where a vehicle's motion is constrained to only one axis - either absolutely, as with a train, or in the typical case of a car - when being driven responsibly! In these cases, attitude is very closely linked to trajectory (in two dimensions, anyhow). This technique is not sufficient for measuring attitude under more complex dynamics, or where velocities are so low that measurement uncertainty masks the signal of interest. Consequently, a practical attitude-determination system requires additional sensors.
SPAN GNSS/INS Solutions
GNSS receivers can be tightly integrated with inertial measurement units (IMUs) to form a combined GNSS/INS (GNSS/inertial navigation system). In these systems, the IMU provides rapid angular rotation and acceleration measurements, complementing the absolute accuracy of the GNSS positions. A GNSS/INS implementation can take advantage of the strengths of each type of system to provide accurate, continuous solutions for applications involving high dynamics, frequent GNSS signal outages, and other operational and environmental factors. Certain aspects of GNSS/INS can prove challenging for some applications, however. One particular challenge is that the orientation of the IMU with respect to the GNSS antenna must be accurately known and typically must remain fixed. This is appropriate for an installation on a rigid body, but proves challenging when considering a body with independently moving parts (such as a skydiver).
Another solution suitable for applications with good sky visibility involves the use of multiple GNSS receivers with antennas at different locations on the vehicle or body. Geometric analysis of the vectors between the independent points can yield attitude information. With two receivers, two attitude dimensions can be measured (e.g., pitch and yaw), while a third receiver adds full three-dimensional attitude determination. Accuracy of a GNSS-only attitude system depends on the geometry of the antenna array, measurement accuracy, and the baseline computation technique. If we take it as a given that antennas are oriented and spaced to provide good geometry, the performance comes down to the accuracy of the baseline computation technique. The simplest baseline computation is done by differencing the positions reported by two antennas at the same time, and computing the vector between them. The resultant angle and baseline accuracy depend partly on the inter-antenna spacing (angular estimates will improve for longer baselines) and partly on the solution accuracies themselves. A simplistic approximation of heading accuracy can be computed by just converting position error into angular error. Over a long baseline of, say, 300 meters (a typical tanker or cruise ship length), single-point RMS accuracies of 1.2 meters could provide heading accuracies on the order of
The foregoing equation is not really accurate because it accounts for full two-dimensional error, while only the error perpendicular to the baseline is of concern; it does, however, provide a good “order of magnitude” estimate. Using the same approximation on a shorter baseline of 10 meters - a small aircraft's wingspan, for example - that same heading error increases to nearly 14 degrees. The problem is magnified on even shorter baselines until, at some point, the vector is meaningless. Clearly, further refinement of the individual solution accuracies is needed for short baselines. If receivers can be operated in a dual-frequency real-time kinematic (RTK mode), individual solution accuracy can be improved to the range of one centimeter or better, assuming a relatively short distance to a reference base station. With such accuracy, we can now theoretically achieve a sub-degree heading accuracy over a baseline as short as one meter. In fact, even as distance from a fixed base station increases, this heading accuracy would not degrade significantly on short-baseline installations because errors would remain highly correlated on the installed antennas. At very long distances from base stations, though, integer ambiguity resolution becomes less reliable and increases the probability of a large blunder affecting position accuracy. Although the accuracy from a dual-RTK setup is attractive for many applications, there is a significant drawback to using RTK in such a system: it requires a fixed base station installation somewhere, and for this base station to be in communication with all receivers installed in the attitude-determination system. Furthermore, each receiver must be in communication with a central processing system that reduces raw positions to attitude vectors. These communication links could be either complex to install or completely unavailable.
Multi-GNSS Attitude with ALIGN
There is a way to achieve the RTK-level of heading accuracy without a base station. NovAtel has developed an RTK-based “moving base” solution that greatly simplifies the task of determining heading and baseline from multiple GNSS receivers.
Much as in the RTK method, a “Master” receiver with this proprietary design periodically sends correction data to one or more similarly configured Rover receivers, along with its own position at the same time. The Rovers then compute their own position relative to the Master using, ideally, fixed-integer RTK techniques. (Less accurate “float” solutions are also possible under poor signal conditions.) In effect, the equipment adapts the traditional RTK mode to function with a different base position at every epoch.
Benefits of ALIGN for This Event
As compared with the dual-RTK method described earlier, NovAtel's proprietary design has some clear advantages. First, the base infrastructure (i.e., fixed base station) and communications requirements are markedly reduced. Without a requirement for absolute accuracy to be at the centimeter level (needing only the relative accuracy for attitude determination), receivers can operate in a single-point mode with no need for ground-based differential corrections. Additionally, the NovAtel system's computations are done on-board the Rover receiver with native firmware features, eliminating the role of a central processor to accomplish that same task. The second advantage to this moving-base solution is less obvious: an inherent increase in heading/baseline accuracy occurs when using it as opposed to a dual RTK solution. In a two-receiver setup, the dual-RTK method involves two independent baselines (Base - Rx1, Base - Rx2) with an associated inaccuracy for each baseline. Differencing these two positions could double the inaccuracy in some circumstances. Conversely, operating those same two receivers in the moving-base mode computes only a single baseline (Moving Master-Rover), and usually that will be a shorter baseline than in an RTK setup (Fixed Base-Rover).
In support of Andrew's training objectives, and to learn more about the performance of NovAtel's products in an unusual environment, we began investigating what equipment we could realistically deploy on a human body in a wingsuit scenario. Given the obvious restrictions of the tight-fitting wingsuit, we concluded that the OEM615's smaller size and power consumption were ideal for this man-mounted application - while still allowing the use of dual-frequency GPS+GLONASS ALIGN at 20Hz with up to 50 Hz raw measurement logging.
- OEM615 receiver itself is slightly less than 36 cm3 and only 24 grams
- When loaded with ALIGN firmware, generates high-precision, real-time heading and pitch angles between two receivers.
- Power was provided by a 1.3 A-h, 12 V battery
- Two compact (69 mm diameter; 22 mm height; 162 grams) L1/L2 GPS/GLONASS active antennas, the G5Ant-2AMNS1 from NovAtel's subsidiary, Antcom Corporation
- For data collection we employed an ASUS Eee PC from ASUSTeK Computer Inc., with data logged via USB in a console-based script.
To support post-processing analysis we also set up a SMART-MR10™ receiver in the drop zone and collected static carrier phase and pseudorange data for the entire day of test (approx 12 hours). While it is not typically meant to serve as a precise base station, the SMART-MR10 is equipped with the same Pinwheel® Antenna technology as NovAtel's survey-grade GPS-700 series of antennas, and includes an integral OEMV-3G receiver that is commonly used in surveying equipment. The integrated package of the SMART-MR10 made for a simpler setup and still provided excellent performance.
A Cessna 206 “jump plane” and pilot were arranged for the entire day of August 2, 2011, at a drop zone near Innisfail, Alberta (52°04'40” N, 114°01'30” W). Setup began around 8:00 AM, with base station setup and a final check of the aircraft and skydiving equipment. Andrew and his wingsuit crew rehearsed their techniques for executing a coordinated exit from the plane, and discussed specific flight plans for the day.
The two OEM615 receivers were strapped (and taped) to Andrew's legs near his feet, with the battery pack similarly strapped to one leg. A small pocket of space within the wingsuit, just above Andrew's feet, provided barely enough space for all of this equipment. The data collection PC was carried in a small backpack, worn in front on Andrew's chest.
With one dual frequency GPS+GLONASS antenna mounted on each foot, Andrew's roll and yaw (heading) could be measured independently, (from the ALIGN HEADING log's pitch and heading outputs respectively).
In total, we conducted seven test jumps on the event day, with an average interval of about 90 minutes between jumps during which we re-packed the parachute, verified the data integrity, and conducted a rudimentary data analysis to decide if any changes were required to optimize jump performance.
Post-Processing and Analysis
With high-quality data sets obtained in the foot-to-foot configuration, we set about on a post-processing mission to extract more information about our performance in the free-fall environment. This was a multi-step process involving several NovAtel utilities and techniques, as described below.
Step 1: Video Time-Synchronization
Each jump was videotaped by an in-air videographer. In order to correlate effects observed in the data with real-world dynamics seen on the video (including such important events as time of the jump itself), we created a very simple synchronization method. After the videographer had begun filming during the airplane's flight up, he shot a short sequence of our wingsuit flyer, Andrew Levson, hitting a key on the collection PC to create a time mark in the GPS data itself. At this point, the data PC output an obvious visual cue (a color change), which was then correlated with the GPS time mark in post-mission analysis. For the rest of the jump, the video was run continuously with events measured from the synchronization mark. Using this method, we fairly easily correlated several events in the video with obvious changes in dynamics.
Step 2: Base Station Coordinates
Our static base station at the Innisfail Airport collected data for approximately 12 hours continuously during the event day. To use it as a reference station for the individual jumps, we needed to determine its coordinates as precisely as possible. We used GrafNav in PPP (precise point positioning) mode, accessing downloaded precise satellite ephemeris and clock information to estimate a position. The long (12-hour) data set, combined with precise post-mission information, yielded a base station estimate accurate to approximately two centimeters or better, based on the solution standard deviation. Given that this is roughly two orders of magnitude better than typical single-point accuracy, we deemed it accurate enough to serve as a reference position for further post-processing.
Step 3: Aerial “Truth” Trajectories
In our wingsuit application, we lack a defined “truth” to compare against. To get our best possible estimate, though, we processed the raw range data collected by each airborne receiver (at 20 hertz) in differential mode with GrafNav. This software resolves carrier-phase ambiguities in much the same way as a real-time kinematic (RTK) application, but it processes the data both forwards and backwards to improve ambiguity resolution in the case of outages. Using GrafNav, we were able to obtain a roughly centimeter-level of accuracy for each receiver for the majority of each flight (except as limited by data outages). Obviously the raw range data used to create the GrafNav “truth” is the same as that used to generate the real-time single-point solution; so, the two trajectories are fundamentally correlated. However, the addition of fixed base station data to the GrafNav solution allows us to detect and reject any major blunders in range data. With this, we are confident that those GrafNav solutions identified as having fixed-integer ambiguities are, in fact, correct to within about two centimeters. However, the uncertainty in the base station position (± two centimeters) is still a factor.
As with the single-point position, we also lack truth data for receiver-to-receiver heading and pitch. We do have two methods to obtain semi-independent estimates of “truth” heading/pitch, though. The first is to use a difference of postprocessed GrafNav solutions (discussed previously). In concept, this would be similar to computing heading between two receivers operating in RTK mode, with the associated accuracy. The second method involves using GrafMov software to compute a moving-baseline solution between the two airborne receivers, in much the same way that ALIGN itself works. The added redundancy of having both forward and backward processing with an independent engine gives added confidence that we are likely to identify any major blunders. In addition to the GrafMov/GrafNav ALIGN calculations, we have one further constraint on our data that serves as an excellent sanity check of the real-time ALIGN solution. With the antennas mounted on Andrew's heels, the baseline length between them was fixed during free-fall to a maximum of roughly 90 centimeters by his wingsuit. Given this knowledge, we can very easily identify any major blunders in the ALIGN solution when the baseline length is reported to be longer than this. So, even though we are operating without a true “fixed” baseline, we still have an absolute limit based on physical parameters that provides us with more opportunity to check our performance.
ALIGN real-time measurements versus GrafMov and GrafNav Post-processed results. This figure compares the jumper's real-time heading against the results from the two different post-processed methods. GrafMov heading (green data points) was established by processing forwards and backwards through the data, with an independent processing engine from that used by the real-time ALIGN firmware. GrafNav (blue data points) shows two RTK trajectories calculated by processing forwards and backwards through the data, with the heading then computed from RTK position (right heel) to RTK position (left heel). This method produces statistically less accurate results than computing a relative heading.
We continued the work back in Calgary with extended post-processing and analysis from the event day data collection. We were able to extract trajectory and velocity for each antenna as well as heading, pitch and baseline solution from the antenna pair. We compared the real-time solutions with forward and backward post-processed GrafNav and GrafMov solutions (these are NovAtel products from our Waypoint Products Group), and examined the following outputs from the NovAtel system:
From each antenna:
- Position (Latitude, Longitude, and Height)
- Velocity (Horizontal 2D, Vertical, and 3D)
The 3D model shows real-time record of the ALIGN Master receiver's trajectory. The average vertical descent during glide stage after jump is approximately 83.8 miles per hour before the parachute deployment. During chute deployment, there was less than two seconds of satellite signal loss tracking due to antennas pointing to ground as Andrew's orientation changes.
Wingsuit ALIGN Master receiver velocities versus time (figure below - top left): This figure shows the 2D, horizontal and vertical velocities recorded from the ALIGN Master receiver. After chute deployment, the average vertical descent is approximately 18.6 miles per hour.
From the antenna pair:
- Baseline length
Wingsuit real-time ALIGN heading measurements versus GrafMov post-processed heading measurements (figure below - top right): This figure shows strong agreement with the ALIGN Master 3D model. Data has been filtered to show ALIGN in fixed ambiguity mode. ALIGN fixed ambiguity data starts approximately 15 seconds after jump and with 11 seconds outage at chute deployment. Note that the ALIGN trajectory is offset at right angles to the jumper's heading as a result of the wingsuit setup and geometry.
Wingsuit real-time ALIGN roll measurements versus GrafMov Post-processed roll measurements (figure below - bottom left): This figure shows strong agreement between ALIGN and GrafMov. There is significant roll at time 183-187 to allow photograph of parachute canopy.
Wingsuit real-time ALIGN baseline measurements versus GrafMov Post-processed baseline measurements (figure below - bottom right): The distance between Andrew's heels is not constant, varying from 35cm to 85cm.
These analyses show a close correspondence between the results of the real-time and post-processed data. For more detailed discussion of these results, see the ANALYSIS tab.
As an OEM supplier of GNSS precise positioning technology, our goal is to provide our customers with the most reliable, robust, high-performance solutions possible. While it is not our role to develop end-user applications, we are excited about the possibilities the technologies utilized in our wingsuit project present. A few that come to mind are detailed below:
Sporting applications such as wingsuit competitions:
- Large group formation flying: World records depend on the accurate positioning of divers with a formation pattern. Currently, the quality of group jumps is measured by taking a photo of the formation and overlaying a grid on the picture (as seen in the side bar). Divers have to be within a certain distance of the borders of the diamond-shaped cells of the grid. As the formation gets bigger, it gets harder to keep everyone inside designated grid locations. Real-time GNSS with audio alerts could help improve the execution of the divers to be properly positioned.
- Longest distance flown: GNSS can accurately track where the diver exits the plane and where the parachute deploys to properly calculate the distance traveled in free fall (canopy descent distance should not be included).
Safety of life applications such as search and rescue or firefighting:
GNSS technology can greatly enhance the flexibility and efficiency of conducting critical safety of life activities. With the first person out of a plane guiding others down to a landing site determined by the initial touchdown point, peer-to-peer relative vectoring eliminates the need to send data back to a base station or require team members to try to hit a preplanned location. This has the ability to save money, time and lives.
Unmanned aerial vehicles (UAVs):
While many UAV integrators currently utilize NovAtel GNSS receivers for precise positioning and attitude measurements, our ALIGN heading solution provides further opportunity to manage swarms of UAVs by offering the ability to develop systems for mid-air de-confliction and collision avoidance.
Aerial Delivery System:
Imagine the efficiency of self-assembling packages - jeeps, food, and ammunition, for example - that land in appropriate rows on the ground in relation to each other. In such a system, air crews do not have to sort the packages out on the plane, as the cargo containers, outfitted with a relative vectoring system, would do this in space as the parachutes float down.