Skimming the article quickly, I couldn't understand where the "ground truth" comes from in their algorithm. You have 10-15 phones within 10 km of each other, and each one pings the GPS satellite to get it's approximate position, sure, but then how do you go from there to the corrected locations?
That doesn't answer the question. Receivers that are typically useful for a local area augmentation system are fixed down, with an accurately-known location. That known location is then the ground truth, and the GPS signals are measured against that and the difference is broadcast to other mobile receivers nearby that use that information to become more accurate. In this scheme, it is unclear how this can be replaced by multiple mobile phones that don't have an accurately-known location.
* Receivers in the metal-reinforced concrete canyons of a big city suffer from multi-path reflections of buildings, leading to bigger uncertainty. You can improve those signals with a (still pretty uncertain) rural reference. This reference typically also sees low angle satellites, and it receives all signals without reflections.
* You can improve the uncertainty of some of your references by running analysis on it's position history and compare it with (highly certain) map data. One of your rural signals is moving at 78 mph? High likelihood to be in the left lane on that highway. Moved with 30mph, now suddenly stopped? Waiting at that specific traffic light, probably first car.
With crosschecks between enough references, some averaging and smoothing, you can produce a usable reference. Add a couple of absolute references (no reason for the company to not have a fixed antenna in the 50 largest cities each), and you're competing with current commercial reference services.
It's also worth noting, that for many of the uncertainties of the GPS signal (satellite clock jitter, atmospheric conditions), you don't need to update your position delta every second. It's already an improvement if you get a good reference update every minute or so.
If you have a sufficiently large deployment, there's a pretty good chance that at any point in time some of the phones will be stationary. Filtering out the fast-moving ones is pretty easy, leaving the stationary and slow-moving ones.
The whole reason technologies like DGPS work is because receivers in roughly the same location will have roughly the same offset. If you average out the stationary receivers, you will be left with a shared common drift - which can be corrected for. Slow-moving receivers probably don't affect the average too much, and you'd be able to filter them out over time as they'd have movement beyond the shared common drift.
Devices will enter the stationary set as they stop moving, and move out of it as they start moving again. Even someone stopped at a traffic light for a few dozen seconds might be able to contribute to it.
In the academic literature it is called "cooperative positioning" although we use different computational techniques than traditionally tapped by academics. Under the hood we are using techniques from RTK and PPP but in the form of a multi-receiver fueled ensemble optimization to determine the error correction. For testing we use a traditional RTK survey rig to test the results of our error corrections on phones as ground truth. Otherwise it is an exercise in managing and removing error to get as close as possible to ground truth.
> GPS accuracy on cell phones just isn’t good enough.
Is this true? Not good enough for what? They say further down that gaming and rideshare companies are showing interest. So is that their customers? Embedding their technology into apps to better stalk 3rd party humans in the real world?
This seems to an outsider like an insignificant problem that only applies to the least scrupulous companies, but maybe I don't understand.
It really isn't. If you go to wilderness, away from phone signals and wifi networks, which are already used to improve accuracy and speed in the cities, GPS just isn't enough.
But then again, gaming and rideshare companies only care about users in cities, so I'm not sure what their interests are.
How is it not enough? Lock times are worse, sure, without aGPS but unless you're in the middle of a dense forest GPS should have no issue locking pretty much anywhere.
What the hell are you talking about? GPS works just fine in 0 connection situations. I used it just fine in airplane mode and it got a fix within 10 meters.
Japan launched their own satellites that have a high angle in the sky because of all the urban canyons that they have in major cities:
> Instead, their ground traces are asymmetrical figure-8 patterns (analemmas), designed to ensure that one is almost directly overhead (elevation 60° or more) over Japan at all times.
This is also partly why more GPS signals (L2C, L5) were added: multiple signals at different frequencies from the same satellite allow for detecting atmospheric and multi-path issues:
It depends on your use case if current mobile pne GPS is good enough. Rideshare companies lose millions of dollars because drivers end up on the wrong block, wrong side of street, wrong level of an airport and have to circle. Automotive companies very much want to use mobile phone GPS to enable collision avoidance with pedestrians and bikes but there are way too many false positives. Fitness tracking can be wildly inaccurate for tracking accurate paces and fastest known time (leaderboards). On our end we don't ever store or compute a user's location just the error correction. The user's device determines their location and they decide who they want to share that location with. There is definitely problems with unscrupulous companies aggregating data and that is the antithesis of what we've worked to architect. There are tons of wonderful use cases that don't require stalking people. Making drone delivery more resilient to outages. Better search and rescue capabilities. Under the hood there is a powerful relative distance metric that gives us an even more precise distance between devices. Which opens the door the interesting location based gaming mechanics - like playing a game of digital tag. Hope that helps provide some context and our philosophy on privacy.
Maybe I'm biased living in a farming community, but is there anywhere that you are likely to find many cell phones available to provide correction data where you won't also find fixed base stations?
A single fixed base station normally expects ~1cm accuracy on the roving receiver. There is nothing in the article to suggest that they are going for even greater accuracy or precision, merely that they want to avoid fixed base stations in favour of many roving stations.
That'd have been great 20 years ago, but now that the base stations are already there, what have you gained? Maybe some limited utility in remote places that normally don't see human activity that somehow have many people with cell phones show up, while needing high accuracy GPS and forgetting their tripod-mount base station at home – but beyond that?
I think the novel part of the team's work is mobile phones can't directly use base stations for error corrections. We don't want to use mobile phones to replace base stations for the traditional surveying use case but instead use GNSS measurements from multiple mobiles to improve their own positioning. The same way having multiple GNSS constellations improves the accuracy of your phone because you have more satellite to receive measurements from. If you multiply that again by having multiple phones getting measurements you have more data to solve for location with. One user's phone's may be obstructed while another had great clear views you can use those additional views to improve the positioning for the obstructed phone.
Is that the case? The article suggests that the correction data obtained from the phones is like the correction data obtained from a fixed station. Presumably multiple phones are needed to provide correction data only because those phones are also moving themselves, using many sources to also correct for that movement.
The novel part seems to be that they think they can build a business selling correction data back to the customers who provided it to them, not having to rely on third-party fixed base stations or installing their own.
That is accurate - the error correction comes from aggregating measurements from multiple mobile devices instead of using a base station. The goal is to provide the service to mobile app developers who use location based services. They already have relationships with their users and our goals is simply to provide them a better performing location service.
There are academic papers on using multiple smartphones to synthesize centimeter accuracy based on smartphone typical meter-scale accuracy. Apparently the obvious approach, which is calculating a dual frequency-based position, which is what survey GPS does, is not available from smartphone GPS, even though Android APIs give you access to signal phase data. I suppose if you had more than one phone you could do better. But "crowdsourcing" seems like a stretch. Why? What's the TAM? What's the business model? What are the use cases? The article mentions anonymization, but that just points out the privacy concerns about participating in a thing that depends on you sharing location data, however obfuscated, with strangers.
I could see Qualcomm buying the patents, if any, for beamforming 5GSA optimization.
Regular GPS claims ~1m accuracy, but in urban areas it's really more like ~10m, and it gets confused about which side of the street a pedestrian is on.
If you have a dozen people trying to navigate the same neighborhood of Boston* on foot, they could all be using the same app, all sharing data with each other and all benefiting from that data.
Does phase data require holding the receiver still? Anytime I've seen a survey GPS receiver, the surveyors seem to leave it in the same spot for a few minutes.
*There's some loss of generality here. In a grid layout place like New York this would be unnecessary, but in a less dense place it would be unviable.
I use hardware that collects RTK corrected GPS coordinates for pole line engineering, and even with RTK you can get some pretty wild errors. One building in a small town with a metal roof and metal siding caused the reported location of the adjacent pole to be more than 10 meters off. I ended up manually correcting the coordinates via the satellite view on the map.
I've seen cases where inaccurate positioning of poles by just 60 centimeters is enough to make pole loading analysis fail a pole.
I'd imagine there is at least something to be gained from calculating position/velocity patterns of receivers (stationary on a desk, moving on a road in a car, etc) and then cleaning up data based on that.
Or perhaps the goal is to use data from dual freq GPS receivers to helping out the continuing majority of single frequency receivers?
Or perhaps they are just leaning on fixed base stations in urban areas to "bootstrap" and focusing on the P2P aspect as marketing, plus a convenient permissions grab for backhauling personal location information.
There is more detail on the testing results and approach in this Medium post. https://medium.com/@zephr.xyz/gps-the-technology-we-all-use-.... Happy to answer questions about the team's work and thanks for interest in it all. Always hard to go into deep technical detail in a media piece.
Interviewed with this team a while back, so it's cool to see them come out of stealth. I was super impressed with their technical chops; would definitely recommend reaching out to anyone who's considering applying.
30 comments
[ 0.23 ms ] story [ 77.8 ms ] thread* Receivers in the metal-reinforced concrete canyons of a big city suffer from multi-path reflections of buildings, leading to bigger uncertainty. You can improve those signals with a (still pretty uncertain) rural reference. This reference typically also sees low angle satellites, and it receives all signals without reflections.
* You can improve the uncertainty of some of your references by running analysis on it's position history and compare it with (highly certain) map data. One of your rural signals is moving at 78 mph? High likelihood to be in the left lane on that highway. Moved with 30mph, now suddenly stopped? Waiting at that specific traffic light, probably first car.
With crosschecks between enough references, some averaging and smoothing, you can produce a usable reference. Add a couple of absolute references (no reason for the company to not have a fixed antenna in the 50 largest cities each), and you're competing with current commercial reference services.
It's also worth noting, that for many of the uncertainties of the GPS signal (satellite clock jitter, atmospheric conditions), you don't need to update your position delta every second. It's already an improvement if you get a good reference update every minute or so.
The whole reason technologies like DGPS work is because receivers in roughly the same location will have roughly the same offset. If you average out the stationary receivers, you will be left with a shared common drift - which can be corrected for. Slow-moving receivers probably don't affect the average too much, and you'd be able to filter them out over time as they'd have movement beyond the shared common drift.
Devices will enter the stationary set as they stop moving, and move out of it as they start moving again. Even someone stopped at a traffic light for a few dozen seconds might be able to contribute to it.
Is this true? Not good enough for what? They say further down that gaming and rideshare companies are showing interest. So is that their customers? Embedding their technology into apps to better stalk 3rd party humans in the real world?
This seems to an outsider like an insignificant problem that only applies to the least scrupulous companies, but maybe I don't understand.
But then again, gaming and rideshare companies only care about users in cities, so I'm not sure what their interests are.
> Instead, their ground traces are asymmetrical figure-8 patterns (analemmas), designed to ensure that one is almost directly overhead (elevation 60° or more) over Japan at all times.
* https://en.wikipedia.org/wiki/Quasi-Zenith_Satellite_System
This is also partly why more GPS signals (L2C, L5) were added: multiple signals at different frequencies from the same satellite allow for detecting atmospheric and multi-path issues:
* https://en.wikipedia.org/wiki/GPS_Block_III#New_navigation_s...
* https://en.wikipedia.org/wiki/GPS_signals#Modernization_and_...
Maybe like a rake receiver / antenna diversity, they can compound the multiple measurements from GPS satellites to get improved location information.
That'd have been great 20 years ago, but now that the base stations are already there, what have you gained? Maybe some limited utility in remote places that normally don't see human activity that somehow have many people with cell phones show up, while needing high accuracy GPS and forgetting their tripod-mount base station at home – but beyond that?
The novel part seems to be that they think they can build a business selling correction data back to the customers who provided it to them, not having to rely on third-party fixed base stations or installing their own.
I could see Qualcomm buying the patents, if any, for beamforming 5GSA optimization.
If you have a dozen people trying to navigate the same neighborhood of Boston* on foot, they could all be using the same app, all sharing data with each other and all benefiting from that data.
Does phase data require holding the receiver still? Anytime I've seen a survey GPS receiver, the surveyors seem to leave it in the same spot for a few minutes.
*There's some loss of generality here. In a grid layout place like New York this would be unnecessary, but in a less dense place it would be unviable.
I've seen cases where inaccurate positioning of poles by just 60 centimeters is enough to make pole loading analysis fail a pole.
Or perhaps the goal is to use data from dual freq GPS receivers to helping out the continuing majority of single frequency receivers?
Or perhaps they are just leaning on fixed base stations in urban areas to "bootstrap" and focusing on the P2P aspect as marketing, plus a convenient permissions grab for backhauling personal location information.