Passing along...How google largely solved urban canyon issue

[Kimo Crossman]

Skip the first hour to see presentation by  Dr. Frank Van Diggelen

https://vimeo.com/505359672

Pretty cool stuff using L5 chipping rate, machine learning and 3d building models for quite a lot of cities.

[gpsfan]

Cool stuff, there is some info about that here too https://www.gpsworld.com/google-to-improve-urban-gps-accuracy-for-apps/

Apparently the 3D models are being pushed via Google Play Services, the Pixel 4a and 5 already have v2 and it's coming to others. Not sure how to check for that though.

[Robert Enger]

The talk by Van Diggelen, as well as the statement from Broadcom in the gpsworld article, suggest that getting V3 is the important goal.

The higher chip-rate of the L5 signals is purported to reduce the demodulation ambiguity introduced by multipath reception of a satellite's signal.   

In Van Diggelen's presentation, the individual multipath components in the L5 are asserted to be narrow enough that they no longer overlap. 

 If the receiver can now discern individual rays of the multipath signal, one can imagine an extremely sophisticated machine model learning to leverage the multiple ray inputs, together with satellite ephemeris and the urban 3D model, to gain improved positioning accuracy.  But imagining and implementing are different things.  It will be interesting to see what real-world performance is provided by V3.

--
You received this message because you are subscribed to the Google Groups "GPSTest" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gpstest_andro...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gpstest_android/f6df04f1-b2d0-4aa5-917c-b00eca264fafn%40googlegroups.com.

[gpsfan]

Yeah a few things are not very clear :

- how does that database they push through Google Play Service help with the "chicken/egg" problem  ? Using Google Maps Live I can see how using the camera the server based algorithms can identify where you are and apply corrections but with the offline model ?

- assuming that problem is indeed solved how do they use the data (only 20Mb) ? Since the sat configuration at this time is X signals and you're receiving Y signals then you're around here ?

Too bad Google haven't come up with demo tools to show how their stuff works, right now there must be about 99.99% of Android users who have never heard of it ;-)

[gpsfan]

Just came across this : https://onenav.ai/solution/

L5, Machine Learning, 3D Buildings, hum...

[Knut]

Looks like the 3D models are the main distinction/contribution from Google here, 2nd freq support is only mentioned in context of Broadcom chip. I was scanning the details hoping to find even the slightest hint of committment by Google to dual freq support on the Pixel 4a 5G, but nothing...

[Robert Enger]

The video stresses the impact that L5 has on urban positioning accuracy.  The win (they claim) is the 5G signal offers narrower (in time) pulses, reducing or eliminating signal overlap when multipath conditions occur.  In the images, they show individual multipath components being discernable.  

In the talk the speaker almost dismisses the use of the L1 signal.  The promised win is not due to dual frequency, but sophisticated use of the L5 signals (and their multipath components).  

And the speaker indicated that the win won't be delivered until Version-3.

--

You received this message because you are subscribed to the Google Groups "GPSTest" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gpstest_andro...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/gpstest_android/fecb97c0-9bd3-48d4-a883-7c167e0310b3n%40googlegroups.com.

[gpsfan]

5G ? That's unrelated ;-)

But yes the L5 signal has 30 meter PR uncertainty vs the 300 of the L1 so that's going to help here. Like I posted the other day Google are not the only ones making that claim about L5 and 3D Machine Learning, etc... oneNAV are in fact developing an L5 only navigation solution https://onenav.ai/solution/

It's unlikely they're going alone on this, or they've licensed the technology from Google.

[Knut]

Understood, I was just trying to read the "fine print" to see whether Google had made any commitment regarding dual-freq GNSS support for its recently released Pixel Model "4a 5G". Apparently not. Pity.

[gpsfan]

So I got a chance to try "Live View" in Google Maps yesterday to navigate on foot in Paris and must say I was pretty impressed. It quickly locates you with the camera when you lift your phone and it scans the nearby buildings. Then when you start walking you have to lower your phone (you get a warning message to do that) and it shows the usual breacrumb track to follow. No more bouncing around from block to block.

As for the 3DML positioning I wasn't too impressed by the tracks I recorded but using GNSS Logger where you can see both the GPS and the Fused positions it did show some improvements. Without tracks being shown on the screen it's hard to get a good handle on it though. I'll see if I can import the log in the Windows analysis tool to see actual tracks.

On Tuesday, February 2, 2021 at 5:10:33 PM UTC+1 kimocr...@gmail.com wrote:

[gpsfan]

So I did some quick testing with a brand new Pixel 6 in a dense urban environment at the foot of tall buildings and can't say I saw much improvement from "Fused Location" over "GPS Location" using the GNSS Logger app. I had verified that there was some Google Maps "Live" data so presumably Google's "secret sauce" machine learning algorithms would have been in effect. Same usual "wrong side of the street" problems due to reflected signals. No wonder Google and Van Diggelen have gone mum on the subject, not that it interested too many people in the first place anyway...