To clarify the scope of the initial MLS use in FxOS: We are first looking at 
Tarako devices, which don’t have a GPS chip. So in order to provide user-value, 
all we have to do is beat “no position estimate”. 

One primary way to achieve this is using GeoIP, which will give you country or 
city-level position estimates. That’s not great for many use-cases, but it’s 
better than nothing. One example might be an app showing you restaurants around 
you. The alternatives are either to show a default world-wide view (with no 
position estimate) or to zoom the map view to the country / city you are 
actually in. The latter still requires the user to pan and zoom a bit to get to 
the right area, but it’s better than a world-wide view. While GeoIP is pretty 
bad for mobile networks, a lot of the target users will use WiFi networks to 
access the internet, as the data connections are too unreliable/slow/expensive. 
GeoIP works better for those landline-based WiFi networks. In addition the code 
is already in FxOS to send observed cell and wifi networks along to MLS. So 
once we get more and better data for the countries in question, the user 
experience will get better without any need to update the client software.

A second use-case on more capable hardware is around misconfigured or missing 
A-GPS support in FxOS devices. One part of A-GPS is to inject a coarse position 
estimate into the GPS chip with an accuracy required to be in the order of 
100-200km. This helps the GPS chip to figure out which satellites should be 
visible in the sky and look for their specific signals. As this use-case only 
requires a very coarse grained position estimate, it’s also something we can 
provide based on the current MLS. This helps with reducing the “time to first 
fix” from a worst case of 12 minutes to much less time.

Of course we want to provide better position estimates and make the service 
useful for other use-cases. But that requires a mixture of getting much more 
data and better algorithms and approaches on collecting and processing that 
data. Those things are underway, but don’t have to block MLS use in FxOS, as we 
can already provide some user value.

On 25.03.2014, at 06:05 , Adrian Custer <[email protected]> wrote:
> It would be interesting to have the discussion of the work you are doing to 
> bring the whole system up to production quality on the
> https://mozilla-ichnaea.readthedocs.org/en/latest/calculation.html
> web page. That should probably include a discussion of all the different 
> error terms you are considering in the position analysis and how multiple 
> observations attenuate the different terms.

The short answer is: We aren’t doing any real work in the current codebase. The 
assumption so far is that we only get a single or very few observations for any 
given cell or wifi network, as most of the areas are only stumbled once or very 
infrequently for each cell network / cell standard combination. With so little 
data per network, there’s not a whole lot of algorithms you can use. We are 
fully aware that this leads to position estimates with low accuracies. To cover 
the widest area we are going to rely on cell networks. As a result of 
limitations in GSM networks, the devices will only send us a single cell 
observation for the currently connected cell. The best we can do here is to 
equate the estimated user position with the center of the cell and a range 
estimate of the cell size (this method is generally known as CellId). That 
leads to accuracies in the kilometer range for urban areas or tens of 
kilometers in suburban/rural areas. The GPS inaccuracies or inaccuracies from 
sensor / timing mismatch aren’t having a big impact at this scale.

I’m more than interested in hearing about your work and your own conclusions. 
We will look at better algorithms in the near future. But those are a secondary 
priority after getting any data at all.

We are following a very incremental approach here, where we try to provide 
value at each step of the process and not wait until we can beat the current 
market offerings from competitors with many years of experience.

Hope this clarifies the current situation a bit.
Hanno
_______________________________________________
dev-geolocation mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-geolocation

Reply via email to