Hi Felix, sorry for the late reply.
On 19.04.2014, at 18:51 , Felix Baumann <[email protected]> wrote: > But my idea was to make a different database containing the exact location of > cell towers (not logical cells!): lat/lon and the adress and pictures of the > tower and their type (GSM/UMTS/etc.) and which logical cells belong to them + > their direction maybe. > This data would be verified in a certain way. Only verified members/users can > upload data. And we could include existing databases of towers. This is an interesting idea, and indeed I misunderstood you. I’m unsure if it scales though. From what we’ve seen active stumblers are only capturing data in a small part of the world. Basically most of us live and work in the same place and most of us don’t go out of our way and go traveling around into all the remote places of this world. One conclusion we took away from that, is that we need different strategies to gather data, one of which is background collection in browsers/phones (via opt-in), where casual users are much more likely to engage and contribute. But those users are only likely to do this as a side-effect of going about their daily life and they should not have to take any constant explicit actions. This means whatever we do must happen transparently in the background for the majority of users. As an additional concern, it’s hard enough to gather data in all the rural and remote places. And the cell infrastructure is far from static. Rollout of new cell networks, like LTE take place, existing antennas are reconfigured to provide better service in populated areas, old equipment is replaced, carriers buy each other and thus logical cell ids change. All this means, we don’t only need to gather data once, but rather constantly update it. We have to consider this, and figure out what the extra value of individually verified data would be. I’m assuming that we would only get this sort of verified data in a small part of the world, and especially in those parts of the world, where we also got active contributors and thus good and frequent regular stumbling data. The basic idea of the current approach is to trust more data over less data. And so in the places where we could get verified data, we also most likely got more good than bad data already and would continue to do so. As an additional vector we’ll also start storing and using search traffic at some point. The search traffic can act as a way to verify the internal consistency of the data (at point A, I only see cell tower C1 and C2 now, but so far there was also C3, so probably C3 is gone) or act as a way to relate new data to the existing one (now there’s a new C4 in the neighborhood of C1 and C2, which we haven’t seen before at all). Again this works by trusting more searches over fewer / older data points. My current conclusion is that implementing individually verified cell locations isn’t effective, as the additional value is small. Hope that clarifies my current thinking and isn’t too confusing :) Hanno _______________________________________________ dev-geolocation mailing list [email protected] https://lists.mozilla.org/listinfo/dev-geolocation
