Location Aware - Beware
By Mike Banahan
Fuelled by drunkenness and hubris, we set about building Somewherenear, a location-aware website that was, well, not so much location-aware as location-aware-ready. Since we don't know any way of getting positioning information out of the phone operators yet, whether they be I-Mode, WAP or anything else, for the moment all of its variants still ask you to tell it where you think you are. It would be so nice to be able to dispense with that — just having the ‘where is the nearest pub’ button — but that's not what we can do yet. Now it's built and working, it has made us think a bit harder about what location-aware / position-aware can really offer.
I was lucky enough to get invited along to the WROX Wireless Developers' Conference in Amsterdam to talk about how we built the thing. One of several things I learnt was that location-awareness is widely seen as a crucial component amongst those that will go to make up the alleged ‘killer application’ for mobile data. I've argued elsewhere that the WAP killer application is not an application at all, but instead the ability to bill for services, and though I've modified that opinion slightly after the conference, still think it's underestimated.
There are numerous compelling arguments about what location-aware services will be able to do for you. Opinions vary a bit about just how accurate that location has to be, and what it will be used for, but hardly anybody seems in doubt that it's going be very important. Some people tell me that street-level directions will be essential (something like “fifteen yards ahead, turn right, walk 100 yards, then left”), others that advertising push (“half price socks here for the next ten minutes”) will be the winner, others have yet other opinions. As a natural Luddite and blessed with a highly sceptical disposition, I'm not prepared to back a lot of those yet, apart from agreeing with the general trend. Geography matters — partly because it just matters and partly because it implies timeliness; what's a mile away is probably interesting, what's fifty miles away much less so, simply because it takes so long to get there. I am certainly sceptical about the need for directions, since I've yet to see a phone that knows which way it is pointing and therefore what ‘ahead’ or ‘left’ or ‘right’ mean. And I know a good slice of the population, close family members included, for whom the concepts of right and left are distinctly hazy. It strikes me that if you are really interested in finding a place, asking passers-by or hailing a taxi are at least as reliable as trying to download maps and directions into a mobile phone. Anyhow, I start to digress.
Sergio Rios gave an excellent presentation at the conference about how positioning is likely to be calculated, using various forms of technology. He also went into some issues around privacy and made some suggestions about the level of accuracy that can be obtained from position-aware systems. After that he described Ericssons' proposed standard for querying position, which had me trying hard not to laugh out loud, but I managed to quell the urge because the rest of the talk was so interesting. The basics of that — not entirely relevant here but worth mentioning — are that a consumer of position data has to ask the positioning server (at the telco's end) what is the position of a list of known subscribers; a list is returned. What I found amusing was the almost complete uselessness of that approach. It will work if a small group (I'll bet it's nearly always only one) of your users ask for specific position-related information. In that case you can go back to the position server with a request for the location of that specific handset. If you want to broadcast to a group of unknown users clustered around a point, you will have to ask for the position of everyone and then weed out the contenders for yourself. That will be hilarious when you have several million subscribers. I hope I misunderstood.
Anyhow, getting back to the point. If we simplify the detail, there are three ways of getting your location using current mobile technology. Either you use satellite positioning (GPS), or your handset triangulates its position from signals it hears from nearby cells and then gives it back to you, or the cells triangulate the handset's position in a nearly indentical model. The latter two don't really differ too much from the application builder's point of view. The GPS version can be augmented a bit by using information from a nearby cell to prime the GPS algorithms so they can resolve much more quickly than they would unaided. Claims of accuracy of within something like 30 metres are being made, depending on the particular technology deployed. Each type of delivery system (there are lots of different phone standards in use, especially in America) has its own quirks and peculiarities. In the US, carriers now have a legal duty to supply positioning information to the emergency services; in Europe there is essentially only the GSM system and it's commercial rather than regulatory pressure that is leading the development of position awareness. So far as I can tell, it is all in the future and there is not yet any public demonstration of how well it all works. So 30 metres is still a claim and we don't know what we will really get.
Years of being lied to by all kinds of people, especially those in marketing, have led me to be sceptical of almost any too-good-to-be-true claim. So instead of just meekly trotting out the received wisdom of what you can get from position-aware-services, I've done some testing of my own. The results so far are somewhat different from the rosy picture I've been hearing from the suits.
About three years ago I bought my first GPS receiver, a Magellan 2000. I wrote a package that runs on a laptop in the car and which gives me a moving-map display based on data coming out of the Magellan serial port. After a couple of years of driving around (probably about 40,000 miles) I think I've got used to the accuracy and reliability you get. Six months ago the car was burgled and so the twerps who did it got an ancient GPS receiver and a laptop with a broken case that will only run Linux unless you know how to boot Windows in 4MB. That prompted me to buy an up to date Garmin handheld. Since then, interested in what it's like to have accurate real-time positioning information, I've carried it around with me and looked at what it says. There is nothing like real experience to counter a bit of marketing spin.
What you learn is that GPS receivers are thirsty — 4 AA batteries do well to last 20 hours — and just how unreliable they are. With a clear view of the sky, no obstructions whatsoever, they will give you extraordinarily accurate figures. Block their view of the satellites and the situation is different. My old Magellan was useless anywhere except on open roads or very lightly urbanised areas, even with an external aerial. In towns it just gave up; the shielding effect of buildings was devastating and even a tree nearby could be a problem. The new receiver is obviously a generation later and since the old one was stolen the US operators of the GPS system have switched off the ‘selective availability’ feature which made civilian users get inaccurate data. If I'm lucky, with a window seat in a train the new receiver will be able to provide information something like 75% of the time. In a plane or a building, or a high-sided street it still fails. And it can take several minutes to lock-on when it's first switched on or even when it comes back into coverage after a black-out. If it is carefully positioned to give it a clear view of the sky, it is astonishing, but you have to take care.
The conclusion that I draw is that whilst GPS is staggeringly accurate — something like 3-5 metres at best — that's when you are lucky. Get inside a building or under a wet tree and it's likely not to work at all. Cars are difficult too, they have tin roofs that block the satellite signal completely. My Garmin is fine if I wedge it on the dasboard under the windscreen or put an external aerial on it, but in the car in my pocket it's dead.
The obvious counter to this experience is that it must be the fault of GPS and that triangulation will be better. I would say it's unwise to bet on that. For accurate positioning — within a few tens of metres — you have to use triangulation, so at least two and preferably three cells must overlap and be providing good signal strength. But hang on a second: to date, the whole art of cell design has been to eliminate overlap except at the edges. If you don't have overlap and triangulation, all you get is the location of the base station and in some cases an idea about how far away you are from it. I'd bet that most of the time you will be lucky if you get to within a few hundred metres and at times, even the nearest kilometre will be considered to be a good result. In urban areas, where accurate location is most important, you will be getting multipath reception with signals bouncing off walls, tunnelling down streets and generally travelling in unpredictable ways. If you think that you are going to get accurate timing signals for triangulation, then I beg to differ. You may be able to figure out how far the signal travelled to reach you, but that is not the same as your radial distance from base.
My concerns are based a) on real experience with GPS and b) speculation when it comes to triangulation. I make no claims to specific expertise on triangulation, except that I've been interested in radio propagation since I was a boy, still hold an amateur radio licence and specialised in radio transmission when I did my BSc. 25 years ago. And I can draw triangles.
Most of the time you will be doing well to know anything more than the location of your primary cell's base station. In a metropolitan area that might tell you to within a few hundred yards where you are, but often it will be much wider. Don't expect anything to do with GPS for years; it's power-hungry and too easily blocked. In rural areas you could be doing well to be located within miles of your true position. The point here is not what can be done in ideal circumstances — the real point is what you can guarantee to the customer. My exeriences with GPS have led me never to trust it and always to check by landmarks too.
So it might be time to think again about advertising push and the like. Instead of the vacuous claims of the marketing types, is there anybody out there who actually knows and dare speak up? What is the real reliable accuracy of mobile positioning, and what should we be banking on? I would love to know. I simply don't believe the claims I have been hearing, not least because the people making those claims have just about nil technical credibility. The descriptions I have heard of how positioning will be done have not improved my faith. Please, prove me wrong in practice. I don't want to know about what is achieved under ideal conditions, I want to know what the consumer experience will really be like.
If you do know the answers but your company policy prohibits public discussion, send me your thoughts from an anonymous email address like Hotmail and I'll summarise the results here.
Mike Banahan, 13th July 2000