[aprssig] RE: Indoor Tracking

[Jon Adams]

Jim's got some good questions here, and as he points out, the challenge of
knowing an asset's location is of a significant interest to the industry.
There are people using the traditional RFID tags, active or passive, but you
only really know where something is if it's near the reader. In a large
warehouse (or at a hamfest) it's rare that the asset gets close to the
reader unless you intended it.

Folks have already started using significant processing to take advantage of
the RSSI from 802.11 and 802.15.4 devices in order to calculate a estimated
position. It's processor intensive, needs a infrastructure that knows where
it is and has sufficient processing capability to do the math and provide
the final estimate of position to the roving device. But there's only so
much you can do with RSSI, given you don't really know the propagation
environment, but only can estimate it. Multipath plays havoc with the
measurement, requiring multiple independent measurements over time and from
different devices, and then post-processing that mess of raw observables
into something that resembles a X/Y(/Z) with DOP. The best I've heard of on
either .11 or 15.4 is about 2-3 m for non-moving things. For moving things,
it's not pretty.

There are plans now underway in 15.4a to add to the PHY the ability to do
precise ranging, subject to the amount of processing power and money you
want to spend. The current PHY of choice is Ultra Wideband (UWB), and the
expectation is fixed asset positions in the 1-20cm range for precision
location and more like metric for the easy stuff. There, again, the "tags"
will likely do no calculation, just send a message or respond to a message
from an infrastructure device, and let someone else do the heavy lifting.

The cellular GPS guys keep claiming that they're almost there on doing
location within a structure, but I'm afraid most of those guys (I know a
few) don't get out very much. The practicality of using unassisted GPS in an
indoor environment makes it worthless for any practical application. They've
got all sorts of schemes for indoor psuedosats, using network assist, etc.,
but really, once you walk in the door, all bets are off. The multipath sucks
majorly, and attempting to know what floor you're on is really difficult,
unless you augment the system with plenty of well surveyed infrastructure
devices that add cost and complexity. In addition, the sheer RF noise
environment works against you, since GPS is at the noise floor and part 15
devices are screaming in comparison. So I think the industry has waved off
the pitch by saying that if you're in a building, you know where you are...
At least to 75m...

At a conference I attended last year in Dallas, everyone was outfitted with
an intelligent "badge" which was more of a PDA with 2-way 900MHz radio
inside. The radios connected to one another via multihop packet, you'd input
your virtual business card and this would slowly propagate out to everyone
else. If someone decided that they wanted to meet up with you, they'd flag
your card in their badge's directory, it would send a message to your badge,
your badge would beep and give a range (less than 3m, less than 10m, etc.)
to the inquirer and you could then choose to meet. This badge did its work
using strictly RSSI - I'm not sure how it dealt with multihop (if it even
did deal with it). It was quite cool, but took enough fumbling with it that
I never had time to really play and exercise it to its capabilities. Don't
even remember what company it was.

RSSI is the only practical thing you've got in an indoor environment, and
still, to locate yourself with respect to the facility itself, you need
infrastructure points that are well-placed, surveyed, and have good
radiation characteristics. The mobile device has to have sufficient
processing that it can estimate some sort of range based upon RSSI from
those infrastructure devices, as well as knowing (probably through regular
beacons, kinda like GPS) what the "absolute" locations of the beacons are. 

Seems like it might be easier to to the electronic badge idea, which really
only needs measured signal strength to estimate a range (but not direction).
BTW, for most indoor scenarios, the 1/r^2 stuff is replaced with some higher
power exponent, due to multipath, absorption, etc. However, every
environment is very different, and it's tough to get to a common term that
fits most situations. I've seen values from 2.5 to 4 for the power,
depending on how nasty the environment is...

Jon n7uv


 
------------------------------

Message: 7
Date: Tue, 31 May 2005 10:54:43 -0700
From: Jim Lux <jimlux at earthlink.net>
Subject: [aprssig] indoor APRS
To: aprssig at lists.tapr.org
Message-ID: <6.1.1.1.2.20050531103818.01e79008 at mail.earthlink.net>
Content-Type: text/plain; charset="us-ascii"; format=flowed

Lots of interesting posts yesterday about location finding indoors.

As it happens, this is an area of great interest in the wireless comm 
business world, and there's also quite a lot of research interest (check 
out the work of Deborah Estrin and her students at UCLA, for 
instance).  Just about all the techniques described in the various emails 
have been done in one form or another, so you'd want to avoid "reinventing 
the wheel".

A couple generic comments:

1) What's the goal here?  Is it to push your location out to others, or is 
it to let your station know where it is?   The usual APRS model is that the 
station autonomously figures out where it is (either by keyboard entry, or 
from GPS, or any other technique that appeals to you), and then it 
broadcasts that position to all and sundry.  In general, you don't actually 
care what your position is (e.g. there are a number of "displayless" 
trackers around), just that it gets somewhere else.

So, if in a particular location there's some magic box that can figure out 
where you are, and push that information out, that would meet the need.

The appeal of the latter approach is that if you do it right, the roving 
nodes don't need ANY modification.  They could beacon their RF signal 
(indicating they're not getting any GPS fixes), and the magic box figures 
out where they're coming from, and deals with it accordingly.  No adding 
hardware to the rovers, no icky IR pods, etc.


2) If the rovers DO need to know their position, AND, they need to know it 
autonomously, then that's a generic localization problem, and one that has 
been studied copiously in the robotics world.  Google for "Where Am I?" by 
Borenstein at UofMich.  It's a several hundred page report that describes 
just about every method known to mankind. 
(http://www-personal.engin.umich.edu/~johannb/ will get you started).

In the APRS world, if you had the "magic locator box", then it could send 
the positions back to the rovers in a properly formatted message. I don't 
know if one has been defined in the APRS spec, but there should be one, 
because this scheme is at least as common in the robotics world as self 
localization.  Semantically, the message would be:  "From: MASTERBOX To: 
W6RMK-15 Payload: Your position is xxxx."  Whether W6RMK-15 reformats and 
repeats it is another story.  Or, whether the message should really be 
"From: MASTERBOX To:BIGWIDEWORLD Payload: W6RMK-15's current position is 
estimated to be xxxxx."  and let the rest of the world sort out who's got 
the best fix on W6RMK-15 (assuming that W6RMK-15 is also sending out it's 
own guess of the location).

3) As to mechanizations...

Assuming you want to use the 144.39 MHz transmission.   With a remarkably 
small number of receiving stations, just using power alone can get you 
fairly accurate positioning. (especially if all you want to know is which 
room you're in)  Sure, there's multipath, etc., but averaging over a 
reasonable time span (you're not driving at 60 mi/hr, after all) should 
work well.  You can also use some direction of arrival information.  I 
wouldn't use pseudo doppler... use a classical pair of figure 8 patterns 
(Adcock) or 3 antennas with coherent receivers, and do the processing on a 
PC at baseband.

If you've got multiple stations, you can use time of arrival (use the first 
arrival to edit out multipath), which is essentially a fancy phase 
measurement. You need stable oscillators at each station, but that's not 
all that tough these days, and you can transmit a low level continuous cal 
tone as well.

The best strategies use a combination of signal strength, direction of 
arrival, and multiple stations.  There's a lot of work in the acoustic area 
for this kind of thing (if you want to do high quality speech recognition, 
being able to figure out where the speaker(s) is (are) helps a whole lot).