Re: [Geopriv] RFC 3825 Updates

From: Henning Schulzrinne ^lt;hgs@cs.columbia.edu>
Date: Fri Feb 02 2007 - 11:47:23 EST

Let me try a new angle that just might bridge the point vs. area
discussion. I think both somewhat miss the point (sorry). We need to
think probabilistically, not deterministically.

For all measurements, we're talking about a probability distribution
of measurements. There are three cases:

(1) The probability of being outside the area is zero. This is only
true if the area circumscribes, say, a property line and there is no
realistic chance that a device associated with that area can be
outside that area. This has nothing to do with measurement errors as
such, just physical limitations of where, say, Ethernet sockets are.
Obviously, even in buildings with DHCP, this is sometimes fuzzy,
e.g., if there's a wireless hotspot that radiates outside the
building. Nothing is said about the likelihood distribution within
that area.

A fuzzed (privacy-obfuscated) area is a special case. In other words,
the actual measurement is much more precise than what's conveyed, so
the object is (virtually) guaranteed to be somewhere inside the
indicated ranges of coordinates, with equal probability.

(2) The circle or other area describes a probability distribution
induced by measurement errors. In almost all cases, this really means
that x%, for some large x (say 90 or 95, but could be 66 in some FCC
regs), of the time the true location is indeed inside that area and
for (100-x)% it is outside. This seems to be the typical definition.
However, nothing is said about the probability distribution within
that area - the recipient of the information has to assume that the
likelihood of the object being at any given point within the area is
equal.

In reality, the latter isn't often quite true. Thus, the most likely
location is close to the centroid of the area. However, just having
an area doesn't tell you one way or the other.

(3) A point with an uncertainty region. Here, the assumption is that
the highest-probability location is at the point, with an uncertainty
region of the x% kind around it. This is the typical measurement
uncertainty/significant figures case.

Thus, the answer is not "point or area", but "both".

Unfortunately, 3825 and many other specs are often unclear what
assumptions they make on the probability distribution. 3825 addresses
both the fuzzed case and (3), but the probability interpretation
differs.

Henning

On Feb 2, 2007, at 11:07 AM, Brian Rosen wrote:

>> Thanks very much for giving your opinion (I'm rather shocked you
>> withheld it in the first place :) ). Just one question, do the
>> values conveyed in 3825 constitute a point or an area?
> A point
>
>
> _______________________________________________
> Geopriv mailing list
> Geopriv@ietf.org
> https://www1.ietf.org/mailman/listinfo/geopriv

_______________________________________________
Geopriv mailing list
Geopriv@ietf.org
https://www1.ietf.org/mailman/listinfo/geopriv
Received on Fri, 2 Feb 2007 11:47:23 -0500

This archive was generated by hypermail 2.1.8 : Fri Feb 02 2007 - 11:47:15 EST