# Neo1973 Update!

Tim Newsom cephdon at gmail.com
Tue Jun 5 18:18:06 CEST 2007

```On Tue, 5 Jun 2007 7:52, Mauro Iazzi wrote:
> On 05/06/07, Tim Newsom <cephdon at gmail.com> wrote:
>> If your tracking movement with 2 3D accelerometers... What would
>> another
>> one provide.
>> As far as I can tell (I am not an expert...)
>> Tracking all 6 vectors will tell you absolute movement in space.  I.e,
>> when 2 vectors point in the same direction with the same magnitude at
>> approximately the same acceleration as gravity.. Its probably laying or
>> positioned flat on that side.
>
> "Probably" is the key here. with two 3d (linear) accelerometers you
> cannot sense rotation around the axis between the two in an inertial
> frame of reference.
> Moreover you cannot distinguish if the Neo is laying face down or
> pushed downwards with 2mg force. This example is somewhat artificial,
> but means that you can probably find more realistic (though
> complicated) movements that are not distinguishable with only two
> accelerometers.
>
> You must then consider the errors which sum up, if you try to track. A
> rough mental estimate gives that you can sum up as much as 1 meter of
> error in ten seconds if you have a precision of 10^-3g over
> acceleration measure. (it does not mean that you are 1 meter away from
> the real position, it means that you can only be sure that you are at
> most 1 meter away from that).
>
> Most of the time you will need good assumptions to get any information
> from raw data:
>
> http://www.wiili.org/index.php/Motion_analysis
>
> can be of some help. No linear accel, no rotation, no tilt, are
> assumptions which can give some meaning to the data and can be done
> for single application, where you can assume the user will have some
> particular behaviour (or you require it).
>
> Still absolute tracking won't probably be anyhow realizable.
>
> --mauro

So are you saying that 3 3d accelerometers in a line with 2 on the end
and 1 in the middle will allow you to distinguish between rotation
around the center axis, etc?

It would seem to me that there are some realistic assumptions which can
be made to reduce error under normal usage.  In addition, in a
navigation sense it would seem that you can use gps to provide error
correction and thus be at least as precise (or maybe not far from it) as
the gps between times when you are out of gps signal (I.e. Tunnel) etc.

Other than a navigation use, accelerometers will be useful for
manipulating applications, but without a compass module, pointing or
other types of "external" information apps might not be possible
anyway.  If that's true, then each program will have some assumptions
built in for normal usage.
Errors can be mostly ignored since what will usually matter will be the
differences between vectors in very short timeframes OR the difference
between the start vectors and the current vectors.  If the phone is
suddenly dropped or thrown that's probably detectable as an extreme
motion and maybe ignorable. /shrug
--Tim

```