[Tinyos-devel] Re: [Tinyos Core WG] time sync interfaces

[John Griessen]

Philip Levis wrote:
> 
> On Dec 18, 2007, at 2:30 PM, Omprakash Gnawali wrote:
> 
>> On Dec 18, 2007 2:09 PM, David Gay <dgay42 at gmail.com> wrote:
>

>>> PS: I suspect with reasonable evidence that millisecond level timesync
>>> is practical.
>>>
>>
>> Or you might need more i

> True, but I think the question is what's the 90%. 

> Phil

To be a HIL, the app can't use specific timers, so it will be limited to
using feedback to correct errors and may have lots of jitter.  It's not going to be
smoothly nudging the time, since it is the same app for all different hdwe.

"Lots  of jitter" is going to be a random
function that likely will kind of fit to a Poisson's distribution where 70% is tighter
than the worst case by 40%.  The places to look for parts of the "worst case jitter"
are in the RX of a packet preamble to when you put a time reading with it, and
whether that has some dependence on interrupts or not.  The accuracy of crystals
won't cause much difference in the worst case numbers if the Xtal is  2% or a
3% of nominal value.  A HIL app will be error correcting all the time, so it will
force a local clock back to where it calculates the master clock is.
The master clock is some specific node on the network.

You know... a HIL timesync app may not be very useful for field work...but it
is really a decent start on a precision time protocol app when you get the right
tunable local clocks, and deterministic timestamping of the RX preamble with that same
tunable local clock's time.

A HIL timesync app would be useful as a guide for a real future app
that depends on certain generic kinds of hardware, and once you got the better hardware,
plus a very minor Hpl module for it,
the app would just sstart giving you more precision as it is.

John Griessen

-- 
Ecosensory   Austin TX
tinyOS devel on:  ubuntu Linux;   tinyOS v2.0.2;   telosb ecosens1