Checksum: 47072
Path: utzoo!utgpu!dennis
From: dennis@gpu.utcs.utoronto.ca (Dennis Ferguson)
Date: Tue, 30-May-89 22:41:16 EDT
Message-ID: <1989May30.224116.2101@gpu.utcs.utoronto.ca>
Organization: Mechanical Engineering, University of Toronto
Newsgroups: comp.dcom.lans
Subject: Re: Radio Time Clocks
References: <895@osf.OSF.ORG> <3086@daisy.UUCP> <2767@csccat.UUCP> <600@lzaz.ATT.COM> <13437@ut-emx.UUCP> <1989May27.222428.4127@utzoo.uucp> <1970@ccncsu.ColoState.EDU> <16468@bellcore.bellcore.com>
Reply-To: dennis@gpu.utcs.utoronto.ca (Dennis Ferguson)

In article <16468@bellcore.bellcore.com> karn@jupiter.bellcore.com (Phil R. Karn) writes:
>Has anybody built a clock that uses CHU's encoded time signals?

Not yet, really, but I've been working on and off on a CHU-synchonized NTP
server for the past few months.  Maybe a month more to get things tested
to the point where I know whether it can be trusted or not.

The nice thing about CHU, beyond the fact that if a serial time code is
what you want you can get by with just a radio and a 300 bps modem, is that
the data arriving at your computer's serial port is generated directly from
the atomic clock itself, rather than being second hand via a radio clock.
This allows you to process using some heavier duty algorithms than your
average clock manufacturer is likely to be able to put together, and also
gives you more intimate knowlege of how the receiver is doing so you can
take approprate action (the Heathkit clock wouldn't be so bad if it would
tell you it was running into difficulty early enough for you to stop
believing it in time).  The end of every data byte gives you a time sample
(90 per minute in all) meaning you've got lots and lots of samples to feed
to the filters.  NTP eats errors and short term variations of known
(observed, measured) statistical properties for breakfast.

I like the idea of time-synchronizing random hosts (i.e. anything that has
something to do besides time keeping) to a well behaved NTP server
across a network rather than trying to deal with a radio clock directly.
You can get some pretty amazing results with NTP, even across large chunks
of Internet topology let alone across a single LAN, without much effort
on the part of the host.  The grot of dealing with a radio clock can
probably be better handled by something which hasn't got much else to do
besides watch the clock.  For isolated hosts the telephone time service
might be a better bet.  Unless you live in Ft. Collins or Ottawa, or
can afford a GOES clock, getting good, reliable time from a radio time
source can take a lot of work.

Dennis Ferguson
