From: "Dave Emery" <die@pig.die.com>
Date: Sat, 27 Jan 1996 14:15:22 +0800
To: jimbell@pacifier.com (jim bell)
Subject: NOISE NOISE NOISE - clocks and other irrelevance
In-Reply-To: <m0tfyb0-00090XC@pacifier.com>
Message-ID: <9601270540.AA10607@pig.die.com>
MIME-Version: 1.0
Content-Type: text/plain



> 
> A peripheral I've long wanted to see, commonly available:  ACCURATE  time, 
> broadcast to the millisecond/microsecond/nanosecond, available from sources 
> as varied as TV VIR's, FM subcarriers, and other sources, available as an 
> easy input (via a peripheral card) to a computer.

	The only technology that I'd trust to be useful much below
10-100 ms is GPS.   The others are unlikely to be controlled well enough
at the source to be trusted.  Current TV broadcasting, for example,
usually involves multiple passes though digital frame stores and time
base correctors  - most homes get the signal via cable which itself
involves significant uncontrolled delays  (juat the thermal changes
in propagation delay  in a long CATV cable and amplifier chain due to weather
changes run into the many microseconds).

	And humans beings being as imperfect as they are, it is hard to
beleive that making sure that the time being broadcast is really kept accurate
is going to be a priority when most people use it for purposes that require
plus or minus a few seconds timing.

> 
> I have a 12-year-old Heathkit "Most Accurate Clock" that I assembled myself, 
> and had the foresight to install it with its computer  interface option. 
> (receives 5, 10, or 15 MHz signals broadcast from Boulder, Colorado, 
> containing "exact" time.) 
> 
> While I've never taken the time to connect it to my PC, it provides 
> (through an RS232 jack) correct time with a rated accuracy of about 5 
> milliseconds, as I vaguely recall. (Even has a dipswitch setup on the bottom 
> to tell it how many 500 mile increments you are away from WWVB... corrects 
> for delay to a first order of magnitude.)
> 
	WWVB is the 60 khz broadcast (which is more accurate due to more
stable propagation) . the  HF ones are WWV.  Commercial time receivers
are available that work off the 60 khz time code (very narrow bandwidth
ASK), but the 60 khz is most used as a standard frequency for long
term tracking of error in local standards.

> (BTW, if anybody knows how to easily connect it to the pc, or has the 
> appropriate software, please tell me  The task isn't difficult from a 
> hardware standpoint; it's just RS-232 serial ASCII timecode at about 9600
> bps which 
> either continuously retransmits or on request.  The problem is the software:

	If you run unix there are some quite sophisticated programs that
can use this specific clock (connected to a serial port) that allow sync
to the full accuracy possible at good times of day (around 1 ms).   The
programs also allow time distribution to other computers on a network -
thus their name  - ntp - which stands for network time protocol (and the
network time program that implements it). This protocol and the various unix
programs that implement it are quite widely used on commercial LANs and
the Internet to sychronize time amoung unix workstations, servers, and
bridges and routers.  Current implementations are capable of tracking
clock oscillator error on a system and adjusting the time periodically
to compensate for the frequency error of the clock and even to predict
(polynomial approximation) the change in frequency error with time.

	The man behind much of this (at least the early research) is
Dave Mills who used to be at louie.udel.edu which hosted a ftp
site for the programs.   An archie search will reveal where they
are kept now, and there is a newsgroup (comp.protocols.time.ntp) for this
which no doubt has a substantial faq file about this. 

	
>  How, exactly, do I INTERFACE such a serial input to the existing computer/RTC 
> combination? (Don't tell me to plug it into an unused serial jack!  I'm not 
> stupid. I'm not a  programmer, and I don't play one on TV! (I know 
> gates, flops, op amps, A/D, D/A, microprocessor hardware design, even some 
> Z-80 assy language, RF,  and I've programmed in Fortran, Basic, APL, Algol, 
> PL/1, Pascal, LISP, but not recently and I don't enjoy it!)
> 
> 

	I suspect that by this point there are several windoze/DOS  programs
to sync a PC to ntp time on a network, and perhaps even a program that will 
accept input from a Heath clock ... although the initial ntp code was
written for unix on Suns.
\

> (Then again, there are those "Receptor" watches which have (at least) similar 
> accuracy, which as I understand it work on FM subcarrier principles.)

	Yes they use the RDS broadcast on the 57 khz subcarrier for this.
Of course there is no certainty the station has the clock set accurately.
> 
> 
> Technology has now supplanted this old monstrosity:  Even with CHEAP GPS 
> receivers, they put out time which is rated in accuracy to well better than 
> 1 microsecond, and probably better than 200 nanoseconds even with S/A turned 
> on, and probably 100 nanoseconds with S/A off.  Once GPS receivers contain 
> equally cheap DGPS receivers, they'll be able to tell you your location to 
> about 1 meter and corresponding time accuracy, about 3 nanoseconds.
> 
	Yup.   And ntp can use several cheap gps products available to sync
a unix clock to high accuracy.

> I'm not particularly familiar with TV VIR signals, but I'd imagine they are 
> timecoded, or at least they COULD be without a lot of effort.  Resolution 
> would be FAR better than 1 microsecond, and accuracy would be primarily 
> limited by knowledge of your location compared to the xmitter.
> 
	Could be is the operative word here,  Many tv program
distribution signals carry frame time codes in the VIT, but who put them
there, how accurately they reflect local time  and where in the delay
chain (before or after the satellite for example) they get inserted is
not well controlled.   Nor is there a standard for the format that
addresses the needs of end users rather than broadcast production, or
any particular effort to ensure that a signal is reliably present in the
over the air transmission.

	Once many years ago (seventies)  the NBS (not NIST yet) tried to
get the TV networks to clock themselves with high accuracy rubidium
standards as a means of distributing standard frequency and time. But
technology has made this meaningless as most tv signals are now
distributed via satellites that move around several kilometers in a day
and clocked into multiple layers of frame stores (often delayed more
than a second)  in digital switching and processing gear and clocked out
with different clocks that often are not very accurate and not in any
way locked to the incoming network.

	TV stations could be made to maintain a local clock sync'd to
GPS and use that to do the final level of clocking out before feeding
the transmitter and could thus ensure that some reference point in some
frame happened at an exact time, but given that a user who can see a TV
signal can probably see GPS signals and can do the same timekeeping himself
for a couple hundred bucks it hardly seems worth it any more.  I do
expect that time codes with modest accuracy (few tens of ms at best)
will become common as part of the Starsite (or whatever they call it
now) program guide distribution on PBS, simply because this has defined
a format that can conveniantly contain time messages multiplexed with
other data and the box displays the time.  DSS and VC-II both also have
this capability, but of course the uncertainty of the satellite delay
limits accuracy and neither has provisions for providing time to other
devices.


> MITM attacks would be far more difficult if both ends of the data 
> conversation agreed on the "exact" time, and could detect transmission 
> delays and CHANGES in transmission delays.  While it would be possible to 
> locally spoof the accurate timecode, a cheap version of a "disciplined 
> oscillator" (which any GPS receiver is going to have, anyway) would detect 
> such short-term spoofing trivially.
> 

	See my public comments on this.



> Occasionally, I've speculated on whether it might be useful to be able to 
> synchronize (or, at least, KNOW) to the PHASE of the 60 Hz power grid.  
> True, I know that the HV grid is 3-phase and most people won't know which 
> phase they're on anyway, but that wouldn't change (at least not frequently!) 
> , and I would imagine that 
> it might be  useful.  You wouldn't necessarily know which CYCLE you're on, 
> either, but again that might be compensated for somehow.  If  your computer 
> were talking, locally, to another computer at 4100 baud (? whatever) (7 bits 
> per symbol(?); equals 28.8kbps) you could "easily" agree on a particular cycle 
> relationship, which is going to be essentially constant over a distance of a 
> few tens or even hundreds of miles.
> 
	This is possible, but I bet the variations in phase in the local
distribution system due to power factor, choice of phase to use, propagation
time through transmission lines and substations and so forth would
mean that phase as observed at two distant sites was rather random
and maybe even subject to shifts over time as load conditions varied.

> What I DON'T know (and some HV transmission engineer will probably be able 
> to tell me, hint hint!) is how STABLE this phase is across the entire 
> country?  I realize that this will probably depend on who'se shipping excess 
>  power to whom at the moment, But I'd imagine the variability will be 
> distinctly limited.
> 
	I've seen some discussions about this, but don't know a reliable
answer.  I do know that the frequency is only 60 hz on the average over
a day and actually wanders up and down quite a bit more than one might
expect as load on the system varies.  I did some measurements of this 22
years ago while debugging some PDP-8E system software I wrote that that
ran a frequency counter (the ratio kind that was very accurate on low
frequencies) and found the diurnal variations surprisingly large and
quite interesting.  I've not repeated the experiment since but suspect
that they still allow the frequency to wander in response to load conditions.

> The biggest attraction of such a system is that the interface would probably 
> be trivial:  Getting it from the P/S is out because they didn't anticipate 
> such a thing.  The easiest interface might be an AC wall xformer with a 
> rectifying limiter and slicer  (Okay, maybe just a resistor and a diode, 
> possibly with the addition of a comparator for precision), driving a 
> readable pin on an otherwise-unused RS-232 interface.  (Possibly 
> installed similar to a dongle.)  Appropriate software (yucch!) would read 
> the square waves, and record the phase at any one time.  Such information 
> could be used to verify the relative synchronization between two different 
> computers, although it would be necessary to identify particular phases, as 
> I mentioned before.
> 
	One could certainly do this, but there are subtlies ... some
places and institutions generate their power locally (and few if any
users know this or know whether or not they are on the grid), UPS
systems are common and wander off of the grid during a power fail,
and many buildings have all three phases floating around wall outlets,
even wall outlets close to each other so such acts as moving plugs
around might very well change the phase.  And power systems switch
phase correction capacitors in and out from time to time as power
factor of large loads varies.  My guess is that to synchronize much
below a ms would be hard, and that random losses and jumps of sync would be
common enough to require lots of special treatment in software.

						Dave Emery  N1PRE
						die@die.com