WTF with my computer clock?

[Jeff Liebermann]

On Tue, 18 Aug 2009 22:08:44 -0700, isw <isw@witzend.com> wrote:

In article <87pl851dui732j35j9oqki89ppqv79s2j3@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:

On Mon, 17 Aug 2009 07:41:19 +0100, "Dave Plowman (News)"
dave@davenoise.co.uk> wrote:

In article <4a88a59a$0$7469$822641b3@news.adtechcomputers.com>,
David Nebenzahl <nobody@but.us.chickens> wrote:
I agree that for most a minute per month is reasonable but I would
expect the same accuracy as my $29.99 Timex wristwatch which is more
like a second a month.

So that kinda begs the question of why computer mfrs. can't (or won't)
include clocks that are at *least* as accurate as a Timex, no? Wouldn't
a computah be a more compelling reason for a more accurate clock? (I
know, $$$ bottom line, right?)

Wonder if it's because a wrist watch is kept at a pretty constant
temperature via the skin?

Do you really expect people to wear a watch when they sleep just to
maintain accuracy? There's quite a difference in temperature between
skin temp (about 37C) and room temperature (about 25C). The same for
a computah. When turned off or in standby, the clock is slightly
above room temperature. When running, it might be as warm as 75C.

Yup, but the long-term average will be pretty good -- gain a little in
the daytime, lose a bit at night (or the other way around; could be
either one depending on how the circuit was set up).

Maybe, if the wearer maintains a regular schedule. That's a fair
assumption, until the wearer changes their usage pattern, such as
going on a ski trip.

Also, please note that the original discussion was over the accuracy
of a computah clock, not a wrist watch. Unless left on continuously,
computers don't maintain a set schedule. Even so, their internal
temperature is affected by the building environment.

Remember the old "Accutron" watches -- the ones with a tuning fork
inside? You could adjust those by deciding which way to lay them on the
table when you went to bed. "12 up" would run at a different rate than
"12 down" because of the effects o gravity on the fork. Also, they ran
noticeably fast on airplane trips, due to thinner air.

<http://members.iinet.net.au/~fotoplot/acc.htm>

I have a 1965 Accutron 214 Space View wrist watch in poor condition.
The specs offered 1 or 2 seconds per day, but only for the first year.
After about 30 years (the last time it ran) and zero service, my guess
is that it was off about 60 seconds per day. I forgot if it was a
gain or loss. The mercury battery leaked inside and it's
unfortunately not currently running. (Yet another project).

You might also be refering to the problem caused by the original steel
watch hands. When they were near the tuning fork coils, the frequency
would lower slightly. The effect was not very big, but still and
error.

The position problem is also not 12 o'clock up versus down. It's 12
o'clock verus 90 degree rotation which is 3 or 9 o'clock. The problem
stems from the tuning fork being vertical or horizontal. The
recommended solution is to lay the watch flat at night. I don't think
it was ever a major problem, just an interesting curiousity for
accuracy fanatics.

A more interesting problem was mechanical vibrations in the 360Hz
range. (the frequency of the tuning fork). When my watch was working,
it would tend to run quite fast if I was working near big synchronous
or induction motors driven by 60Hz such as in my fathers clothing
factory. It was not unusual to gain about a minute, after spending an
hour pushing cloth through an industrial sewing machine (with my hands
on the table). I suspect (guess) that vibration was also the problem
in airplanes, not thin air.

Temperature is of course a problem:
<http://bmumford.com/mset/tech/accutron/index.html>

--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Jeff Liebermann]

On Tue, 18 Aug 2009 22:26:03 -0700, isw <isw@witzend.com> wrote:

But even the "best" receiver still wouldn't solve the multipath problems
that plague the MW bands.

How much accuracy are you looking for in a PC clock? I doubt that WWV
will give you millisecond PC clock accuracy, but it's more than
suitable for nailing it withing one second. Averaged over even a
fairly short period of time, the 2.5/5/10/15/20Mhz frequencies are
quite accurate. For re-synchronizing the clock, the time ticks are
also sufficiently accurate:

WWV Frequency Accuracy
As transmitted: 1 part in 100 billion
As received: 1 part in 10 million

WWV Time Tick Accuracy
As received: 1 millisecond plus propagation delay

WWVB Frequency Accuracy
As transmitted: 1 part in 100 billion
As received: 1 part in 100 billion

The major ionospheric multipath problem is the almost 180 degree phase
reversals from constant path switching as the various incident and
reflected signals fade in and out. Yeah, that's going to be a
problem, but due to the limited accuracy required in a PC, it's not a
big problem.

--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Jeff Liebermann]

On Tue, 18 Aug 2009 18:49:05 +0000 (UTC), "Geoffrey S. Mendelson"
<gsm@mendelson.com> wrote:

Michael A. Terrell wrote:
They replaced all the transmitters and towers at WWVB a few years ago
to improve service. It now reaches Central Florida without a long wire
antenna & tuner. Why would they spend millions and take a couple years
to do the update if they were planning to shut it down?

They will shut it down eventually because of the cost. NTP servers cost
almost nothing, GPS is "free" because there are no incremental costs for
providing the time signals.

NIST funding was increaded from $719 million to $819 million.
<http://www.aip.org/fyi/2009/081.html>

National Institute of Standards and Technology:
The FY 2009 appropriation was $819.0 million.
The Administration's request was $846.1 million, an increase of 3.3
percent or $27.1 million.
The House bill provides $781.1 million, a cut of 4.6 percent or
$37.9 million.
The Senate Appropriations Committee bill would provide $878.8
million, an increase of 7.3 percent or $59.8 million.

<http://www.nist.gov/public_affairs/releases/approps-summary2008-2010.htm>
<http://www.nist.gov/public_affairs/releases/approps-summary2009.htm>

Operating costs of the various standards stations and labs seems to so
small as to not even be mentioned. I didn't see any mention of
shutting down anything.

--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Dave Plowman (News)]

In article <bk6o859u6lsbousrobhmes12ocrhoml1md@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:

Yup, but the long-term average will be pretty good -- gain a little in
the daytime, lose a bit at night (or the other way around; could be
either one depending on how the circuit was set up).

Maybe, if the wearer maintains a regular schedule. That's a fair
assumption, until the wearer changes their usage pattern, such as
going on a ski trip.

Also, please note that the original discussion was over the accuracy
of a computah clock, not a wrist watch. Unless left on continuously,
computers don't maintain a set schedule. Even so, their internal
temperature is affected by the building environment.

But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

--
*A journey of a thousand sites begins with a single click *

Dave Plowman dave@davenoise.co.uk London SW
To e-mail, change noise into sound.

 

[Meat Plow]

On Wed, 19 Aug 2009 18:14:14 +0100, "Dave Plowman (News)"
<dave@davenoise.co.uk>wrote:

In article <bk6o859u6lsbousrobhmes12ocrhoml1md@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:
Yup, but the long-term average will be pretty good -- gain a little in
the daytime, lose a bit at night (or the other way around; could be
either one depending on how the circuit was set up).

Maybe, if the wearer maintains a regular schedule. That's a fair
assumption, until the wearer changes their usage pattern, such as
going on a ski trip.

Also, please note that the original discussion was over the accuracy
of a computah clock, not a wrist watch. Unless left on continuously,
computers don't maintain a set schedule. Even so, their internal
temperature is affected by the building environment.

But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

Size?
Temp?

Does a tiny watch xtal garner any more accuracy merely because of its
size?

Does a watch xtal have a different temperature coefficient?

 

[Meat Plow]

On Wed, 19 Aug 2009 18:19:05 +0000 (UTC), "Geoffrey S. Mendelson"
<gsm@mendelson.com>wrote:

Meat Plow wrote:

Size?
Temp?

Does a tiny watch xtal garner any more accuracy merely because of its
size?

Does a watch xtal have a different temperature coefficient?

You are confusing the hardware clock and software clock in a computer.

The hardware clock is crystal controlled. It is used at boot time to set
the software clock.

The software clock is incremented by the lowest priority interupts, which
causes it to wander off.

There are various schemes to sync it with the hardware clock, but without
an external source, e.g. NTP, the don't work very well as hardware clocks
are not very accurate.

Geoff.

The question asked was;

On Wed, 19 Aug 2009 18:14:14 +0100, "Dave Plowman (News)"
dave@davenoise.co.uk>wrote:
But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

I had nothing confused.

 

[Jeff Liebermann]

On Wed, 19 Aug 2009 18:14:14 +0100, "Dave Plowman (News)"
<dave@davenoise.co.uk> wrote:

But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

Oh yes. The original Accutron was a steel tuning fork osillator. No
crystal of any kind to drive it. It depended totally on mechanical
stability.

Watch crystals come in a few flavors. The original version used
Statek type quartz tuning forks. They're really a mechanical tuning
fork made out of quartz:
<http://www.statek.com/products.php>
They work nicely at low frequencies and do not require a large divider
chain to drive the gears. 32.768Khz was the most common.

As IC technology progressed, it was more economical to use a big
divider chain and a higher frequency crystal such as 3.57945Mhz.
Meanwhile, someone figured out how to shrink the 32.768Mhz crystal, so
the next generation went back to those. (This is a gross over
simplification). The problem is that these relatively low frequency
and small physical size crystals have a terrible temperature
coeficient. Here's a typical data sheet:
<www.abracon.com/Resonators/AB26T.pdf>

The original IBM PC used a 14.31818MHz AT cut crystal. It was much
more stable, but there was no mechanism for adjusting the exact
frequency. There was also no temperature compensation or even the use
of temperature stable capacitors. This sorta explains how it works
and includes at series of curves for AT and SC cut crystals.
<http://www.4timing.com/techcrystal.htm>
The IBM PC oscillator was somewhat of an improvement in stability over
the typical watch crystal, but without an adjustment, it was nearly
useless.

Since 1981, I've looked inside literally hundreds of computahs and
SBC's. Not a single one has a tunable clock oscillator. One or two
used replaceable modular oscillators, which could pre purchased as a
TCXO, but which were usually supplied as a commodity clock oscillator.

These daze, the way to stabilize a TCXO is to first pre-age (beat-up)
the crystal to reduce long term drift. The crystal oscillator is then
characterized over the required temperature range. A table of
frequency versus temperature is generated and saved in a PROM. A PIC
controller on the oscillator takes the measured temperature, reads the
table, and applies the necessary correcting voltage to a varactor to
stabilize the oscillator over a very wide temp range. With this
method, you can take a really awful crystal, and compensate it to
impressive accuracies.

gotta run...


--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[David Nebenzahl]

On 8/19/2009 12:49 PM Jeff Liebermann spake thus:

Since 1981, I've looked inside literally hundreds of computahs and
SBC's. Not a single one has a tunable clock oscillator. One or two
used replaceable modular oscillators, which could pre purchased as a
TCXO, but which were usually supplied as a commodity clock oscillator.

So I wonder if the lowly SX28, one of my favorite little machines to
program (a PIC-like li'l guy) is an exception to this seeming rule?

I ask because, looking at the specs for this CPU, it has some
configuration bits (marked IRCTRIM0-2) that trim the internal RC
oscillator frequency, supposedly in steps of about 3%, up to a maximum
of +/- 8% (yeah, I know, doesn't add up, but whatever). Is this what you
would call a "tunable oscillator"?

These daze, the way to stabilize a TCXO is to first pre-age (beat-up)
the crystal to reduce long term drift. The crystal oscillator is then
characterized over the required temperature range. A table of
frequency versus temperature is generated and saved in a PROM. A PIC
controller on the oscillator takes the measured temperature, reads the
table, and applies the necessary correcting voltage to a varactor to
stabilize the oscillator over a very wide temp range. With this
method, you can take a really awful crystal, and compensate it to
impressive accuracies.

So presumably what I just described is a varactor built into the SX28.


--
Found--the gene that causes belief in genetic determinism

 

[Geoffrey S. Mendelson]

Meat Plow wrote:

Size?
Temp?

Does a tiny watch xtal garner any more accuracy merely because of its
size?

Does a watch xtal have a different temperature coefficient?

You are confusing the hardware clock and software clock in a computer.

The hardware clock is crystal controlled. It is used at boot time to set
the software clock.

The software clock is incremented by the lowest priority interupts, which
causes it to wander off.

There are various schemes to sync it with the hardware clock, but without
an external source, e.g. NTP, the don't work very well as hardware clocks
are not very accurate.

Geoff.

--
Geoffrey S. Mendelson, Jerusalem, Israel gsm@mendelson.com N3OWJ/4X1GM

 

[Jeff Liebermann]

On Wed, 19 Aug 2009 13:15:30 -0700, David Nebenzahl
<nobody@but.us.chickens> wrote:

On 8/19/2009 12:49 PM Jeff Liebermann spake thus:

Since 1981, I've looked inside literally hundreds of computahs and
SBC's. Not a single one has a tunable clock oscillator. One or two
used replaceable modular oscillators, which could pre purchased as a
TCXO, but which were usually supplied as a commodity clock oscillator.

So I wonder if the lowly SX28, one of my favorite little machines to
program (a PIC-like li'l guy) is an exception to this seeming rule?

Is that the Ubicom or Parallax SX28 processor? Dunno, I've never
worked with these. (Reminder: I are not a programmist).

I ask because, looking at the specs for this CPU, it has some
configuration bits (marked IRCTRIM0-2) that trim the internal RC
oscillator frequency, supposedly in steps of about 3%, up to a maximum
of +/- 8% (yeah, I know, doesn't add up, but whatever). Is this what you
would call a "tunable oscillator"?

I can't tell for sure:
<http://www.parallax.com/dl/docs/prod/datast/SX20AC-SX28AC-Data-v1.6.pdf>
See Section 9.0
I don't see any internal or external compensation for temperature
drift. It does have a real time clock, but again, no stabilization.
There is a section in the RC oscillator (FUSE register) which sets the
divider ratio from the RC oscillator. This is really a coarse
adjustment to set the divider ratio to generate an assortment of
frequencies between 31KHz and 4MHz. No way is it intended for fine
tuning for temp compensation.

These daze, the way to stabilize a TCXO is to first pre-age (beat-up)
the crystal to reduce long term drift. The crystal oscillator is then
characterized over the required temperature range. A table of
frequency versus temperature is generated and saved in a PROM. A PIC
controller on the oscillator takes the measured temperature, reads the
table, and applies the necessary correcting voltage to a varactor to
stabilize the oscillator over a very wide temp range. With this
method, you can take a really awful crystal, and compensate it to
impressive accuracies.

So presumably what I just described is a varactor built into the SX28.

I don't think so. I couldn't see such a feature on the data sheet.
Varactors are also chip real estate hogs, and would usually require
substantial documentation and explanation to impliment. I don't see
any of that in the data sheet.


I sorta blundered across this:
"NTP temperature compensation"
<http://www.ijs.si/time/temp-compensation/>

 

[isw]

In article <bk6o859u6lsbousrobhmes12ocrhoml1md@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:


Re: Accutron watches

-- snippage --

A more interesting problem was mechanical vibrations in the 360Hz
range. (the frequency of the tuning fork). When my watch was working,
it would tend to run quite fast if I was working near big synchronous
or induction motors driven by 60Hz such as in my fathers clothing
factory. It was not unusual to gain about a minute, after spending an
hour pushing cloth through an industrial sewing machine (with my hands
on the table).

I used mine for sports car rallies, and needed to synch it to WWV fairly
often (weekend rallies), but it was damn hard to set to the nearest
second even though I had the jeweler install the "hack" feature. I
learned to adjust it to run just slow enough so that accidental knocks
and so on would never put it ahead of time during the week. Then, simply
by giving it a good "thump" on the edge, I could overdrive the fork
briefly (it would do a three-tooth push on the driven gear instead of
the usual two), which would make it gain a good fraction of a second. A
few of those would get the thing spot on.

I suspect (guess) that vibration was also the problem
in airplanes, not thin air.

Bulova said it was air density.

Temperature is of course a problem:
http://bmumford.com/mset/tech/accutron/index.html

As well as every other momentum-transfer effect that plagues tuning
forks. Interestingly, they also affect those 32,768 Hz. crystals because
they are physically shaped like tuning forks (that's the only
oscillatory mode that can run that slowly in such a small piece of
quartz).

John Harrison's marine chronometer, developed for the British navy in
the mid-1700's, was good for about a minute a month, which was
considered the lowest accuracy usable for navigation. The Accutron was
the first "commercial" watch to have the same accuracy.

Isaac

 

[isw]

In article <508d5d4c04dave@davenoise.co.uk>,
"Dave Plowman (News)" <dave@davenoise.co.uk> wrote:

In article <bk6o859u6lsbousrobhmes12ocrhoml1md@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:
Yup, but the long-term average will be pretty good -- gain a little in
the daytime, lose a bit at night (or the other way around; could be
either one depending on how the circuit was set up).

Maybe, if the wearer maintains a regular schedule. That's a fair
assumption, until the wearer changes their usage pattern, such as
going on a ski trip.

Also, please note that the original discussion was over the accuracy
of a computah clock, not a wrist watch. Unless left on continuously,
computers don't maintain a set schedule. Even so, their internal
temperature is affected by the building environment.

But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

Most do. In many cases, the actual "CPU clock" of a couple of GHz. or
so, is derived from that same crystal, upconverted by a digital
phase-locked loop.

Isaac

 

[isw]

In article <licp855cgviq32hh2vtiugmno0gs4n12d9@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:

On Wed, 19 Aug 2009 13:15:30 -0700, David Nebenzahl
nobody@but.us.chickens> wrote:

On 8/19/2009 12:49 PM Jeff Liebermann spake thus:

Since 1981, I've looked inside literally hundreds of computahs and
SBC's. Not a single one has a tunable clock oscillator. One or two
used replaceable modular oscillators, which could pre purchased as a
TCXO, but which were usually supplied as a commodity clock oscillator.

So I wonder if the lowly SX28, one of my favorite little machines to
program (a PIC-like li'l guy) is an exception to this seeming rule?

Is that the Ubicom or Parallax SX28 processor? Dunno, I've never
worked with these. (Reminder: I are not a programmist).

I ask because, looking at the specs for this CPU, it has some
configuration bits (marked IRCTRIM0-2) that trim the internal RC
oscillator frequency, supposedly in steps of about 3%, up to a maximum
of +/- 8% (yeah, I know, doesn't add up, but whatever). Is this what you
would call a "tunable oscillator"?

I can't tell for sure:
http://www.parallax.com/dl/docs/prod/datast/SX20AC-SX28AC-Data-v1.6.pdf
See Section 9.0
I don't see any internal or external compensation for temperature
drift. It does have a real time clock, but again, no stabilization.
There is a section in the RC oscillator (FUSE register) which sets the
divider ratio from the RC oscillator. This is really a coarse
adjustment to set the divider ratio to generate an assortment of
frequencies between 31KHz and 4MHz. No way is it intended for fine
tuning for temp compensation.

These daze, the way to stabilize a TCXO is to first pre-age (beat-up)
the crystal to reduce long term drift. The crystal oscillator is then
characterized over the required temperature range. A table of
frequency versus temperature is generated and saved in a PROM. A PIC
controller on the oscillator takes the measured temperature, reads the
table, and applies the necessary correcting voltage to a varactor to
stabilize the oscillator over a very wide temp range. With this
method, you can take a really awful crystal, and compensate it to
impressive accuracies.

So presumably what I just described is a varactor built into the SX28.

I don't think so. I couldn't see such a feature on the data sheet.
Varactors are also chip real estate hogs, and would usually require
substantial documentation and explanation to impliment. I don't see
any of that in the data sheet.


I sorta blundered across this:
"NTP temperature compensation"
http://www.ijs.si/time/temp-compensation/

Putting a crystal in a temperature-stabilized "oven" is a well known
technique for generating a stable frequency (the telco folks and the
broadcast folks have been doing that for over 75 years, at least).

I have thought for a long time that it would be "neat" to glue a
resistor to the crystal case, and use heat to control the frequency.
You'd pulse-width modulate the power going to the resistor...

Isaac

 

[isw]

In article <32e27l.h00.19.4@news.alt.net>,
Meat Plow <meat@petitmorte.net> wrote:

On Wed, 19 Aug 2009 18:14:14 +0100, "Dave Plowman (News)"
dave@davenoise.co.uk>wrote:

In article <bk6o859u6lsbousrobhmes12ocrhoml1md@4ax.com>,
Jeff Liebermann <jeffl@cruzio.com> wrote:
Yup, but the long-term average will be pretty good -- gain a little in
the daytime, lose a bit at night (or the other way around; could be
either one depending on how the circuit was set up).

Maybe, if the wearer maintains a regular schedule. That's a fair
assumption, until the wearer changes their usage pattern, such as
going on a ski trip.

Also, please note that the original discussion was over the accuracy
of a computah clock, not a wrist watch. Unless left on continuously,
computers don't maintain a set schedule. Even so, their internal
temperature is affected by the building environment.

But is there any real difference between a 'quartz' watch and a PC clock?
They both rely on a low cost crystal?

Size?
Temp?

Does a tiny watch xtal garner any more accuracy merely because of its
size?

Because of the oscillatory mode, low-frequency watch crystals are
notoriously inaccurate.

Does a watch xtal have a different temperature coefficient?

Yes; poor, for the same reason.

Isaac

 

[Jeff Liebermann]

On Wed, 19 Aug 2009 23:28:19 -0700, isw <isw@witzend.com> wrote:

Putting a crystal in a temperature-stabilized "oven" is a well known
technique for generating a stable frequency (the telco folks and the
broadcast folks have been doing that for over 75 years, at least).

Yep. Motorola land mobile radios have had OCXO oscillators since the
1960's. However, in the mid 1970's, most land mobile radios switched
to TCXO (temperature compensated xtal osc), which draw less power, and
are less prone to burning out.

I have thought for a long time that it would be "neat" to glue a
resistor to the crystal case, and use heat to control the frequency.
You'd pulse-width modulate the power going to the resistor...

Yep. Some of the really cheap land mobile radios did that. There was
a metal clip, holding a resistor, sometimes with some silicon grease.
The problem with that scheme is that the lack of thermal insulation
means the resistor is going to burn plenty of excessive power heating
the nearby components and chassis. Same problem with a computah. Some
styrofoam insulation and a plastic can, is usually sufficient
insulation.

Note that there are quite small OCXO's that would work very nicely in
a PC. The small size and internal vacuum insulation means very little
heat loss and fairly fast warm up time.
<http://www.vectron.com/products/ocxo/ocxo_index.htm>

Incidentally, one problem with using an OCXO is that it sucks quite a
bit of power when the computah is turned OFF. If you kill the power
to the oven, the clock oscillator will drift away merrily, and there
goes your accuracy. I also don't think the EPA or Joe Sixpack will
appreciate the power drain. It certainly won't qualify for an Energy
Star rating.


--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558