[time-nuts] Raspberry Pi TCXO Hat

[Richard Laager]

I stumbled across this product listing:
https://www.teradak.com/products/115.html

It's a Pi hat that contains two TCXOs, one at 19.2 MHz for the CPU and
one at 25 MHz for the Ethernet controller. It's a rather permanent
modification, as you have to desolder the stock crystals and replace
them with wires going to the hat. I'm not 100% sure how to order the hat
yet, but they gave a price of $46 (USD) via email.

I'm curious if this would provide any meaningful improvement in system
clock accuracy, for NTP, if I'm already a GPS PPS hat. If there's a
reasonable chance this could be interesting, I'm thinking about ordering
a couple and "sacrificing" a Pi 3 and/or Pi 4.

In terms of mounting both hats, I'm guessing that the TCXO is only using
power from that hat connector, so rather than use the hat connector, I
would just solder some wires to it and the Pi, such that I could stack
the TCXO on top of the GPS hat physically but not need to connect them
electrically.

-- 
Richard

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[David J Taylor via time-nuts]

I stumbled across this product listing:
https://www.teradak.com/products/115.html

It's a Pi hat that contains two TCXOs, one at 19.2 MHz for the CPU and
one at 25 MHz for the Ethernet controller. It's a rather permanent
modification, as you have to desolder the stock crystals and replace
them with wires going to the hat. I'm not 100% sure how to order the hat
yet, but they gave a price of $46 (USD) via email.

I'm curious if this would provide any meaningful improvement in system
clock accuracy, for NTP, if I'm already a GPS PPS hat. If there's a
reasonable chance this could be interesting, I'm thinking about ordering
a couple and "sacrificing" a Pi 3 and/or Pi 4.

In terms of mounting both hats, I'm guessing that the TCXO is only using
power from that hat connector, so rather than use the hat connector, I
would just solder some wires to it and the Pi, such that I could stack
the TCXO on top of the GPS hat physically but not need to connect them
electrically.

Richard
=

Richard,

As a pure sacrificial experiment I would be interested to know your results.

In terms of NTP time keeping, I'm not sure you would gain a lot.  Best 
results here are obtained when the RPi is kept at a steady or very slowly 
changing temperature.  I have one which is in a normally closed cupboard 
butting onto a north-facing outside wall:


 https://www.satsignal.eu/mrtg/raspi1_ntp.html

and this is a Raspberry Pi 1 model B.  By contrast, here's one sitting near 
a central heating radiator:


 https://www.satsignal.eu/mrtg/raspi12_ntp.html

which is an RPi 3 model B.  Both have a constant CPU load (light on RasPi-1, 
14% average on RasPi-12.  Both have GPS/PPS sources attached.


Given that the path to the outside Ethernet world on both of those models is 
via a USB controller, I would expect to see very little improvement with a 
TCXO in a constant temperature environment.  Just perhaps with a RasPi-4 you 
/might/ see an improvement as the Ethernet has a more direct path, but 
that's a board where self-generated heat is greater, so the thermal issues 
might make things worse.


Try it and see!

Cheers,
David
--
SatSignal Software - Quality software for you
Web: http://www.satsignal.eu
Email: david-tay...@blueyonder.co.uk
Twitter: @gm8arv 



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Hal Murray]

rlaa...@wiktel.com said:
> I'm curious if this would provide any meaningful improvement in system clock
> accuracy, for NTP, if I'm already a GPS PPS hat. If there's a reasonable
> chance this could be interesting, I'm thinking about ordering a couple and
> "sacrificing" a Pi 3 and/or Pi 4. 

What are your goals?  NTP server?  Neat graphs?

The Pi 3 has Ethernet on USB.  That adds a layer of jitter to packet timings 
so a Pi 3 even with GPS will never make a great NTP server.

How stable is your temperature?  Self heating as the CPU load changes is 
important even if the room temperature is stable.  A TCXO won't help if the 
temperature is already stable.

If you have 2 identical Pi-s, you could put the TCXO on one and run a 
side-by-side comparison and tell us what happens.


-- 
These are my opinions.  I hate spam.




___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Adrian Godwin]

On Sat, Nov 2, 2019 at 11:59 AM David J Taylor via time-nuts <
time-nuts@lists.febo.com> wrote:

>
> Given that the path to the outside Ethernet world on both of those models
> is
> via a USB controller, I would expect to see very little improvement with a
> TCXO in a constant temperature environment.  Just perhaps with a RasPi-4
> you
> /might/ see an improvement as the Ethernet has a more direct path, but
> that's a board where self-generated heat is greater, so the thermal issues
> might make things worse.
>
>
There's a bunch of 'ifs' there though - on the early Pis, the ethernet chip
was a major source of heat. Presumably it's a different, non-USB part on
the Pi4 so that may or may not have changed.

Does the Pi4 inherently run hotter, or is it just capable of running at a
level where it consumes more power ? If it's under-used and maybe even
intentionally underclocked, it might do better than earlier versions.

It would at least be worth including a Pi4 running the same workload as the
other boards in any parallel test.
___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Adrian Godwin]

Also - I'm unsure of the benefit of having a TCXO for the ethernet clock
unless it runs so hot that ethernet can't sync using the usual
uncompensated crystal. Is there some benefit I'm not seeing ?


On Sat, Nov 2, 2019 at 12:35 PM Adrian Godwin  wrote:

>
>
> On Sat, Nov 2, 2019 at 11:59 AM David J Taylor via time-nuts <
> time-nuts@lists.febo.com> wrote:
>
>>
>> Given that the path to the outside Ethernet world on both of those models
>> is
>> via a USB controller, I would expect to see very little improvement with
>> a
>> TCXO in a constant temperature environment.  Just perhaps with a RasPi-4
>> you
>> /might/ see an improvement as the Ethernet has a more direct path, but
>> that's a board where self-generated heat is greater, so the thermal
>> issues
>> might make things worse.
>>
>>
> There's a bunch of 'ifs' there though - on the early Pis, the ethernet
> chip was a major source of heat. Presumably it's a different, non-USB part
> on the Pi4 so that may or may not have changed.
>
> Does the Pi4 inherently run hotter, or is it just capable of running at a
> level where it consumes more power ? If it's under-used and maybe even
> intentionally underclocked, it might do better than earlier versions.
>
> It would at least be worth including a Pi4 running the same workload as
> the other boards in any parallel test.
>
>
___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[MLewis]

The Pi 4 launched with issues, including heat issues, but has been 
improved. The improvements in power usage are supposed to reduce the 
heat issues. A new bootloader was released. It includes that by default, 
power will be maintained on the power pins, so power is maintained on hats.


https://github.com/raspberrypi/rpi-eeprom/blob/master/firmware/release-notes.md
https://www.raspberrypi.org/downloads/

There's a new tool for updating the Pi 4 bootloader EEPROM.
https://www.raspberrypi.org/documentation/hardware/raspberrypi/booteeprom.md
https://github.com/raspberrypi/rpi-eeprom

https://hackaday.com/2019/10/30/rpi4-now-overclocked-net-booted-and-power-sipping/

"The USB controller will get hot based on its usage now, instead of get 
hot always. If you use it to its fullest it will get just as hot, but if 
you don�t use it much or at all, it will be cooler."


"the bootloader (and associated firmware) actually reduce the power 
consumption of both the USB controller and core SoC, with the USB 
controller accounting for about half the power saving.

...
PCIe ASPM savings actually account for about half of the idle power 
savings realised so far. Throughput loss to mass storage due to ASPM 
appears to be in the ~2% range.


The rest is made up of SDRAM PHY optimisation, clocking improvements, 
changes to the load-step response of the PMIC (allowing us to reduce 
load-step margin on the core voltage). We also have a DVFS scheme which 
makes the first (1GHz) frequency back-off much more energy efficient 
than it was at launch."



Michael

On 02/11/2019 5:13 AM, David J Taylor via time-nuts wrote:

...
Just perhaps with a RasPi-4 you /might/ see an improvement as the 
Ethernet has a more direct path, but that's a board where 
self-generated heat is greater, so the thermal issues might make 
things worse.


Try it and see!

Cheers,
David



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Hal Murray]

artgod...@gmail.com said:
> Also - I'm unsure of the benefit of having a TCXO for the ethernet clock
> unless it runs so hot that ethernet can't sync using the usual uncompensated
> crystal. Is there some benefit I'm not seeing ? 

I had similar thoughts.

One could imagine using Ethernet to distribute a clock Just tap off the 
post-PLL receive clock.  For that to make sense, you would need to start with 
something like a GPSDO rather than a TCXO.

Maybe you could measure the local transmit clock and send that to the receive 
side.  I don't see how to make that work rock-solid.

-- 
These are my opinions.  I hate spam.




___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Achim Gratz]

Richard Laager writes:
> I stumbled across this product listing:
> https://www.teradak.com/products/115.html

That site red-zones my BS detector…

> It's a Pi hat that contains two TCXOs, one at 19.2 MHz for the CPU and
> one at 25 MHz for the Ethernet controller.

The one for the Ethernet controller is useless even for the seemingly
intended application.  Replacing the main crystal with a TCXO may or may
not help NTP performance depending on how much temperature effects are
currently dominating it.

> It's a rather permanent modification, as you have to desolder the
> stock crystals and replace them with wires going to the hat. I'm not
> 100% sure how to order the hat yet, but they gave a price of $46 (USD)
> via email.

You'd be much better off taking not of the injection point and then use
a GPS timing module that can produce a clean 19.2MHz clock.  I still
have that way down on my to-do list; although I was going to try and
inject the clock without actually desoldering the crystal so it would
still work even when the GPS module doesn't provide a clock.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

DIY Stuff:
http://Synth.Stromeko.net/DIY.html

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Bob kb8tq]

Hi

Congratulations !!!

You have just invented Sync-E. Seems like *maybe* somebody beat you to it ….

https://en.wikipedia.org/wiki/Synchronous_Ethernet 


Turns out it is one way to get *frequency* to a device (like a cell site). Just 
how well
it does is (as always) a “that depends” sort of thing. 

Bob

> On Nov 2, 2019, at 5:44 PM, Hal Murray  wrote:
> 
> 
> artgod...@gmail.com said:
>> Also - I'm unsure of the benefit of having a TCXO for the ethernet clock
>> unless it runs so hot that ethernet can't sync using the usual uncompensated
>> crystal. Is there some benefit I'm not seeing ? 
> 
> I had similar thoughts.
> 
> One could imagine using Ethernet to distribute a clock Just tap off the 
> post-PLL receive clock.  For that to make sense, you would need to start with 
> something like a GPSDO rather than a TCXO.
> 
> Maybe you could measure the local transmit clock and send that to the receive 
> side.  I don't see how to make that work rock-solid.
> 
> -- 
> These are my opinions.  I hate spam.
> 
> 
> 
> 
> ___
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe, go to 
> http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Chris Caudle]

On Sat, November 2, 2019 6:37 am, Adrian Godwin wrote:
> Also - I'm unsure of the benefit of having a TCXO for the ethernet clock

Would only be of benefit if using an Ethernet device which timestamps IEEE
1588 packets with the clock running in the NIC.  Typical usage on a linux
based system involves transferring timestamps between NIC clock and system
clock, so if both of those clocks are low drift it can simplify keeping
the system and NIC clocks synchronized.

-- 
Chris Caudle



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Richard Laager]

On 11/2/19 1:15 AM, Hal Murray wrote:
> 
> rlaa...@wiktel.com said:
>> I'm curious if this would provide any meaningful improvement in system clock
>> accuracy, for NTP, if I'm already a GPS PPS hat. If there's a reasonable
>> chance this could be interesting, I'm thinking about ordering a couple and
>> "sacrificing" a Pi 3 and/or Pi 4. 
> 
> What are your goals?  NTP server?  Neat graphs?

Closer to the latter.

> The Pi 3 has Ethernet on USB.  That adds a layer of jitter to packet timings 
> so a Pi 3 even with GPS will never make a great NTP server.

Right, though a Pi 4 may be better.

> How stable is your temperature?  Self heating as the CPU load changes is 
> important even if the room temperature is stable.  A TCXO won't help if the 
> temperature is already stable.

The graph in ntpviz shows the jitter and temp almost perfectly correlated.

> If you have 2 identical Pi-s, you could put the TCXO on one and run a 
> side-by-side comparison and tell us what happens.

Right, that's the sort of test I had in mind.

-- 
Richard

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Richard Laager]

On 11/2/19 2:55 PM, Achim Gratz wrote:
> The one for the Ethernet controller is useless even for the seemingly
> intended application.  Replacing the main crystal with a TCXO may or may
> not help NTP performance depending on how much temperature effects are
> currently dominating it.

Right, I don't expect a benefit there. Since the board has both, if I
got one, I'd probably wire them both up just because.

>> It's a rather permanent modification, as you have to desolder the
>> stock crystals and replace them with wires going to the hat. I'm not
>> 100% sure how to order the hat yet, but they gave a price of $46 (USD)
>> via email.
> 
> You'd be much better off taking not of the injection point and then use
> a GPS timing module that can produce a clean 19.2MHz clock.  I still
> have that way down on my to-do list; although I was going to try and
> inject the clock without actually desoldering the crystal so it would
> still work even when the GPS module doesn't provide a clock.

How would it work to have two crystals connected in parallel?

-- 
Richard

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Hal Murray]

> The graph in ntpviz shows the jitter and temp almost perfectly correlated.

I'd expect drift to be tracking temperature much better than jitter.



-- 
These are my opinions.  I hate spam.




___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Achim Gratz]

Richard Laager writes:
> How would it work to have two crystals connected in parallel?

Not two crystals in parallel, a crystal plus the output of an (external)
reference oscillator.  If the two frequencies are close enough you can
injection lock the crystal oscillator, but the more likely outcome is
that you simply overdrive the input so the presence of the crystal
doesn't really matter while the external signal is present (that's
basically the mode you are using with no crystal present anyway).  I
have no idea if that is going to work (depending on how much power is
delivered to the feedback), but I can always remove the crystal if not.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf rackAttack:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[David J Taylor via time-nuts]

From: Richard Laager

On 11/2/19 1:15 AM, Hal Murray wrote:


rlaa...@wiktel.com said:
I'm curious if this would provide any meaningful improvement in system 
clock

accuracy, for NTP, if I'm already a GPS PPS hat. If there's a reasonable
chance this could be interesting, I'm thinking about ordering a couple 
and

"sacrificing" a Pi 3 and/or Pi 4.


What are your goals?  NTP server?  Neat graphs?


Closer to the latter.

The Pi 3 has Ethernet on USB.  That adds a layer of jitter to packet 
timings

so a Pi 3 even with GPS will never make a great NTP server.


Right, though a Pi 4 may be better.


How stable is your temperature?  Self heating as the CPU load changes is
important even if the room temperature is stable.  A TCXO won't help if 
the

temperature is already stable.


The graph in ntpviz shows the jitter and temp almost perfectly correlated.


If you have 2 identical Pi-s, you could put the TCXO on one and run a
side-by-side comparison and tell us what happens.


Right, that's the sort of test I had in mind.

Richard
===

Folks,

As a matter of interest, I've just compared the reported jitter on a RPi 3B, 
RPi 3B+ and RPi-4B, all PPS synced with classic NTP, all in the same room, 
but with slightly different puck antenna locations.  The lowest 5-hour 
averaged jitter was:


RasPi-15  model 3B   0.98 us
RasPi-22  model 4B   1.08 us
RasPi-18  model 3B+  1.40 us

So either the antenna location (puck inside a room) is important, or these 
results are down in the noise anyway.  RasPi-18 has a puck antenna near a 
south-facing window so I would have hoped it was the best, but apparently 
not.  RasPi-15 and -22 have antennas in a similar location (pucks with a 
magnetic base on top of a Desktop PC).  These are all Wi-Fi connected RPi 
cards so I couldn't compare the Ethernet delay.


Maybe I'll try and setup a 3B/4B Ethernet comparison sometime - I'd like to 
know the result too!


I suspect that a good outside antenna might benefit at least one of these 
RPi cards.  Even the RPi model 3 would make an adequate server for most 
users considering that devices these days are mostly connected over Wi-Fi 
rather than Ethernet!


Where better timing is preferable, I've tended to add cheap PPS boards with 
built-in GPS antennas to assembled Raspberry Pi zero units (e.g. MMDVM 
hotspots), using only the PPS from the board (as other units want the serial 
port pins), and relying on Wi-Fi for the time-of-day information.


Cheers,
David
--
SatSignal Software - Quality software for you
Web: http://www.satsignal.eu
Email: david-tay...@blueyonder.co.uk
Twitter: @gm8arv 



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[shouldbe q931]

On Tue, Nov 5, 2019 at 12:03 PM David J Taylor via time-nuts
 wrote:
>
> Folks,
>
> As a matter of interest, I've just compared the reported jitter on a RPi 3B,
> RPi 3B+ and RPi-4B, all PPS synced with classic NTP, all in the same room,
> but with slightly different puck antenna locations.


Have you considered driving multiple Pi from the same PPS source ? And
to keep their
temperature stable, keeping all three in the same enclosure ?

Cheers

Arne

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[David J Taylor via time-nuts]

From: shouldbe q931

Have you considered driving multiple Pi from the same PPS source ? And to 
keep their temperature stable, keeping all three in the same enclosure ?


Cheers
Arne
==

Yes, that would be an interesting experiment, Arne, but I don't have the 
right kit right now to do that.  I feel that the effects of temperature are 
reasonably well known.


However, I have set up a test between the RPi model B which I used in 
previous tests and a model 4B to see what effect the change of Ethernet 
architecture may have in practice.  The earlier comparison was with a 
BeagleBone Black:


 https://www.satsignal.eu/ntp/BBB-vs-RPi.html

Cheers,
David
--
SatSignal Software - Quality software for you
Web: http://www.satsignal.eu
Email: david-tay...@blueyonder.co.uk
Twitter: @gm8arv 



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[Achim Gratz]

David J Taylor via time-nuts writes:
> The graph in ntpviz shows the jitter and temp almost perfectly correlated.

Based on my experience it should really be a correlation of temperature
rate of change vs. jitter.  On my self-ovenized servers I see on average
around 200ns jitter (close to the theoretical limit of 157ns based on
the 19.2MHz clock) with the occasional peak to about three times the
baseline.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Samples for the Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#BlofeldSamplesExtra

___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.

Re: [time-nuts] Raspberry Pi TCXO Hat

[David J Taylor via time-nuts]

David J Taylor via time-nuts writes:

The graph in ntpviz shows the jitter and temp almost perfectly correlated.


Based on my experience it should really be a correlation of temperature
rate of change vs. jitter.  On my self-ovenized servers I see on average
around 200ns jitter (close to the theoretical limit of 157ns based on
the 19.2MHz clock) with the occasional peak to about three times the
baseline.

Regards,
Achim.


Achim,

I didn't write the part you quoted - blame my non-standard emailer!

What I see on the RPi is a change of offset against rate of change of 
temperature - non-ovened.  It's just as I expect.


Cheers,
David
--
SatSignal Software - Quality software for you
Web: http://www.satsignal.eu
Email: david-tay...@blueyonder.co.uk
Twitter: @gm8arv 



___
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe, go to 
http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
and follow the instructions there.