PK
Pavel Kořenský
Mon, May 13, 2024 8:04 PM
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
(signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing
is beyond repair, because capacitors are old, the whole plastic box is
already decomposing etc. etc. So, I decided to design a brand new GPSDO
with a different approach.
I bought the uBlox F9T timing GPS module:
https://www.sparkfun.com/products/18774 and I designed a relatively
simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase
comparison of 10 MHz signal from oscillator (divided by 16 with
74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if
those two signals are locked, there should be exactly 50% duty cycle
signal on the OC1 output. The sinus signals from local oscillator and
GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of
Pi Pico. Pi Pico software implements a PI regulator (loop run each
100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719)
with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
signal from GPS satellites and is pretty precise by itself. Far better
than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those
times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes
couple of ns here and there. So, with the help of my colleague, we
implemented the Kalman filter which is used for measured duty cycle
filtration in PI loop.
The whole solution is able to recover from smaller sudden phase
differences during couple of seconds. From large sudden phase
differences, the recovery time is around 30-35 seconds, because firstly
the Kalman filter is disconnected and cleared, the normal PI loop lock
fast again and after 30 seconds of lock, the Kalman filter is switched
on again.
Currently the whole design is running on breadboard, using one old
HP-10811A double-oven OCXO as a local oscillator, the power supply is a
chaotic mess of cheap DC-DC converters etc. but the measured results are
quite good. See the picture, where my design is compared to TM4313
GPSDO. I measured both devices with my Agilent 53132A which has a
non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the
whole thing, I will put the design (without local oscillator) on the PCB
and in the box. I want to use my existing Efratom FRS-C as a local
oscillator with much better short term stability than the HP-10811A. The
only difference (from electronic point of view) is that FRS-C has a
control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design,
because I do not have access to third, more precise "reference"
frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
(signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing
is beyond repair, because capacitors are old, the whole plastic box is
already decomposing etc. etc. So, I decided to design a brand new GPSDO
with a different approach.
I bought the uBlox F9T timing GPS module:
https://www.sparkfun.com/products/18774 and I designed a relatively
simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase
comparison of 10 MHz signal from oscillator (divided by 16 with
74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if
those two signals are locked, there should be exactly 50% duty cycle
signal on the OC1 output. The sinus signals from local oscillator and
GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of
Pi Pico. Pi Pico software implements a PI regulator (loop run each
100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719)
with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
signal from GPS satellites and is pretty precise by itself. Far better
than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those
times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes
couple of ns here and there. So, with the help of my colleague, we
implemented the Kalman filter which is used for measured duty cycle
filtration in PI loop.
The whole solution is able to recover from smaller sudden phase
differences during couple of seconds. From large sudden phase
differences, the recovery time is around 30-35 seconds, because firstly
the Kalman filter is disconnected and cleared, the normal PI loop lock
fast again and after 30 seconds of lock, the Kalman filter is switched
on again.
Currently the whole design is running on breadboard, using one old
HP-10811A double-oven OCXO as a local oscillator, the power supply is a
chaotic mess of cheap DC-DC converters etc. but the measured results are
quite good. See the picture, where my design is compared to TM4313
GPSDO. I measured both devices with my Agilent 53132A which has a
non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the
whole thing, I will put the design (without local oscillator) on the PCB
and in the box. I want to use my existing Efratom FRS-C as a local
oscillator with much better short term stability than the HP-10811A. The
only difference (from electronic point of view) is that FRS-C has a
control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design,
because I do not have access to third, more precise "reference"
frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
BK
Bob kb8tq
Mon, May 13, 2024 9:14 PM
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>
> Hello,
>
> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
>
> At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
>
> I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
>
> The whole design works as follows:
>
> The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
>
> The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
>
> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
>
> But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
>
> The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
>
> Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
>
>
> In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
>
> The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
>
> What do you think about my design ? Did I missed something ?
>
> PavelK
>
>
>
>
> <Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
PK
Pavel Kořenský
Mon, May 13, 2024 9:35 PM
Hello,
I know about this problem. This is the reason why I am using the 1.25
MHz from F9T.
Last year, my friend did some serious measurements od various F9T
frequencies. Because he is working in CERN, he has a possibility to use
hydrogen maser disciplined by cesium fountain as a primary frequency
standard and compare various sources to this primary frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
I am further trying to minimise such errors with averaging inside PI
loop. Also, because the duty cycle measurement in Pi Pico is using the
free running Pico 125 MHz internal clock as a source for incrementing
the PIO counters, the results are always a bit averaged (or smudged) and
this also helps.
BTW, it seems that the picture which I attached to my former post was
too big to be sent to the mail list, so here is the link:
https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hello,
I know about this problem. This is the reason why I am using the 1.25
MHz from F9T.
Last year, my friend did some serious measurements od various F9T
frequencies. Because he is working in CERN, he has a possibility to use
hydrogen maser disciplined by cesium fountain as a primary frequency
standard and compare various sources to this primary frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
I am further trying to minimise such errors with averaging inside PI
loop. Also, because the duty cycle measurement in Pi Pico is using the
free running Pico 125 MHz internal clock as a source for incrementing
the PIO counters, the results are always a bit averaged (or smudged) and
this also helps.
BTW, it seems that the picture which I attached to my former post was
too big to be sent to the mail list, so here is the link:
https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
> Hi
>
> What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
>
> The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
>
> Does any of this matter? It very much depends on your application and objectives.
>
> Fun !!!
>
> Bob
>
>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>
>> Hello,
>>
>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
>>
>> At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
>>
>> I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
>>
>> The whole design works as follows:
>>
>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
>>
>> The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
>>
>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
>>
>> But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
>>
>> The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
>>
>> Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
>>
>>
>> In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
>>
>> The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
>>
>> What do you think about my design ? Did I missed something ?
>>
>> PavelK
>>
>>
>>
>>
>> <Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
>> time-nuts mailing list -- time-nuts@lists.febo.com
>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
ZC
Zdenek Chaloupka
Tue, May 14, 2024 5:44 PM
Hi,
the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
Best regards
Zdenek
On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi,
the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
Best regards
Zdenek
> On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>
> Hello,
>
> I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
> Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
> And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
>
> I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
>
> BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
>
> PavelK
>
> Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
>> Hi
>>
>> What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
>>
>> The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
>>
>> Does any of this matter? It very much depends on your application and objectives.
>>
>> Fun !!!
>>
>> Bob
>>
>>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>>
>>> Hello,
>>>
>>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
>>>
>>> At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
>>>
>>> I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
>>>
>>> The whole design works as follows:
>>>
>>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
>>>
>>> The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
>>>
>>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
>>>
>>> But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
>>>
>>> The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
>>>
>>> Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
>>>
>>>
>>> In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
>>>
>>> The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
>>>
>>> What do you think about my design ? Did I missed something ?
>>>
>>> PavelK
>>>
>>>
>>>
>>>
>>> <Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
MD
Magnus Danielson
Tue, May 14, 2024 7:29 PM
Hi Pavel,
Do you happen to have a link to a paper or something that present the
results of the measurements of F9T you refer to?
Cheers,
Magnus
On 2024-05-13 23:35, Pavel Kořenský via time-nuts wrote:
Hello,
I know about this problem. This is the reason why I am using the 1.25
MHz from F9T.
Last year, my friend did some serious measurements od various F9T
frequencies. Because he is working in CERN, he has a possibility to
use hydrogen maser disciplined by cesium fountain as a primary
frequency standard and compare various sources to this primary
frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for
F9T.
I am further trying to minimise such errors with averaging inside PI
loop. Also, because the duty cycle measurement in Pi Pico is using the
free running Pico 125 MHz internal clock as a source for incrementing
the PIO counters, the results are always a bit averaged (or smudged)
and this also helps.
BTW, it seems that the picture which I attached to my former post was
too big to be sent to the mail list, so here is the link:
https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging
bridges”. They do not lock the LO on the F9T to the GPS signal. They
simply correct it to the “nearest edge” when they can. How well this
works and what sort of issues it creates is a “that depends” sort of
thing. You may find that some output frequencies are “nicer” than
others.
The result (at any output frequency) is that you can get a fairly
long “wrong phase” out of the device. It’s not so much a phase step
as a phase lock at the wrong location. The classic answer to this is
to look at the “sawtooth correction” information out of the F9T.
Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and
objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts
time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
(signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole
thing is beyond repair, because capacitors are old, the whole
plastic box is already decomposing etc. etc. So, I decided to design
a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module:
https://www.sparkfun.com/products/18774 and I designed a relatively
simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase
comparison of 10 MHz signal from oscillator (divided by 16 with
74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally,
if those two signals are locked, there should be exactly 50% duty
cycle signal on the OC1 output. The sinus signals from local
oscillator and GPS are connected to the 4046 via AC-coupled Schmitt
buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules
of Pi Pico. Pi Pico software implements a PI regulator (loop run
each 100ms) and is steering a local oscillator with 20bit SPI-DAC (
MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
signal from GPS satellites and is pretty precise by itself. Far
better than was the 1PPS from my old Motorola Oncore UT+ back in
1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and
spikes couple of ns here and there. So, with the help of my
colleague, we implemented the Kalman filter which is used for
measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase
differences during couple of seconds. From large sudden phase
differences, the recovery time is around 30-35 seconds, because
firstly the Kalman filter is disconnected and cleared, the normal PI
loop lock fast again and after 30 seconds of lock, the Kalman filter
is switched on again.
Currently the whole design is running on breadboard, using one old
HP-10811A double-oven OCXO as a local oscillator, the power supply
is a chaotic mess of cheap DC-DC converters etc. but the measured
results are quite good. See the picture, where my design is compared
to TM4313 GPSDO. I measured both devices with my Agilent 53132A
which has a non-disciplined Rubidium standard (Efratom FRS-C) as a
timebase.
In the next phase, I will build a decent linear power supply for the
whole thing, I will put the design (without local oscillator) on the
PCB and in the box. I want to use my existing Efratom FRS-C as a
local oscillator with much better short term stability than the
HP-10811A. The only difference (from electronic point of view) is
that FRS-C has a control voltage 0-5V and HP-10811A has a control
voltage -5V to +5V.
The only thing which I do not know is how to test the final design,
because I do not have access to third, more precise "reference"
frequency in the form of either cesium frequency standard or
hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13
202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi Pavel,
Do you happen to have a link to a paper or something that present the
results of the measurements of F9T you refer to?
Cheers,
Magnus
On 2024-05-13 23:35, Pavel Kořenský via time-nuts wrote:
> Hello,
>
> I know about this problem. This is the reason why I am using the 1.25
> MHz from F9T.
> Last year, my friend did some serious measurements od various F9T
> frequencies. Because he is working in CERN, he has a possibility to
> use hydrogen maser disciplined by cesium fountain as a primary
> frequency standard and compare various sources to this primary
> frequency standard.
> And the result was, that 1.25 MHz is one of the "best" frequencies for
> F9T.
>
> I am further trying to minimise such errors with averaging inside PI
> loop. Also, because the duty cycle measurement in Pi Pico is using the
> free running Pico 125 MHz internal clock as a source for incrementing
> the PIO counters, the results are always a bit averaged (or smudged)
> and this also helps.
>
> BTW, it seems that the picture which I attached to my former post was
> too big to be sent to the mail list, so here is the link:
> https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
>
> PavelK
>
> Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
>> Hi
>>
>> What you will run into on the F9T output are often called “hanging
>> bridges”. They do not lock the LO on the F9T to the GPS signal. They
>> simply correct it to the “nearest edge” when they can. How well this
>> works and what sort of issues it creates is a “that depends” sort of
>> thing. You may find that some output frequencies are “nicer” than
>> others.
>>
>> The result (at any output frequency) is that you can get a fairly
>> long “wrong phase” out of the device. It’s not so much a phase step
>> as a phase lock at the wrong location. The classic answer to this is
>> to look at the “sawtooth correction” information out of the F9T.
>> Getting that to match up with the HF synthesized output can be tricky.
>>
>> Does any of this matter? It very much depends on your application and
>> objectives.
>>
>> Fun !!!
>>
>> Bob
>>
>>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts
>>> <time-nuts@lists.febo.com> wrote:
>>>
>>> Hello,
>>>
>>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
>>> Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
>>> (signal generator, counter etc.).
>>>
>>> At the beginning of 2024, the unit died an I found that the whole
>>> thing is beyond repair, because capacitors are old, the whole
>>> plastic box is already decomposing etc. etc. So, I decided to design
>>> a brand new GPSDO with a different approach.
>>>
>>> I bought the uBlox F9T timing GPS module:
>>> https://www.sparkfun.com/products/18774 and I designed a relatively
>>> simple circuit with 74HCT4046A and Pi Pico.
>>>
>>> The whole design works as follows:
>>>
>>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase
>>> comparison of 10 MHz signal from oscillator (divided by 16 with
>>> 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally,
>>> if those two signals are locked, there should be exactly 50% duty
>>> cycle signal on the OC1 output. The sinus signals from local
>>> oscillator and GPS are connected to the 4046 via AC-coupled Schmitt
>>> buffers (74LVC1G97).
>>>
>>> The OC1 output duty cycle is measured constantly by two PIO modules
>>> of Pi Pico. Pi Pico software implements a PI regulator (loop run
>>> each 100ms) and is steering a local oscillator with 20bit SPI-DAC (
>>> MAX5719) with a precision 5V reference (LT1021).
>>>
>>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
>>> signal from GPS satellites and is pretty precise by itself. Far
>>> better than was the 1PPS from my old Motorola Oncore UT+ back in
>>> 1999. At those times, there was a SA in GPS signal.
>>>
>>> But the GPS signal is not ideal, there are phase differences and
>>> spikes couple of ns here and there. So, with the help of my
>>> colleague, we implemented the Kalman filter which is used for
>>> measured duty cycle filtration in PI loop.
>>>
>>> The whole solution is able to recover from smaller sudden phase
>>> differences during couple of seconds. From large sudden phase
>>> differences, the recovery time is around 30-35 seconds, because
>>> firstly the Kalman filter is disconnected and cleared, the normal PI
>>> loop lock fast again and after 30 seconds of lock, the Kalman filter
>>> is switched on again.
>>>
>>> Currently the whole design is running on breadboard, using one old
>>> HP-10811A double-oven OCXO as a local oscillator, the power supply
>>> is a chaotic mess of cheap DC-DC converters etc. but the measured
>>> results are quite good. See the picture, where my design is compared
>>> to TM4313 GPSDO. I measured both devices with my Agilent 53132A
>>> which has a non-disciplined Rubidium standard (Efratom FRS-C) as a
>>> timebase.
>>>
>>>
>>> In the next phase, I will build a decent linear power supply for the
>>> whole thing, I will put the design (without local oscillator) on the
>>> PCB and in the box. I want to use my existing Efratom FRS-C as a
>>> local oscillator with much better short term stability than the
>>> HP-10811A. The only difference (from electronic point of view) is
>>> that FRS-C has a control voltage 0-5V and HP-10811A has a control
>>> voltage -5V to +5V.
>>>
>>> The only thing which I do not know is how to test the final design,
>>> because I do not have access to third, more precise "reference"
>>> frequency in the form of either cesium frequency standard or
>>> hydrogen maser.
>>>
>>> What do you think about my design ? Did I missed something ?
>>>
>>> PavelK
>>>
>>>
>>>
>>>
>>> <Snímek obrazovky 2024-05-13
>>> 202444.png>_______________________________________________
>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
MD
Magnus Danielson
Tue, May 14, 2024 7:32 PM
Hi Zdenek,
How is edges of the 1.25 MHz corrected? Often?
Just a straight divided down of 125 MHz gives that clock, but not the
GNSS result, so some form of hand-over I expect to occurr.
Cheers,
Magnus
On 2024-05-14 19:44, Zdenek Chaloupka via time-nuts wrote:
Hi,
the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
Best regards
Zdenek
On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi Zdenek,
How is edges of the 1.25 MHz corrected? Often?
Just a straight divided down of 125 MHz gives that clock, but not the
GNSS result, so some form of hand-over I expect to occurr.
Cheers,
Magnus
On 2024-05-14 19:44, Zdenek Chaloupka via time-nuts wrote:
> Hi,
>
> the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
>
> Best regards
> Zdenek
>
>
>> On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>
>> Hello,
>>
>> I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
>> Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
>> And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
>>
>> I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
>>
>> BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
>>
>> PavelK
>>
>> Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
>>> Hi
>>>
>>> What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
>>>
>>> The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
>>>
>>> Does any of this matter? It very much depends on your application and objectives.
>>>
>>> Fun !!!
>>>
>>> Bob
>>>
>>>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>>>
>>>> Hello,
>>>>
>>>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
>>>>
>>>> At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
>>>>
>>>> I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
>>>>
>>>> The whole design works as follows:
>>>>
>>>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
>>>>
>>>> The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
>>>>
>>>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
>>>>
>>>> But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
>>>>
>>>> The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
>>>>
>>>> Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
>>>>
>>>>
>>>> In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
>>>>
>>>> The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
>>>>
>>>> What do you think about my design ? Did I missed something ?
>>>>
>>>> PavelK
>>>>
>>>>
>>>>
>>>>
>>>> <Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
>>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>> _______________________________________________
>> time-nuts mailing list -- time-nuts@lists.febo.com
>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
ZC
Zdenek Chaloupka
Tue, May 14, 2024 8:15 PM
Hi Magnus,
the receiver corrects the output frequency/phase every millisecond. The fact that there is a factor of 100 to internal source clock helps to have a cleaner frequency, I suspect.
Best regards
Zdenek
On 14 May 2024, at 20:32, Magnus Danielson via time-nuts time-nuts@lists.febo.com wrote:
Hi Zdenek,
How is edges of the 1.25 MHz corrected? Often?
Just a straight divided down of 125 MHz gives that clock, but not the GNSS result, so some form of hand-over I expect to occurr.
Cheers,
Magnus
On 2024-05-14 19:44, Zdenek Chaloupka via time-nuts wrote:
Hi,
the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
Best regards
Zdenek
On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi Magnus,
the receiver corrects the output frequency/phase every millisecond. The fact that there is a factor of 100 to internal source clock helps to have a cleaner frequency, I suspect.
Best regards
Zdenek
> On 14 May 2024, at 20:32, Magnus Danielson via time-nuts <time-nuts@lists.febo.com> wrote:
>
> Hi Zdenek,
>
> How is edges of the 1.25 MHz corrected? Often?
>
> Just a straight divided down of 125 MHz gives that clock, but not the GNSS result, so some form of hand-over I expect to occurr.
>
> Cheers,
> Magnus
>
> On 2024-05-14 19:44, Zdenek Chaloupka via time-nuts wrote:
>> Hi,
>>
>> the reason why 1.25 MHz is so clean, is that internal clock for 1PPS/freq. output is based on internal 125MHz clock (that is why there is 8ns quantization error for 1PPS). Sawtooth correction information is of no use at 1.25 MHz, as it is output only for some tens of hertz, if I remember correctly.
>>
>> Best regards
>> Zdenek
>>
>>
>>> On 13 May 2024, at 22:35, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>>
>>> Hello,
>>>
>>> I know about this problem. This is the reason why I am using the 1.25 MHz from F9T.
>>> Last year, my friend did some serious measurements od various F9T frequencies. Because he is working in CERN, he has a possibility to use hydrogen maser disciplined by cesium fountain as a primary frequency standard and compare various sources to this primary frequency standard.
>>> And the result was, that 1.25 MHz is one of the "best" frequencies for F9T.
>>>
>>> I am further trying to minimise such errors with averaging inside PI loop. Also, because the duty cycle measurement in Pi Pico is using the free running Pico 125 MHz internal clock as a source for incrementing the PIO counters, the results are always a bit averaged (or smudged) and this also helps.
>>>
>>> BTW, it seems that the picture which I attached to my former post was too big to be sent to the mail list, so here is the link: https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
>>>
>>> PavelK
>>>
>>> Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
>>>> Hi
>>>>
>>>> What you will run into on the F9T output are often called “hanging bridges”. They do not lock the LO on the F9T to the GPS signal. They simply correct it to the “nearest edge” when they can. How well this works and what sort of issues it creates is a “that depends” sort of thing. You may find that some output frequencies are “nicer” than others.
>>>>
>>>> The result (at any output frequency) is that you can get a fairly long “wrong phase” out of the device. It’s not so much a phase step as a phase lock at the wrong location. The classic answer to this is to look at the “sawtooth correction” information out of the F9T. Getting that to match up with the HF synthesized output can be tricky.
>>>>
>>>> Does any of this matter? It very much depends on your application and objectives.
>>>>
>>>> Fun !!!
>>>>
>>>> Bob
>>>>
>>>>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts <time-nuts@lists.febo.com> wrote:
>>>>>
>>>>> Hello,
>>>>>
>>>>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop (signal generator, counter etc.).
>>>>>
>>>>> At the beginning of 2024, the unit died an I found that the whole thing is beyond repair, because capacitors are old, the whole plastic box is already decomposing etc. etc. So, I decided to design a brand new GPSDO with a different approach.
>>>>>
>>>>> I bought the uBlox F9T timing GPS module: https://www.sparkfun.com/products/18774 and I designed a relatively simple circuit with 74HCT4046A and Pi Pico.
>>>>>
>>>>> The whole design works as follows:
>>>>>
>>>>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase comparison of 10 MHz signal from oscillator (divided by 16 with 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally, if those two signals are locked, there should be exactly 50% duty cycle signal on the OC1 output. The sinus signals from local oscillator and GPS are connected to the 4046 via AC-coupled Schmitt buffers (74LVC1G97).
>>>>>
>>>>> The OC1 output duty cycle is measured constantly by two PIO modules of Pi Pico. Pi Pico software implements a PI regulator (loop run each 100ms) and is steering a local oscillator with 20bit SPI-DAC ( MAX5719) with a precision 5V reference (LT1021).
>>>>>
>>>>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS signal from GPS satellites and is pretty precise by itself. Far better than was the 1PPS from my old Motorola Oncore UT+ back in 1999. At those times, there was a SA in GPS signal.
>>>>>
>>>>> But the GPS signal is not ideal, there are phase differences and spikes couple of ns here and there. So, with the help of my colleague, we implemented the Kalman filter which is used for measured duty cycle filtration in PI loop.
>>>>>
>>>>> The whole solution is able to recover from smaller sudden phase differences during couple of seconds. From large sudden phase differences, the recovery time is around 30-35 seconds, because firstly the Kalman filter is disconnected and cleared, the normal PI loop lock fast again and after 30 seconds of lock, the Kalman filter is switched on again.
>>>>>
>>>>> Currently the whole design is running on breadboard, using one old HP-10811A double-oven OCXO as a local oscillator, the power supply is a chaotic mess of cheap DC-DC converters etc. but the measured results are quite good. See the picture, where my design is compared to TM4313 GPSDO. I measured both devices with my Agilent 53132A which has a non-disciplined Rubidium standard (Efratom FRS-C) as a timebase.
>>>>>
>>>>>
>>>>> In the next phase, I will build a decent linear power supply for the whole thing, I will put the design (without local oscillator) on the PCB and in the box. I want to use my existing Efratom FRS-C as a local oscillator with much better short term stability than the HP-10811A. The only difference (from electronic point of view) is that FRS-C has a control voltage 0-5V and HP-10811A has a control voltage -5V to +5V.
>>>>>
>>>>> The only thing which I do not know is how to test the final design, because I do not have access to third, more precise "reference" frequency in the form of either cesium frequency standard or hydrogen maser.
>>>>>
>>>>> What do you think about my design ? Did I missed something ?
>>>>>
>>>>> PavelK
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> <Snímek obrazovky 2024-05-13 202444.png>_______________________________________________
>>>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>>> _______________________________________________
>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>> _______________________________________________
>> time-nuts mailing list -- time-nuts@lists.febo.com
>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
PK
Pavel Kořenský
Tue, May 14, 2024 8:25 PM
Hi Magnus,
I will ask him, but I do not know about any publicly available paper. We
just talked about this topic some time ago and he gave me couple of tips
for "good frequencies".
One of his students recently finished the Mgr. level diploma thesis in
which the student constructed the GPSDOCXO with F9T and this student is
also using the 1.25MHz. So, maybe there will be some comparison in this
thesis. The diploma thesis should be freely available as soon as the
student will finish final exam etc. Maybe in couple of weeks or
something like that.
This student is using different design of PLL, with precision A/D
converter measuring the voltage on double balanced mixer, but the used
GPS receiver is the same, so I suppose that the 1.25 MHz is a "good"
frequency for overall usage.
PavelK
Dne 14.05.2024 v 21:29 Magnus Danielson via time-nuts napsal(a):
Hi Pavel,
Do you happen to have a link to a paper or something that present the
results of the measurements of F9T you refer to?
Cheers,
Magnus
On 2024-05-13 23:35, Pavel Kořenský via time-nuts wrote:
Hello,
I know about this problem. This is the reason why I am using the 1.25
MHz from F9T.
Last year, my friend did some serious measurements od various F9T
frequencies. Because he is working in CERN, he has a possibility to
use hydrogen maser disciplined by cesium fountain as a primary
frequency standard and compare various sources to this primary
frequency standard.
And the result was, that 1.25 MHz is one of the "best" frequencies
for F9T.
I am further trying to minimise such errors with averaging inside PI
loop. Also, because the duty cycle measurement in Pi Pico is using
the free running Pico 125 MHz internal clock as a source for
incrementing the PIO counters, the results are always a bit averaged
(or smudged) and this also helps.
BTW, it seems that the picture which I attached to my former post was
too big to be sent to the mail list, so here is the link:
https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
PavelK
Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
Hi
What you will run into on the F9T output are often called “hanging
bridges”. They do not lock the LO on the F9T to the GPS signal. They
simply correct it to the “nearest edge” when they can. How well this
works and what sort of issues it creates is a “that depends” sort of
thing. You may find that some output frequencies are “nicer” than
others.
The result (at any output frequency) is that you can get a fairly
long “wrong phase” out of the device. It’s not so much a phase step
as a phase lock at the wrong location. The classic answer to this is
to look at the “sawtooth correction” information out of the F9T.
Getting that to match up with the HF synthesized output can be tricky.
Does any of this matter? It very much depends on your application
and objectives.
Fun !!!
Bob
On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts
time-nuts@lists.febo.com wrote:
Hello,
I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
(signal generator, counter etc.).
At the beginning of 2024, the unit died an I found that the whole
thing is beyond repair, because capacitors are old, the whole
plastic box is already decomposing etc. etc. So, I decided to
design a brand new GPSDO with a different approach.
I bought the uBlox F9T timing GPS module:
https://www.sparkfun.com/products/18774 and I designed a relatively
simple circuit with 74HCT4046A and Pi Pico.
The whole design works as follows:
The 4046 is using its phase comparator OC1 (exclusive-or) for phase
comparison of 10 MHz signal from oscillator (divided by 16 with
74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally,
if those two signals are locked, there should be exactly 50% duty
cycle signal on the OC1 output. The sinus signals from local
oscillator and GPS are connected to the 4046 via AC-coupled Schmitt
buffers (74LVC1G97).
The OC1 output duty cycle is measured constantly by two PIO modules
of Pi Pico. Pi Pico software implements a PI regulator (loop run
each 100ms) and is steering a local oscillator with 20bit SPI-DAC (
MAX5719) with a precision 5V reference (LT1021).
The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
signal from GPS satellites and is pretty precise by itself. Far
better than was the 1PPS from my old Motorola Oncore UT+ back in
1999. At those times, there was a SA in GPS signal.
But the GPS signal is not ideal, there are phase differences and
spikes couple of ns here and there. So, with the help of my
colleague, we implemented the Kalman filter which is used for
measured duty cycle filtration in PI loop.
The whole solution is able to recover from smaller sudden phase
differences during couple of seconds. From large sudden phase
differences, the recovery time is around 30-35 seconds, because
firstly the Kalman filter is disconnected and cleared, the normal
PI loop lock fast again and after 30 seconds of lock, the Kalman
filter is switched on again.
Currently the whole design is running on breadboard, using one old
HP-10811A double-oven OCXO as a local oscillator, the power supply
is a chaotic mess of cheap DC-DC converters etc. but the measured
results are quite good. See the picture, where my design is
compared to TM4313 GPSDO. I measured both devices with my Agilent
53132A which has a non-disciplined Rubidium standard (Efratom
FRS-C) as a timebase.
In the next phase, I will build a decent linear power supply for
the whole thing, I will put the design (without local oscillator)
on the PCB and in the box. I want to use my existing Efratom FRS-C
as a local oscillator with much better short term stability than
the HP-10811A. The only difference (from electronic point of view)
is that FRS-C has a control voltage 0-5V and HP-10811A has a
control voltage -5V to +5V.
The only thing which I do not know is how to test the final design,
because I do not have access to third, more precise "reference"
frequency in the form of either cesium frequency standard or
hydrogen maser.
What do you think about my design ? Did I missed something ?
PavelK
<Snímek obrazovky 2024-05-13
202444.png>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi Magnus,
I will ask him, but I do not know about any publicly available paper. We
just talked about this topic some time ago and he gave me couple of tips
for "good frequencies".
One of his students recently finished the Mgr. level diploma thesis in
which the student constructed the GPSDOCXO with F9T and this student is
also using the 1.25MHz. So, maybe there will be some comparison in this
thesis. The diploma thesis should be freely available as soon as the
student will finish final exam etc. Maybe in couple of weeks or
something like that.
This student is using different design of PLL, with precision A/D
converter measuring the voltage on double balanced mixer, but the used
GPS receiver is the same, so I suppose that the 1.25 MHz is a "good"
frequency for overall usage.
PavelK
Dne 14.05.2024 v 21:29 Magnus Danielson via time-nuts napsal(a):
> Hi Pavel,
>
> Do you happen to have a link to a paper or something that present the
> results of the measurements of F9T you refer to?
>
> Cheers,
> Magnus
>
> On 2024-05-13 23:35, Pavel Kořenský via time-nuts wrote:
>> Hello,
>>
>> I know about this problem. This is the reason why I am using the 1.25
>> MHz from F9T.
>> Last year, my friend did some serious measurements od various F9T
>> frequencies. Because he is working in CERN, he has a possibility to
>> use hydrogen maser disciplined by cesium fountain as a primary
>> frequency standard and compare various sources to this primary
>> frequency standard.
>> And the result was, that 1.25 MHz is one of the "best" frequencies
>> for F9T.
>>
>> I am further trying to minimise such errors with averaging inside PI
>> loop. Also, because the duty cycle measurement in Pi Pico is using
>> the free running Pico 125 MHz internal clock as a source for
>> incrementing the PIO counters, the results are always a bit averaged
>> (or smudged) and this also helps.
>>
>> BTW, it seems that the picture which I attached to my former post was
>> too big to be sent to the mail list, so here is the link:
>> https://www.rajce.idnes.cz/pkorensky/album/agilent-53132a/1635108363
>>
>> PavelK
>>
>> Dne 13.05.2024 v 23:14 Bob kb8tq napsal(a):
>>> Hi
>>>
>>> What you will run into on the F9T output are often called “hanging
>>> bridges”. They do not lock the LO on the F9T to the GPS signal. They
>>> simply correct it to the “nearest edge” when they can. How well this
>>> works and what sort of issues it creates is a “that depends” sort of
>>> thing. You may find that some output frequencies are “nicer” than
>>> others.
>>>
>>> The result (at any output frequency) is that you can get a fairly
>>> long “wrong phase” out of the device. It’s not so much a phase step
>>> as a phase lock at the wrong location. The classic answer to this is
>>> to look at the “sawtooth correction” information out of the F9T.
>>> Getting that to match up with the HF synthesized output can be tricky.
>>>
>>> Does any of this matter? It very much depends on your application
>>> and objectives.
>>>
>>> Fun !!!
>>>
>>> Bob
>>>
>>>> On May 13, 2024, at 4:04 PM, Pavel Kořenský via time-nuts
>>>> <time-nuts@lists.febo.com> wrote:
>>>>
>>>> Hello,
>>>>
>>>> I used a modified Brooks Shera GPSDO (with some Gollidge OCXO an
>>>> Motorola OnCore UT+ GPS) since 1999 as a timebase for my workshop
>>>> (signal generator, counter etc.).
>>>>
>>>> At the beginning of 2024, the unit died an I found that the whole
>>>> thing is beyond repair, because capacitors are old, the whole
>>>> plastic box is already decomposing etc. etc. So, I decided to
>>>> design a brand new GPSDO with a different approach.
>>>>
>>>> I bought the uBlox F9T timing GPS module:
>>>> https://www.sparkfun.com/products/18774 and I designed a relatively
>>>> simple circuit with 74HCT4046A and Pi Pico.
>>>>
>>>> The whole design works as follows:
>>>>
>>>> The 4046 is using its phase comparator OC1 (exclusive-or) for phase
>>>> comparison of 10 MHz signal from oscillator (divided by 16 with
>>>> 74HCT393) and 1.25 MHz signal from F9T GPS (divided by 2). Ideally,
>>>> if those two signals are locked, there should be exactly 50% duty
>>>> cycle signal on the OC1 output. The sinus signals from local
>>>> oscillator and GPS are connected to the 4046 via AC-coupled Schmitt
>>>> buffers (74LVC1G97).
>>>>
>>>> The OC1 output duty cycle is measured constantly by two PIO modules
>>>> of Pi Pico. Pi Pico software implements a PI regulator (loop run
>>>> each 100ms) and is steering a local oscillator with 20bit SPI-DAC (
>>>> MAX5719) with a precision 5V reference (LT1021).
>>>>
>>>> The GPS output signal 1.25 MHz is in F9T internally locked to 1PPS
>>>> signal from GPS satellites and is pretty precise by itself. Far
>>>> better than was the 1PPS from my old Motorola Oncore UT+ back in
>>>> 1999. At those times, there was a SA in GPS signal.
>>>>
>>>> But the GPS signal is not ideal, there are phase differences and
>>>> spikes couple of ns here and there. So, with the help of my
>>>> colleague, we implemented the Kalman filter which is used for
>>>> measured duty cycle filtration in PI loop.
>>>>
>>>> The whole solution is able to recover from smaller sudden phase
>>>> differences during couple of seconds. From large sudden phase
>>>> differences, the recovery time is around 30-35 seconds, because
>>>> firstly the Kalman filter is disconnected and cleared, the normal
>>>> PI loop lock fast again and after 30 seconds of lock, the Kalman
>>>> filter is switched on again.
>>>>
>>>> Currently the whole design is running on breadboard, using one old
>>>> HP-10811A double-oven OCXO as a local oscillator, the power supply
>>>> is a chaotic mess of cheap DC-DC converters etc. but the measured
>>>> results are quite good. See the picture, where my design is
>>>> compared to TM4313 GPSDO. I measured both devices with my Agilent
>>>> 53132A which has a non-disciplined Rubidium standard (Efratom
>>>> FRS-C) as a timebase.
>>>>
>>>>
>>>> In the next phase, I will build a decent linear power supply for
>>>> the whole thing, I will put the design (without local oscillator)
>>>> on the PCB and in the box. I want to use my existing Efratom FRS-C
>>>> as a local oscillator with much better short term stability than
>>>> the HP-10811A. The only difference (from electronic point of view)
>>>> is that FRS-C has a control voltage 0-5V and HP-10811A has a
>>>> control voltage -5V to +5V.
>>>>
>>>> The only thing which I do not know is how to test the final design,
>>>> because I do not have access to third, more precise "reference"
>>>> frequency in the form of either cesium frequency standard or
>>>> hydrogen maser.
>>>>
>>>> What do you think about my design ? Did I missed something ?
>>>>
>>>> PavelK
>>>>
>>>>
>>>>
>>>>
>>>> <Snímek obrazovky 2024-05-13
>>>> 202444.png>_______________________________________________
>>>> time-nuts mailing list -- time-nuts@lists.febo.com
>>>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
>>
>> _______________________________________________
>> time-nuts mailing list -- time-nuts@lists.febo.com
>> To unsubscribe send an email to time-nuts-leave@lists.febo.com
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
F
fff@spukhafte.net
Tue, May 21, 2024 7:06 PM
Hi Pavel,
I’m curious if you considered using the uBlox M8F, which is designed to discipline external oscillators? You can get corrective data either through any of its interfaces, or from an external DAC.
Fred
Hi Pavel,
I’m curious if you considered using the uBlox M8F, which is designed to discipline external oscillators? You can get corrective data either through any of its interfaces, or from an external DAC.\
\
Fred
PK
Pavel Kořenský
Tue, May 21, 2024 9:53 PM
Hi Fred,
no, I did not considered this particular module.
I looked briefly at the datasheet some time ago, but this module seems
to be oriented for mobile/GSM tasks. It has a weird output frequency and
the DAC SPI control allows to control max. 16bit DAC.
And 16bit DAC is really not ideal for a Rb standard which has the EFC
input which is able to tune only +/- 1 x 10^-9 with voltage swing 0-5V
range and long term drift of +/-5 x 10^-10/year.
The F9T which I am using is not cheap (actually, it is expensive as
hell), but has two programmable outputs and I am using the 20bit DAC
with the Pi Pico. Of course, there will always be some noise on the DAC,
but extra 4 bits of precision allows some fine tuning.
PavelK
Dne 21.05.2024 v 21:06 Fred Fierling via time-nuts napsal(a):
Hi Pavel,
I’m curious if you considered using the uBlox M8F, which is designed to discipline external oscillators? You can get corrective data either through any of its interfaces, or from an external DAC.
Fred
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi Fred,
no, I did not considered this particular module.
I looked briefly at the datasheet some time ago, but this module seems
to be oriented for mobile/GSM tasks. It has a weird output frequency and
the DAC SPI control allows to control max. 16bit DAC.
And 16bit DAC is really not ideal for a Rb standard which has the EFC
input which is able to tune only +/- 1 x 10^-9 with voltage swing 0-5V
range and long term drift of +/-5 x 10^-10/year.
The F9T which I am using is not cheap (actually, it is expensive as
hell), but has two programmable outputs and I am using the 20bit DAC
with the Pi Pico. Of course, there will always be some noise on the DAC,
but extra 4 bits of precision allows some fine tuning.
PavelK
Dne 21.05.2024 v 21:06 Fred Fierling via time-nuts napsal(a):
> Hi Pavel,
>
> I’m curious if you considered using the uBlox M8F, which is designed to discipline external oscillators? You can get corrective data either through any of its interfaces, or from an external DAC.\
> \
> Fred
> _______________________________________________
> time-nuts mailing list -- time-nuts@lists.febo.com
> To unsubscribe send an email to time-nuts-leave@lists.febo.com
