[time-nuts] Re: FPGA-based GPS NTP server project

[HÃ¥kan T Johansson]

Hi,

thanks for all the tips!  With some NEO-7M and NEO-6M boards together with 
a 0.5 ns sampling of the PPS and sawtooth correction message it is now 
able to reach a jitter of about 2 ns of the PPS vs. the tracking.

Another thing came up:

For monitoring, I have found it quite useful to have an estimate of the 
current tracking resolution in the NTP responses.  However, the root 
dispersion field does not have resolution enough (smallest non-0 value is 
15.3 us) and thus only give useful reports when tracking is way off, e.g. 
during initial lock.  So I am (ab)using the precision field to report 
log_2 of a running average of the absolute differences between the actual 
and expected NTP timestamps at the PPS signals.  With the previous 8 ns 
sampling without sawtooth correction it reached around -26 to -27, and now 
instead between -29 and -30.

Are there adverse effects of having a dynamically varying NTP precision 
field?

Cheers,
HÃ¥kan




On Sun, 12 Feb 2023, Bob Camp wrote:

> Hi
>
> You don’t have to spend what the F9T’s cost simply to knock down the jitter. There are a
> number of uBlox modules that have sawtooth correction data and cost < $40 mounted on
> on a board.
>
> Bob
>
>> On Feb 12, 2023, at 6:51 AM, HÃ¥kan T Johansson <f96hajo at chalmers.se> wrote:
>>
>>
>> Hi,
>>
>> perhaps the uBlox ZED-F9T would be a suitable choice, on a RCB-F9T board?
>> That looks like 8 ns jitter, and the UBX-TIM-TP message should give the sawtooth.
>>
>> A 0.5 ns sampler I actually have on the FPGA.
>>
>> Cheers,
>> HÃ¥kan
>>
>>
>> On Sun, 12 Feb 2023, Ed Marciniak wrote:
>>
>>> Your mediatek GPS receiver is probably a really lousy pick for low jitter.
>>> There are options from Navspark like the NS-T, other receiving chips/modules
>>> from the same, as well as some uBlox parts that would do better.
>>> There’s options with under 20ns jitter on those, and if memory serves
>>> there’s one that’s available that with a good sky view and antenna might be
>>> more like 8ns jitter.
>>> Get Outlook for iOS
>>> ____________________________________________________________________________
>>> From: Bob kb8tq via time-nuts <time-nuts at lists.febo.com>
>>> Sent: Saturday, February 11, 2023 4:38:23 PM
>>> To: Discussion of precise time and frequency measurement
>>> <time-nuts at lists.febo.com>
>>> Cc: HÃ¥kan T Johansson <f96hajo at chalmers.se>; Bob kb8tq <kb8tq at n1k.org>
>>> Subject: [time-nuts] Re: FPGA-based GPS NTP server project
>>> Hi
>>> If you want to get the short term jitter down to around a second or so,
>>> set it up with a GPS module that puts out a sawtooth correction. You
>>> also need to get the Arty A7 measuring at the 1 ns level. That’s do-able
>>> without insane amounts of effort.
>>> Bob
>>>> On Feb 11, 2023, at 4:09 AM, HÃ¥kan T Johansson via time-nuts
>>> <time-nuts at lists.febo.com> wrote:
>>>>
>>>>
>>>> Hi,
>>>>
>>>> I have a NTP server project based on a Digilent Arty A7 FPGA board with a
>>> MediaTek GPS MT3329.  Board-to-board NTP tests over network show about 20 ns
>>> jitter, with long-term wanders of 100 ns, presumably due to my limited
>>> sky-view.  More details:
>>>>
>>>> http://fy.chalmers.se/~f96hajo/fakernet/board_Arty_A7-35-GPS-NTP.html
>>>>
>>>> Cheers,
>>>> HÃ¥kan
>>>> _______________________________________________
>>>> time-nuts mailing list -- time-nuts at lists.febo.com
>>>> To unsubscribe send an email to time-nuts-leave at lists.febo.com
>>> _______________________________________________
>>> time-nuts mailing list -- time-nuts at lists.febo.com
>>> To unsubscribe send an email to time-nuts-leave at lists.febo.com
>
>