[time-nuts] Re: function generator

[Erik Kaashoek]

Some time ago I needed a output at 10,000000.001Hz so I tried to do that 
with a SI5351.
Using pure integer math (as the PLL and divider register are integers) i 
search for a combination of 3 divider/multipliers that gave the least error.
If the reference frequency is not integer related to the internal PLL 
frequency and  multiplier/divider registers you always will have limited 
accuracy as there is a fractional error.
The amount of error will depend on the number of digits in the 
multiplier/divider register lengths and the care you take to search for 
the best solution.
If the DDS has a PLL driven clock this could be the cause.
Erik PD0EK

On 22-11-2021 18:14, Jeremy Elson wrote:
> Andy,
>
> I was thinking in the same direction - the device can create arbitrary
> waveforms so they're almost certainly using DDS. So I assumed it was some
> sort of round-off error. I just can't quite convince myself of this because
> it seems like round-off errors would be bigger, i.e. an error of one part
> in 1E11 suggests a buffer of length 1E11 which seems unlikely.
>
> On Mon, Nov 22, 2021 at 9:06 AM Andy Talbot <andy.g4jnt at gmail.com> wrote:
>
>> Could the Rigol be using a DDS / NCO lookup approach to its frequency
>> generation?  In which case you'll be subject to rounding errors on the DDS
>> increment, whose maximum magnitude will be  dependent on the register
>> length.
>>
>> Andy
>> www.g4jnt.com
>>
>>
>>
>> On Mon, 22 Nov 2021 at 16:53, Jeremy Elson <jelson at gmail.com> wrote:
>>
>>> I did not see any such setting in the Rigol, but I'll check again. in
>> April
>>> I did write to Rigol to report the problem and had the following
>> (abridged)
>>> conversation with support:
>>>
>>> Me:
>>>
>>> "I recently tried to use your DG1022Z signal generator to generate one
>>> pulse-per-second (pulse mode, frequency 1.000000hz, width 10
>> microseconds).
>>> However, it appears there is a small frequency error of one part in 1e11,
>>> i.e. the pulse per second gets later by about 6 nanoseconds every 1,000
>>> seconds." [More technical description abridged, including a link to a
>>> graph.]
>>>
>>> Rigol:
>>>
>>> "Please find the following datasheet for DG1000Z, the accuracy is
>> +/-1ppm.
>>> If your pulse is 1second with 10us width, 6ns per 1000s is in the
>> accuracy
>>> range."  [They attached an image of a page from the DG1000Z datasheet,
>>> showing a line that said "Accuracy: +/- 1ppm of the setting value"]
>>>
>>> Me:
>>>
>>> "Is this the specification even when the unit is provided with an
>> accurate
>>> external clock?"
>>>
>>> Rigol:
>>>
>>> "I would say Yes. The internal processing circuit will effect the clock
>>> signal,  harmonics and phase noise will result the frequency variance."
>>>
>>> -Jeremy
>>>
>>> On Sun, Nov 21, 2021 at 11:54 PM Poul-Henning Kamp <phk at phk.freebsd.dk>
>>> wrote:
>>>
>>>> --------
>>>> Jeremy Elson writes:
>>>>
>>>>> [...] I plugged an RbXO into the DG1022Z and simply asked the
>>>>> DG1022Z to divide it down to a 1pps signal. That is, I configured it
>> to
>>>>> create a pulse with a small duty cycle and 1hz frequency using the
>>>> external
>>>>> clock as a reference. To my surprise, it introduces a small but
>>>>> measurable error of one part in 1E11 in the dividing-down. You can
>> read
>>>> the
>>>>> full story of this in my post from earlier this year:
>>>> When you feed such instruments an external clock, you often have to
>>> change
>>>> one or more calibration constants for the internal (OC)XO's offset to
>>> zero.
>>>> Another problem is that if you use the square wave output of a DDS
>> based
>>>> generator, they often produce the square from the sine output of
>>>> the DDS chip with a schmitt-trigger, which causes lousy jitter.
>>>>
>>>> --
>>>> Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
>>>> phk at FreeBSD.ORG         | TCP/IP since RFC 956
>>>> FreeBSD committer       | BSD since 4.3-tahoe
>>>> Never attribute to malice what can adequately be explained by
>>> incompetence.
>>> _______________________________________________
>>> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe
>> send
>>> an email to time-nuts-leave at lists.febo.com
>>> To unsubscribe, go to and follow the instructions there.
>>>
>> _______________________________________________
>> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe send
>> an email to time-nuts-leave at lists.febo.com
>> To unsubscribe, go to and follow the instructions there.
>>
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com -- To unsubscribe send an email to time-nuts-leave at lists.febo.com
> To unsubscribe, go to and follow the instructions there.