Hi

If you are counting on your loop noise to spread your tones out - indeed not a 
good idea. There are several ways you can “go quiet” in your loop….

Bob

On Sep 6, 2014, at 2:10 PM, Magnus Danielson <[email protected]> wrote:

> Hi Bob,
> 
> Indeed. The way to keep the MCU PWM doing reasonable stuff is to use a higher 
> rate, and then update the PWM value in sync with the wrap-around, and then 
> alter the value (dither or whatever) so that the average has higher 
> precision. First degree sigma-delta is actually not a bad strategy and fairly 
> simple to do.
> 
> With FPGA you can do more funky stuff, which is what I do.
> 
> It is even better if you can use a linear DAC of sufficient rate and 
> resolution.
> 
> If you do hold-over functionality with static steering value, that is you 
> stop updating the EFC-steering (this is what most folks do), then the 
> resolution of the full steering can dominate the initial frequency offset. 
> So, one should think about that too.
> 
> As you go into hold-over, all of a sudden you can run into idle-tones in a 
> way that normal dynamics would dither out.
> 
> Cheers,
> Magnus
> 
> On 09/06/2014 07:01 PM, Bob Camp wrote:
>> Hi
>> 
>> One of the easy things to do with PWM is to dither the LSB. That gives you 
>> one more bit of precision. It still keeps the main tone at the same place.
>> 
>> Your worst case tone happens at 50% duty cycle (perfect square wave). If you 
>> do your 50/50 as a square wave at Fmax(not Fmin), your fundamental “worst 
>> tone” is at your highest frequency rather than the lowest. Not easy with MCU 
>> PWM’s, pretty simple with an FPGA.
>> 
>> By far the best thing to do is to clock your PWM at a nice high frequency 
>> (like a couple hundred MHz). That way you get lots of bits and your 
>> fundamental tone is still pretty high. Again, nice for 400 MHz clock FPGA’s, 
>> not so much for $0.50  MCU’s.
>> 
>> Bob
>> 
>> On Sep 6, 2014, at 12:52 PM, Magnus Danielson <[email protected]> 
>> wrote:
>> 
>>> Hi Bob,
>>> 
>>> Agreed. I often find that modulations eats your margin out.
>>> 
>>> PWM is interesting in this regard. PWM has the property that the lowest 
>>> frequency has the highest amplitude and the overtones then decay with 1/f 
>>> from that. For a given clock rate, as you add a bit of PWM precision, you 
>>> half the PWM repetition rate and thus move the frequency down... where we 
>>> are more sensitive to the modulation it causes, and the 1/f slope of the 
>>> oscillator does not help.
>>> 
>>> I designed a PWM-like signal that has reversed PWM spectrum so that the 
>>> highest frequency has the strongest amplitude. The 1/f of the oscillator 
>>> integration makes the modulation flat among the different bits and much 
>>> easier to handle phase-noise wide.
>>> 
>>> Another approach is sigma-delta style modulation, which noises out the 
>>> amplitude. Higher-degree sigma-delta needs to avoid idle-tones for optimum 
>>> result.
>>> 
>>> Thus, paying attention to these details pays of with simplifying the effort 
>>> to achieve good phase-noise properties.
>>> 
>>> There is more dangers that can occur in PWM-space, but this should be 
>>> enough of a starting-point.
>>> 
>>> Cheers,
>>> Magnus
>>> 
>>> On 09/06/2014 01:39 PM, Bob Camp wrote:
>>>> Hi
>>>> 
>>>> Yes indeed, as you go below 1 Hz (or 1 radian/sec) all the things that 
>>>> “help” you roll off wise now hurt you. If you are worried about sidebands 
>>>> inside 1 Hz, you need to change a sign here and there. The only thing that 
>>>> saves you is that the noise floor is now coming up pretty fast.
>>>> 
>>>> If you modulate a crystal oscillator, the loaded frequency of the crystal 
>>>> is changed to accomplish the modulation. When your FM swings 100 Hz high, 
>>>> your crystal is tuned 100 Hz high. When your modulation swings 100 Hz low, 
>>>> your crystal is tuned 100 Hz low. The Q has no impact in this case. No I 
>>>> did not believe it worked that way until I did it …. Since then I’ve built 
>>>> a *lot* of VCXO’s with modulation bandwidths >> than their crystal Q 
>>>> bandwidths. The biggest problem comes from crystal spurs rather than 
>>>> crystal Q.
>>>> 
>>>> Bob
>>>> 
>>>> On Sep 6, 2014, at 6:09 AM, Magnus Danielson <[email protected]> 
>>>> wrote:
>>>> 
>>>>> Bob,
>>>>> 
>>>>> On 09/06/2014 03:00 AM, Bob Camp wrote:
>>>>>> Hi
>>>>>> 
>>>>>> Oddly enough (and yes it is odd) you can modulate an oscillator well 
>>>>>> outside the crystal’s bandwidth. The bigger issue is that the EFC does 
>>>>>> not pull the crystal very far on a normal OCXO. The FM modulation index 
>>>>>> drops to very small numbers pretty fast as you go up in modulation 
>>>>>> frequency.
>>>>>> 
>>>>>> You typically only worry about modulation sidebands that are above the 
>>>>>> phase noise floor. Since phase modulation sidebands go down as 1/Fmod on 
>>>>>> an FM modulator (for small modulation index) they get pretty low pretty 
>>>>>> fast.
>>>>>> 
>>>>>> If your OCXO has an EFC range of 0.1 ppm at 10 MHz, it will swing 1 Hz 
>>>>>> p-p (+/- 0.5 Hz) for the full EFC voltage. At 5 Hz, you have a 
>>>>>> modulation index of 0.1. Of course if you are multiplying to 10 GHz, the 
>>>>>> index could be quite large. This gets back to the “this all depends on 
>>>>>> what you are doing”.
>>>>>> 
>>>>>> If your EFC is 5V, a reasonably quiet signal would have noise below 0.5 
>>>>>> mV. That’s already 80 db down. A very quiet supply should be in the < 5 
>>>>>> nV / sqrt(Hz) range.  That would put the noise down 180 db.
>>>>>> 
>>>>>> It’s unlikely that your OCXO has a phase noise spec of -180 dbc / Hz at 
>>>>>> 10 Hz. We may already be done …
>>>>>> 
>>>>>> To bring all the numbers together:
>>>>>> 
>>>>>> At 1 Hz the modulation will do a sideband X db down at your desired 
>>>>>> frequency.
>>>>>> 
>>>>>> You will drop 20 db by the time you get to 10 Hz simply due to the 1/F 
>>>>>> FM->PM.
>>>>> 
>>>>> Since the oscillator integrate frequency into phase, you have a 
>>>>> 1/(2*pi*f) factor. The typical LaPlace model for an oscillator is Ko/s, 
>>>>> where Ko is the input sensitivity of the oscillator.
>>>>> A more complete model needs to include the Q of the crystal, naturally, 
>>>>> unless you are "in-band" of that Q where it has less drastic properties.
>>>>> 
>>>>>> Bottom line - it’s not all that hard to get a quiet enough EFC voltage.
>>>>> 
>>>>> Agreed.
>>>>> 
>>>>> I've found that thinking about systematic noises of low frequency (i.e. 
>>>>> comparator frequency and overtones) as well as loop dynamics is what one 
>>>>> should think about. Lack of DAC resolution hurts.
>>>>> 
>>>>> Cheers,
>>>>> Magnus
>>>>> _______________________________________________
>>>>> time-nuts mailing list -- [email protected]
>>>>> To unsubscribe, go to 
>>>>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>>>> and follow the instructions there.
>>>> 
>>>> _______________________________________________
>>>> time-nuts mailing list -- [email protected]
>>>> To unsubscribe, go to 
>>>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>>> and follow the instructions there.
>>>> 
>>> _______________________________________________
>>> time-nuts mailing list -- [email protected]
>>> To unsubscribe, go to 
>>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>>> and follow the instructions there.
>> 
>> _______________________________________________
>> time-nuts mailing list -- [email protected]
>> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>> and follow the instructions there.
>> 
> _______________________________________________
> time-nuts mailing list -- [email protected]
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.

_______________________________________________
time-nuts mailing list -- [email protected]
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.

Reply via email to