[time-nuts] spur prediction DDS software

[Bruce Griffiths]

Bruce Griffiths wrote:
> Magnus Danielson wrote:
>> On 27/01/11 14:33, jimlux wrote:
>>> On 1/27/11 5:17 AM, Luis Cupido wrote:
>>>> Hi,
>>>>
>>>> Is there a DDS spur prediction software around ?
>>>
>>> Not generically..
>>>
>>> There is a dissertation out there with some matlab code. I'll see if I
>>> can find a link
>>>
>>> Most of the time what I do is write a little program in matlab/octave,
>>> run a bunch of samples, and FFT it. That way you can model things like
>>> error feedback or truncation effects.
>>
>> What I do is to write a little C program that keeps accumulating away.
>> Whenever the wrapped value becomes lower than the previous value I 
>> have a cycle completed. Just print the values of each such phase-wrap 
>> and you will readily see the phase movements. You will also notice 
>> that this is a number of phase movements overlayed ontop of each 
>> other. Each such overlay structure has an integer number of cycles 
>> over the DDS period. It really takes very little time to break them 
>> out and get their relative amplitudes using this method. This is a 
>> healthy exercise to do to learn what DDS phase noise can result in.
>>
>> Thinking a bit more about it, you realize that you can see the same 
>> pattern as being the result of a sawtooth being wrapped up on itself 
>> around the nyquist frequencies. Pondering some more over it you 
>> realize that there is DDS frequencies in which these overtones will 
>> wrap on top of each other in the wrapped spectrum, This condition 
>> will avoid close-in spurs, so an analogue clean-up has a fair chance 
>> of doing something with them. It also becomes clear that small shifts 
>> can produce larger shifts of spurs...
>>
>> Having a fixed M/N situation where N=2^n (common for many DDSes) will 
>> limit the frequencies with the "good" condition of spurs being far 
>> away, but shifting to a variable N improve things in this context, 
>> but makes some control aspects messy.
>>
>> A problem with using a 2^(n-1) < N < 2^n is that the phase-sequence 
>> will run short compared to the usual binary bits taken from the top 
>> of the DDS, so this will needs to be handled. It takes some extra 
>> tricks to avoid the phase-jump and hence sawtooth phase modulation.
>
> The AD9913 (when operating in the variable modulus mode) merely adds 
> an extra 1 (lsb) to the main accumulator whenever the auxiliary 
> (variable modulus) accumulator overflows.
> Thus to avoid phase modulation of the output the combined result of 
> the 2 accumulators has to result in the main accumulator always having 
> zeros in the lower b-p bits where p is the width of the truncated 
> phase word and b is the width of the main accumulator.
>
More precisely the phase truncation error must be constant (ie the phase 
truncation error, the least significant b-p bits of the main accumulator 
should always have the same value).
This requires that the auxiliary accumulator overflow every cycle and 
the ls p-b bits of the main frequency tuning word have the value 2^(p-b) -1.
ie using the auxiliary accumulator doesnt add any extra phase truncation 
spur free frequencies compared to those available when only the main 
accumulator is used.
Thus although using the auxiliary accumulator may allow frequencies like 
10MHz - 10Hz to be generated from a 100MHz DDS clock they will not be 
phase truncation spur free.

>>
>> The good thing about DDSes is that you with very cheap hardware can 
>> get a linear scale over a large frequency range. It will not 
>> necesserily align up to the intended frequency (as dialed in). More 
>> elaborate schemes will reduce these effects.
>>
>> ... and that was prior to do phase quantization prior to sine shaping.
>>
>> Cheers,
>> Magnus
>>
>> _______________________________________________
>> time-nuts mailing list -- time-nuts at febo.com
>> To unsubscribe, go to 
>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
>> and follow the instructions there.
>>
> Bruce
>
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to 
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>