[time-nuts] DDS - higher frequecies

[Jim Lux]

On 11/25/12 4:30 PM, Hal Murray wrote:
>
> Suppose I have an A/D running at 1 MHz.  The standard simple minded approach
> is that it will work for any input signal with a bandwidth up to 1/2 MHz.  We
> usually think of that in the baseband, but it also works for, say  1.25 to
> 1.5 MHz.  The input signal gets aliased down into the baseband.  (and if you
> are unlucky, which is easy, some of the aliasing reflects back and overlaps
> so you can't tell X-y from X+y)
>
> There is similar math for D/A, the reverse direction.  I think this applies
> for a DDS making higher frequencies than simple arithmetic would allow it to
> generate.
>

Yes.. you generate all the aliases.. Fx, Fs-Fx, Fs+Fx, 2Fs-Fx, 2Fs+Fx, etc.


> Does anybody have a good web page for how that works?  My simple expectations
> are that it would have to generate lots of harmonics and then go through a
> filter to get rid of all the wrong stuff.  I'm missing the step where all the
> harmonics come from.

Not exactly harmonics, but aliases.

What you are doing is convolving the sampling function (a series of 
ideal impulses, either in frequency or time domain)  with the other signal.

It's basically the opposite of an undersampling IF

ANd, because the typical output has a sample/hold, it's not a series of 
impulses, but a staircase, so the frequency domain has a sin x/x shape 
to it.

>
> Are they just really tiny and I have to do a lot of good filtering and
> amplification?

Yes..
And all the usual things about timing jitter aggravating the higher 
order outputs more than the first order, etc.


>
> Do I need something other than a traditional DDS for this sort of stuff?
>
>
>