ADS4225: Finding a proper clock source in conjunction with a pll for the ADC

Emmanouil, 

We are taking a look at your questions and will get back to you asap!

Yusuf

Manos,

Contact the clocking forum regarding the pll questions of the LMX device. See if the attached documents help regarding your other questions.

Regards,

Jim

2352.Clocking High Speed Data Converters - 3_17_2013.pptx2084.ADC Basics.pdf

Thanks Jim,

The undersampling procedure is what troubles me the most. While sampling sine, triangle, square and other waveforms of known shape is an relatively easy task, doing the same for real world waveforms that have no standard shape is something that as i understand cannot be reconstructed with 2-samples per period which is not the case for sine, triangle and other known shapes according to Nyquist theorem.

So i was wondering if there are any guidelines or links you could provide as to how undersampling is performed so that a non-standard waveform can be sampled.

Regarding the PLL i will refer to the forum you suggested.

The docs you provided are really nice. Thanks for that.

Regards

Manos Tsachalidis

Hi Manos,

You've actually fallen into a bit of a trap. What you're referring to is actually "fitting" of data to a known waveform. Sampling a triangle wave at 2 samples per period will actually violate Nyquist sampling theorem. The reason being is that a periodic continuous time triangle wave has "infinite" frequency spectrum - caused by the sharp transition at the peak and trough of the triangle wave - and therefore the entire signal does not fit within one Nyquist zone. The only reason you can faithfully reassemble the triangle wave is because you "know" what the signal is supposed to be and you fit it to a known model. It seems you're looking at the sampling process as a way to "fit" a sampled signal to a known shape.

The theorem is much more restrictive. The signal frequency content must entirely fit within one Nyquist zone (e.g. DC to Fs/2 or Fs/2 to Fs). This means that analog filtering must be sufficient to eliminate all signals outside of the desired Nyquist zone to avoid corruption of the desired signal. Sampling a signal in 2nd Nyquist zone will then result in "aliasing" to the 1st Nyquist zone. Aliasing is how we perceive the output information, however a key to the sampling theorem is that the frequency domain is actually periodic. The 1st nyquist "image" is mirrored around Fs/2 (thus why undersampling a signal in 2nd Nyquist results in a flipped spectrum) and then the 1st and 2nd Nyquist zones are repeated infinitely at a period of Fs. The 2nd Nyquist image of this "infinite" frequency spectrum will then contain exactly the information we need to reconstruct the signal - remember that we analog filtered out any information that exists in other Nyquist zones.

If I wanted to prove this to myself, I would first undersample a signal in 2nd Nyquist zone. I would then interpolate the signal by a factor of 32 by stuffing 31 zeros between each sample and digitally band-pass filter the 2nd Nyquist image. You will see that the signal looks exactly like you expect (like below). Note that I didn't use any "existing knowledge" about the signal during this reconstruction, I simply filtered out the image in the repeating frequency spectrum that corresponded to my desired signal.

The key to the real Nyquist sampling theorem is that to be able to faithfully reconstruct the signal all of the information (frequency content) for the signal must sit entirely in one Nyquist zone, but not necessarily Nyquist zone one. This means that undersampling can result in a reconstructable signal as long as the signal sits entirely in a chosen Nyquist zone.

Regards,
Matt Guibord