0
$\begingroup$

There is a lot of literature on sampling rates. Typically, as a rule of thumb around 20% sampling rate margin to bandwidth (e.g. 1.2 x 2 x B) can be used for setting the optimum Nyquist-compliant sampling rate. I have a baseband signal going from -3.75 GHz to 3.75 GHz (3.75 GHz bandwidth). At the moment I'm using a 9 Gsps sampling rate (20% margin), but I'm wondering if it would make sense to use 8 Gsps (typical available ADCs value), which would mean only 6.7% margin.

My signal is mainly Gaussian noise by the way.

Improve this question
$\endgroup$
3
  • $\begingroup$ That will depend a lot on the quality of you anti-aliasing filters and how sensitive your pass band edge's are to residual aliasing. $\endgroup$
    – Hilmar
    Commented Aug 28 at 9:10
  • $\begingroup$ I agree that the passband and attenuation band performance will have an impact on aliasing. However, the filter is not designed. What is realistic to assume? $\endgroup$
    – Albert
    Commented Aug 28 at 9:29
  • $\begingroup$ This will come down to a trade off between the benefits of a lower sample rate and the cost or complexity of a more aggressive anti-aliasing filter. How that trade off looks depends on the specifics of your situation. Chances are that the "rule of thumb" comes from a lot of people doing the trade-off analysis for their application and came up with "20% is the sweet spot". FWIW: even in audio 20% is pretty much the standard. $\endgroup$
    – Hilmar
    Commented Aug 28 at 17:15

1 Answer 1

Reset to default
0
$\begingroup$

This is really a consideration to the specific analog anti-alias filter used ahead of the ADC and becomes a part of the overall noise figure budget of the measurement device (noise figure being how much margin you allow for implementation errors). Increasing the sampling rate margin decreases the complexity of this filter design, but increases the sampling rate required. The complexity of that filter is also driven by how much rejection is needed in the alias regions that would ultimately fold into the operating bandwidth of the signal of interest- and for radio design specifically (which is my familiarity) that rejection is typically driven by an amount of allowable desensitization due to interference, jamming, and other adjacent channels that may be operating simultaneously with the signal. This would be part of the Minimum Performance Requirements or other specifications for the receiver. I imagine these same considerations would carry to many other applications where an ADC is used for purpose of measuring sensitive signals. I personally use a starting range of 20% to 30% to get a ball-park value but then as the design proceeds that becomes part of the trade space based on the available ADC devices, sampling clock options, and analog chain (especially the anti-alias filter) and overall system requirements for sensitivity, dynamic range and noise.

I detail a lot of this in my existing answer to DSP.SE#52646. where I specifically show how the concept of cascaded noise figure design (typical radio receiver design approach) can be used in a mixed signal case for determining ADC factors such as sampling rate and number of bits.

As for further details and considerations for the anti-alias filter, please see my existing answer for DSP.SE #63191 as there are a lot of points in that post that help avoid over-designing or over-constraining the filter based on considerations specific to the signal under interest, mainly the actual minimum bandwidth required for processing the signal, and the spectral properties of the maximum expected interference, noise, gain and other signals that may be present ahead of the anti-aliasing filter.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks Dan. Your answer is clear. I should look at realistic low-pass anti-aliasing filter designs to see what is feasible. But in the end it's a combination of achieved performance vs. technological challenges I guess... $\endgroup$
    – Albert
    Commented Aug 28 at 14:17
  • $\begingroup$ @Albert Yes there isn't any one specific answer and I think it does require a system design perspective to make the appropriate trades between the analog and digital boundary and the related technologies and techniques that can be employed (I have good familiarity in both areas so often land in that role professionally). $\endgroup$
    – Dan Boschen
    Commented Aug 28 at 23:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.