0
\$\begingroup\$

I understand that oversampling will reduce quantization noise floor.

Would oversampling also reduce the noise floor of the noises in band of interest?

*******EDIT 20150320*********

The correct answer is from the comment of Brian Drummond.

Oversampling combined with proper DSP will help reducing quantization noises and thereby increase ADC resolution, under given conditions. Details can be found at: http://www.silabs.com/Support%20Documents/TechnicalDocs/an118.pdf

However, it cannot reduce in-band noises.

\$\endgroup\$
3
  • 4
    \$\begingroup\$ Not if the noise is present in-band before sampling. \$\endgroup\$
    – user16324
    Commented Mar 18, 2015 at 11:51
  • 1
    \$\begingroup\$ ^ this. You can't reduce thermal noise, it's there and you get it all. \$\endgroup\$ Commented Mar 18, 2015 at 12:01
  • 1
    \$\begingroup\$ You should ask on dsp.stackexchange.com \$\endgroup\$
    – RawBean
    Commented Mar 18, 2015 at 12:49

2 Answers 2

5
\$\begingroup\$

You can't reduce noise that is already present at base band but, by appropriate filtering and over-sampling you can reduce out of band noise aliasing into the base band when the ADC converts. Aliasing: -

enter image description here

Here, a signal (in red), is grossly under-sampled and it produces the blue signal. The blue signal is not representative of the original red signal and this is called an aliased signal. This is due to not designing a secure-enough anti-alias filter before the ADC. The anti-alias filter would attempt to significantly reduce the amplitude of the red signal before it entered the ADC.

It's the same for noise - noise above the Nyquist frequency of an ADC can alias down into the base band after conversion and this adds to the original base band noise. Here's a picture of several signals of which one is below the Nyquist frequency, the rest becoming aliased: -

enter image description here

Looking at the left picture, signal(1) is converted and remains at its correct frequency on the right hand picture but, all the other signals (those above fs/2) are aliased down into the base band.

So, if you use over-sampling, the anti alias filter becomes more effective and signals 2, 3, 4 and 5 become much more heavily filtered by the anti alias filter. There will still be some aliasing artifacts but significantly reduced.

Noise is just a wide-band signal and is treated no differently by the ADC: -

enter image description here

\$\endgroup\$
5
  • \$\begingroup\$ Thanks Andy, I like your picture and explanation. I did not choose it as the correct answer, because it seems that it does not answer the question. \$\endgroup\$ Commented Mar 20, 2015 at 9:26
  • \$\begingroup\$ @richieqianle I think andyaka answered your question in the first couple of lines. He says "You can't reduce noise that is already present at base band" but then goes on to explain that if out of band noise is aliased into the baseband by the sampling process, then oversampling can help reduce that. But I think he's not explaining that this helps to relax the anti-aliasing filter requirements, making filtering easier/ more effective. \$\endgroup\$
    – akellyirl
    Commented Mar 20, 2015 at 9:53
  • \$\begingroup\$ @akellyirl you can take a horse to water but you can't make them drink LOL \$\endgroup\$
    – Andy aka
    Commented Mar 20, 2015 at 10:13
  • \$\begingroup\$ @akellyirl I would explain why oversampling helps reduce quantization noises and does not work for in-band noises instead; after all, that is what is not understood. \$\endgroup\$ Commented Mar 20, 2015 at 12:18
  • \$\begingroup\$ @richieqianle Just to be clear, my answer is NOT about quantization noise. It's about aliasing out of band noise (or signals) into the baseband. I also pointed out that noise that is already in the baseband cannot be reduced. You seem to think otherwise but that is your prerogative. What you propose as another answer (the silicon lab link) only works when natural noise is greater than quantized noise btw. \$\endgroup\$
    – Andy aka
    Commented Mar 20, 2015 at 13:36
0
\$\begingroup\$

Oversampling (i.e. sampling at frequencies significantly higher than the Nyquist rate) reduces neither the in-band noise nor the quantization noise. It makes better antialiasing possible, which is important, but that's all it does.

\$\endgroup\$
2

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.