Separation of signal from noise in sigma delta ADC

Hi, I am working on sigma delta modulator based ADC.

I have output bitstream from the modulator and would like to measure SNR and SNDR using FFT.

How could I possibly separate the signal and noise from the output of modulator?

Also, is there anyway to separate the noise and distortion from the total noise from the modulator output above or can I only do that from an analysis of signal harmonics ?

Thank you,

Well, use the following techniques:

1) Adaptive Viner filter
2) wavelet denoising with hard and soft thresholding and minimax thresholding
3) Empirical mode decomposition (EMD) and singular spectrum analysis (SSA) (I'm the expert in EMD, i've been working on it for more than 2 years, and with great pleasure i'll explain you all the possibilities of this method)
4) Fourier denoising (see www.basegroup.ru, unfortunately, it's in Russian, nevertheless you'll find there the page, devoted to initial signal processing, which includes denoising)

By the way, you can estimate the effectiveness of denoising not only by SNR, but by RMSE (Root mean square value) as well.

I wonder, what is SNDR?

With respect,

Dmitrij

Thank you for your reply, Dmitrij.

SNDR is the ratio of signal to noise+distortion.

Could you please explain more about EMD and SSA method in details ?

I really appreciate for your help.

Thank you.

Dear sehun119!!!

Another member of this forum has already applied to me with the request to explain about EMD. I've already written a small report for him/her. Please, see the question in this forum by t.m.keerthi "projects on DSP applications". Read this brief description and ask questions, if you are interested in the topic. I'll answer them with great pleasure!!!

With respect,

Dmitrij

Added after 12 minutes:

I have one more idea about denoising!!!

You know, that in the majority of cases, we collide with in signal processing, noise is usually a high oscillating process. This denotes, that it's PSD (power spectrum density), is high-pass, so it's close to zero in low-band frequency domain. Therefore, it's possible to use lowpass filter in order to get rid of harmful noise component.

But, of course, you'll have to adjust the parameters of the filter to the concrete noise, otherwise, the signal can be spoilt significantly. The parameters are the cutoff frequency and the attenuation in the stop-band. If you know the kind of noise, at least, approximately (it's variance, PSD) change the parameters of the filter until the output signal of the filter is zero (or close to zero with the given extent of accuracy). The input signal of your filter - is noise!!!

When the needed parameters are found, apply your filter to the combination of signal+noise and see the results. At first, you can do it on model signals, which can be easily generated in Matlab. The estimation of the effectiveness of denoising may be done by computing SNR, SNDR, RMSE. If the results are satisfactory, it's very good.

Besides, I paid your attention on the site www.basegroup.ru. There you can find the method of denoising, based on FFT!!! Unfortunately, it's in Russian, but if necessary, I can find this method and tell you afterwards (I'm from Russia, Saint-Petersburg)!!!

With respect,

Dmitrij