sigma-delta AD question: judging AD ability

[tdy]

hur fungerar en sigma delta ad

I am trying to understand how to judge a AD ability. and I run a transient simulation in cds sprctre for about 5 cycles of input signal. should I do the dft directly at the digital output node? Thanks in advance.

 

[JoannesPaulus]

sigma-delta AD question

Whatever the AD ability is... yes, you can see the non-decimated spectrum of your ADC by taking the dft of the digital output (one bit per sample).

 

[tdy]

sigma-delta AD question

I do not know much about the sampling theory, but the dft process is a sampling process, right? so the more points it samples the more accuracy it achieves. my carrier freq is 10M and the input signal is 10K and the time interval is 500us after it settles. that means I have to sample more than 10000 times to get the anti alias result. but the dft result is different if i use 10,000 points and 1000,000 points, espectially for the low freqency. I suppose the latter is more accurate.

 

[tdy]

sigma-delta AD question

another thing I am not quite sure is for the dft output. should the noise floor at LF be sqrt(KT/fsC) from the dft output?

 

[JoannesPaulus]

Re: sigma-delta AD question

I do not know much about the sampling theory, but the dft process is a sampling process, right? so the more points it samples the more accuracy it achieves. my carrier freq is 10M and the input signal is 10K and the time interval is 500us after it settles. that means I have to sample more than 10000 times to get the anti alias result. but the dft result is different if i use 10,000 points and 1000,000 points, espectially for the low freqency. I suppose the latter is more accurate.

The accuracy of the DFT, as in frequency resolution, depends on the data capture time, i.e. if you want to discriminate a 1kHz frequency, you need to have a 1ms capture. The number of points is fixed by the sampling rate: if you have a sampling rate for your DS-ADC of 40MHz (the rate at which the digital output comes out), you can have 40000 points in 1ms.
It is good practice to use a power of 2, so, I would use either 32768 (with a 1.22kHz frequency discrimination and shorter simulation) or 65536 points (with a 610Hz frequency discrimination but a longer simulation).

 

[tdy]

sigma-delta AD question

a power of 2, that's quite a hint. I will try to validate that. Thanks a lot.