Hi,

I am following the example in "Device Noise Simulation of Delta-Sigma Modulators" from this website.  It suggests to analyze the switched-cap integrator with close loop setting and get the input referred noise. But in the "tdnoise"'s "direct plot form", I cannot get the "input referred noise" output.

Should I use "source" to do pnoise? But in that way, I cannot get the folded noise. Because this integrator is used in my sigma delta ADC, so I'd like to get the folded noise as suggested in another paper from this website. (Simulating Switched-Capacitor Filters with SpectreRF).

Your help is highly appreciated!

Cong

Thank you for your reply, Andrew.

In the past few days, I found something:

1.  the pnoise cannot report the input-referred noise. I am using ic5033 in Linux. The software always says the input is not a port, even I am using a port. I asked somebody and they suspect it is a bug in Cadence. So I use noise factor to calculate the input-referred noise. (F-1)*4kTR, the results make sense.

2. the sigma-delta adc is a continuous-to-discrete circuit, so all I need is to calculate all folded noise into (fsample/2), so if I use "tdnoise", I just need to sweep freqeuncy from 0 to (fsample/2), but for "source", I need to sweep more sidebands to get the integral power. I tried both methods for simple "sample-and-hold" circuit, and the results are almost the same, but "source" takes longer simulation time.

But the problem is "tdnoise" cannot give me input-referred noise, so I will use "source" as you suggested and try to find a way out to get input-referred noise, even it is longer and needs to calculate from noise factor. If you have any new ideas, could you share with me?  Thank you so much!

cong  :)

Hi Andrew,

Thank you so much for your comments.

I did try voltage source to do pnoise, but when I asked for "input-referred noise", the output window still complains there is no port.

As for the sweep frequency range, my understanding is: the ADC will convert a continuous time signal to a discrete time signal. So in frequency domain, the signal's spectrum is duplicated in peroid of "fsample". All duplicated signal's strength are the same. So if the signal or noise's bandwidth is wider than fsample/2, we will see aliasing.

In cadence, the tdnoise analysis can simulate this aliasing effect and only 0-fsample/2 sweep is needed. For example, the KT/C noise power is folded, then if I integrate the power from 0 to fsample/2, I can get KT/C power.

But in "source" simulation, the analysis does simulation on the ADC's _analog_ output, which is usually a step response in time domain, which means the ADC samples the continuous time signal input and outputs a flat step voltage(still analog),  to represent the input signal at the sampling time. So, in frequency domain, the signal spectrum is duplicated first, then modulated by a Sinc function. Therefore, in "source" simulation, the spectrum output's strength is modulated by "Sinc" function. Although all sideband's power are folded  into (0 ~ fsample/2), but it is modulated by sinc function, it is not flat any more. So if I integrate this bandwidth, the power is less than the theory value. But if I integrate all the sidebands, the more sidebands I integrate, the more accurate the power is.

I did not go deep in the math here, I just guess and try. Maybe it is true that the integral of Sinc function equals to one. I am not sure.

If I am wrong, please correct me. Thank you very much again.


Best Regards,

cong