The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Measurements >> Phase Noise and Jitter Measurements >> Strobed noise problem
https://designers-guide.org/forum/YaBB.pl?num=1212503415

Message started by ritchie_ch on Jun 3^rd, 2008, 7:29am

Title: Strobed noise problem
Post by ritchie_ch on Jun 3^rd, 2008, 7:29am

Hi all,
As we know, the strobed noise is valid only in the frequency range of -fo/2 to +fo/2, where the fo stands for output freuqency.
I have got the strobed phase noise of a buffer which is the input buffer for the PFD input. If the ouput frequency of the buffer is 40MHz, thus the phase noise is only valid between 0~20MHz (if half the side band is considered). Then I use the phase noise data to a frequency domain model of the PLL in matlab in order to estimate the output phase noise of PLL. Because of the input phase noise is only valid between 0~20MHz, I could only calculate the phase noise of PLL between 2GHz~2GHz+20MHz and if I want to get the phase noise at higher frequency offset, say 2GHz+40MHz, there is no valid data availble for input phase noise. What should I do?

Any anwser is appreciated!

-Ryan

Title: Re: Strobed noise problem
Post by Ken Kundert on Jun 3^rd, 2008, 8:50am

You are wrong when you say the output from strobed noise is only valid over a range of -f₀/2 to +f₀/2. Rather the noise is periodic with a period of f₀, so you only need to know the noise over the range -f₀/2 to +f₀/2. Once you know if over that range, you know it everywhere.

-Ken

Title: Re: Strobed noise problem
Post by David Lee on Jun 3^rd, 2008, 10:47pm

Do a conventional pnoise analysis of your input buffer + FPD + charge pump + loop filter, at your frequency of interest.

Title: Re: Strobed noise problem
Post by ritchie_ch on Jun 4^th, 2008, 3:07am

Ken Kundert wrote on Jun 3^rd, 2008, 8:50am:

Thanks Ken, David :)
As far as I know, all of the noise power concentrate between -fo/2 ~ +fo/2, and for a simple RC network, the power would be kT/C. If the noise is periodic with a period of fo, the total power would be infinite in the whole frequency range because there are infinite shares of kT/C with a period of fo? I fail to figure out where the extra noise power comes from or do I misunderstand something?
There still is another problem. As you said, the strobed noise is periodic, so there should exist replicas of the noise spectrum between -fo~+fo. For my PLL under test mentioned above, the input clock buffer before PFD operate at 40MHz, so I should get a spectrum of strobed phase noise with replicas at 40MHz, 80MHz, 120MHz, ..., etc.. When I use this spectrum as reference clock phase noise in Matlab, I would see strong spurs at these frequency offsets for the noise at these points runing to infinite. Am I right?

-Ryan

Title: Re: Strobed noise problem
Post by David Lee on Jun 4^th, 2008, 12:32pm

This is an excellent question.

In real life, the output waveform of your input buffer has finite noise power. The spectrum extends from -infinity to +infinity. When you sample the output waveform periodically, the noise samples also have a finite power (finite variance). The spectrum of the noise samples is periodic outside (-fo/2,+fo/2). The total noise power (variance) within (-fo/2,+fo/2) is finite. The fundamental issue is this. You are trying to take information from a discrete-time noise sequence, and apply it to a continous-time system (PLL). To avoid this, do everything in continuous time.

So, run regular pnoise analysis on input buffer + PFD + charge pump + loop filter to obtain accurate result at any frequency f.
This will include noise at these frequencies: f, fo +/- f, 2 fo +/- f, 3 fo +/- f, ...

Title: Re: Strobed noise problem
Post by Ryan Cheung on Jun 5^th, 2008, 5:13am

David,

Thanks for your reply!
So you mean I should use 'noise type=sources' in pnoise to simulate on buffer+PFD+charge pump+loop filter? But this type of simulation gives time averaged noise. Could this correctly describe the phase noise behavior if a digital buffer (with discrete noise attribute) is included?

Another problem. What if I want to simulate the phase noise of a digital buffer at the PLL output. I have to add this discrete-time noise with the phase noise of the PLL (continuous one), and the same problem is unavoidable. Any suggestion is appreciated! :-[

-Ryan

Title: Re: Strobed noise problem
Post by David Lee on Jun 5^th, 2008, 9:12am

Ryan Cheung wrote on Jun 5^th, 2008, 5:13am:

In many PLLs, much of the PLL phase noise comes from the loop filter. Would you use time averaged noise to model that? Or noise at a set of discrete times? Both are approximations of reality, but I think time averaged noise is much more realistic, because the VCO is designed to react to its inputs at all times.

Title: Re: Strobed noise problem
Post by Ryan Cheung on Jun 6^th, 2008, 8:00pm

Thanks,

Yeah, you are right. But my PFD here is a special architecture and we have a tight spec., so any noise source cannot be ignored. So I am very concern about the phase noise contributed from blocks like buffer, PFD, and charge pump, which may be igored in traditional design.
So, if I run PFD+charge pump with 'noise type=sources', and the PFD jitter is dominant, could the simulator give me a good estimation of current noise at loop filter?

-Ryan

Title: Re: Strobed noise problem
Post by David Lee on Jun 10^th, 2008, 12:40pm

By simulating buffer + PFD + charge pump + loop filter together, you can see the total noise at the output of the loop filter, due to all noise sources from your entire circuit. No approximation is needed.

Title: Re: Strobed noise problem
Post by Ryan Cheung on Jun 10^th, 2008, 11:12pm

David,

Thank you very much! I have done the same thing as you said. But the simulator always gives a warning saying that the 'maxsamples' need to be given a large number. What does the 'maxsamples' option mean?

-Ryan