Designer’s Guide Community

wandola

Junior Member

Offline

Posts: 24

Still having trouble understanding spectreRF simulation for S/H
Aug 26^th, 2014, 1:35am

Dear all,

I was trying to simulate the noise of dynamic comparator. I understand that this top has been discussed in many places and still, I couldn't fully understand how the spectreRF works.

The first thing I've tried is Kent Kundert's famous paper "Simulating switched-capacitor filters with spectreRF."

I start with a simple S/H circuit.

I firstly setup the PSS simulation. Then I also setup the Pnoise simulation. I can get the correct output noise over frequency. For 1pF cap, the simulated noise power is about 89uV/sqrtHz. The noise source type is timedomain.

I notice that there is another choice when plotting the output noise: From Right mouse button: Direct plot form--> Integ Output Noise. I got a very strange waveform.

Q1 Can anybody explain this function and how to interpret this waveform?

Q2. I also did simulation with noise type: source. The waveform will be attache in below. I also didnt understand how it comes.

Q3. I also did some PAC simulation besides the PNOISE. The outut waveforms looks very weird.

Back to top

Fig1.png

IP Logged

wandola Junior Member Offline Posts: 24	Re: Still having trouble understanding spectreRF simulation for S/H Reply #1 - Aug 26^th, 2014, 1:45am Here is the simulated output noise.
Back to top	s_H_output_noise.png IP Logged

wandola Junior Member Offline Posts: 24	Re: Still having trouble understanding spectreRF simulation for S/H Reply #2 - Aug 26^th, 2014, 1:47am If I choose the other Integ output noise option, the output waveform is shown below. This is for questions 1. What is this integ output noise?
Back to top	integr_output_noise_001.png IP Logged

wandola Junior Member Offline Posts: 24	Re: Still having trouble understanding spectreRF simulation for S/H Reply #3 - Aug 26^th, 2014, 1:56am For questions 2: noise type: source. The plotted output noise and input noise look very strange.
Back to top	Pnoise_sourcetype_source.png IP Logged

wandola Junior Member Offline Posts: 24	Re: Still having trouble understanding spectreRF simulation for S/H Reply #4 - Aug 26^th, 2014, 1:57am output:
Back to top	pnoise_fstop4MHz_noise_source.png IP Logged

wandola Junior Member Offline Posts: 24	Re: Still having trouble understanding spectreRF simulation for S/H Reply #5 - Aug 26^th, 2014, 1:58am For question 3: the PAC analysis. I set the sideband to 10. How come the waveforms are so strange
Back to top	pac_out10_harmonics.png IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2384
Silicon Valley

Re: Still having trouble understanding spectreRF simulation for S/H
Reply #6 - Aug 26^th, 2014, 1:00pm

Let's focus on the simplest question and work through them.

First, you need to tell us the clock frequency.

Now, 'sources' actually means 'time-averaged noise' or 'the noise averaged over one clock cycle' (I don't know why they call it 'sources'). Since this is a track and hold, I would expect it to have roughly a sin(x)/x shape. The shape will be corrupted somewhat because this is not an ideal system (there is some feed thru, the switches do not act instantaneously, etc.). So the output noise seems reasonable to me at first glance. Why does it not seem reasonable to you?

-Ken

Back to top

IP Logged

wandola

Junior Member

Offline

Posts: 24

Re: Still having trouble understanding spectreRF simulation for S/H
Reply #7 - Aug 26^th, 2014, 6:58pm

Hi Ken, thanks a lot for your prompt replay. I really appreciate it.

First of all, the switch clock frequency is set at 400 kHz.The maxacfreq is set at 50 MHz.

I ran PSS + PNoise simulation for the output noise. Maximum sideband is 30. Pnoise sweeping frequency from 0 to 4 MHz (10 times fclk). If I use noise type: time-domain, i can see the output noise plot is about 89uV/sqrtHz, which is close to calculation sqrt(KT/C)=64uV/sqrtHz.

If I use noise type source, it can be seen that the output noise is 1.3 mV/sqrtHz. and the input noise is higher than the output noise. (the second last grpha). Why is the noise power so high now? And why is input noise power higher than output noise power?

I also did PAC analysis. frequency sweeps from 0 to 10*fclk and maxsideband is 10. And PAC magnitude is 1. The output is shown in the last graph. It seems the sinc function fundamental doesnot ring about the 0. It is around 0.5. Do you understand why?

Thanks a lot for your time.

Back to top

IP Logged

Ken Kundert

Global Moderator

Offline

Posts: 2384
Silicon Valley

Re: Still having trouble understanding spectreRF simulation for S/H
Reply #8 - Aug 28^th, 2014, 1:49pm

The reason why the input referred noise is larger than the output noise is that the sample and hold has an effective gain of roughly 0.5 because of the 50% duty cycle of the clock.

This is also the reason why the gain of the primary sideband oscillates about 0.5 in PAC. The input is directly connected to the output half the time, so at high frequencies the effective gain is 0.5.

-Ken

Back to top

IP Logged