The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> What and how to use SpectreRF Timedomain Pnoise?
https://designers-guide.org/forum/YaBB.pl?num=1421274857

Message started by joebob on Jan 14^th, 2015, 2:34pm

Title: What and how to use SpectreRF Timedomain Pnoise?
Post by joebob on Jan 14^th, 2015, 2:34pm

While Timedomain Pnoise looks like an interesting option for oscillators such as a ring osc where instantaneous noise is really important at the crossing, how does this compare to regular pnoise and the difference in the results? Is it lower? For lab equipment, how does one correlate timedomain pnoise in silicon? Or, is the timedomain noise just a "relative" measurement correlating the RMS period jitter.

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by Ken Kundert on Jan 16^th, 2015, 4:49pm

Noise in oscillators is predominantly low frequency phase noise, so there should be no benefit using the time-domain noise as the noise itself is largely independent of time.

-Ken

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by joebob on Jan 22^nd, 2015, 1:51pm

Ken,

Thanks for your comment. Agree. Phase noise is not really a time dependent measurement. But was interested in correlating somewhat abs jitter to phase noise if include enough sidebands. The regular pnoise gives reasonable correlation to silicon but not all the time.

Also found a difference using SpectreRF Pnoise and AFS Pnoise. AFS Pnoise is by default full spectrum large signal noise and not small signal. This shows differences in the noise floor and also close-in carrier noise. But what to believe?

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by sheldon on Jan 27^th, 2015, 3:48pm

joebob,

Maybe you should check with your Cadence support engineer. The
default for Spectre RF Shooting Newton pnoise is full spectrum noise.
Since you want the detail of the close-in noise, you need to enable
the pnoise Lorentzian option. Did you do that? When you say large
signal noise do you mean transient noise analysis?

Sheldon

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by joebob on Jan 27^th, 2015, 5:59pm

Sheldon,

Thank you for your reply. Yes, I should consult someone who would understand the choices. I did not know Spectre pnoise was full spectrum by default and thought there was an option to set for this. If default, that is good. I understand additionally AFS pnoise is not small signal and there is a difference in the two results.

How accurate is the Lorentzian option since close to the carrier is hard to accurately model? If accurate, why isn't this default? I think there is a button to click to turn it on. I really don't like clicking a button without fully understanding what it does. I think the simulation time is similar but definitely fills in the corner. Are there other ways to model close to carrier in-band noise? TN is an option. There is growing need to model closer to carrier since circuits are more sensitive to in-band noise at high frequencies.

As I understand from what someone told me, AFS Transient Noise is also full spectrum noise and found it not too sensitivity to noisefmax other than maybe some 1/f noise. Having to set noisefmax with high 50x frequencies can cause the time step to drop really low and longer simulation. AFS also evaluates noise at each timestep and is designed to be truly random so think this will be more accurate since see correlation reports to silicon. TN adds like 2-4x time but that dependent on the circuit and options set so is not without cost, especially if dealing with RC parasitics.

thanks again,

- Joebob

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by sheldon on Jan 28^th, 2015, 4:34pm

Joebob,

When on Designer's Guide, I try to avoid commercials so won't
respond to everything. Again talk to the local team.

First a comment, Spectre RF calculates the close-in phase noise
including 1/f noise effects so transient noise is not required. Since
the original equation for close-in phase noise from Lorentz assumes
white noise, using the term Lorentzian maybe a little misleading.
However, it helps most people understand what the option does
with being cryptic.

Yes transient noise can be used for close-in phase noise. However,
it will be inefficient compared to periodic steady-state analysis
with pnoise since
#1) You will need to use a low fundamental frequency for Fourier Analysis
You looking for close-in phase noise and stop time is inversely
proportional to frequency, 1kHz is 10x slower than 10kHz.
#2) You will need to calculate the power spectral density not use
an FFT
Since you are looking for noise you will probably end up
needing to use Welch's averaged periodogram method
method. To achieve good accuracy you will need average
multiple times.

Assuming you want a fundamental frequency of 1kHz and a
clean spectrum, 64x averaging, the transient stop time needs
to be 64ms, for a ~2.4GHz oscillator, that turns out to be ~150M
periods. As a result if you use transient noise analysis, you will
need to run a very long simulation. Been there done that, it is
painful.

Sheldon

Title: Re: What and how to use SpectreRF Timedomain Pnoise?
Post by Ken Kundert on Jan 28^th, 2015, 7:30pm

I am always skeptical of transient noise results. It is difficult to get accurate results because the noise signals are quite small and often overwhelmed by simulator errors. In addition, requesting information about low frequency noise, such as flicker noise, causes the simulations to run extremely slowly. Finally, their primary claimed benefit, that they can handle nonlinear noise effects, is almost never needed. Ironically, to see nonlinear noise effects in an oscillator, you have to simulate very close-in noise, which is extremely slow and largely impractical.

Pnoise analysis, on the other hand, separates the noise signals from the nonlinear transient behavior, making it much more accurate and robust if you do not need nonlinear noise effects, which as I already pointed out, you almost never do.

-Ken