The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> dfferent freq results in pss vs transient
https://designers-guide.org/forum/YaBB.pl?num=1281170406

Message started by casual on Aug 7^th, 2010, 1:40am

Title: dfferent freq results in pss vs transient
Post by casual on Aug 7^th, 2010, 1:40am

in the QLCVCO (cmos type with bottom bias), freq vs vctrl (vc) are plotted using pss and transient methods.
At the vc=0.8V, transient result has a dip to much lower frequency, however the pss results looks ok. In the instantaneous freq plot (transient sim), it confirms that the vco has settled down.

However, which result is reliable? At the end, i got to simulate whole pll in transient mode.

Title: Re: different freq results in pss vs transient
Post by pancho_hideboo on Aug 7^th, 2010, 2:32am

I can't find out any design issue in your post.

supermoment wrote on Aug 7^th, 2010, 1:40am:

At the vc=0.8V, transient result has a dip to much lower frequency, however the pss results looks ok.

Show me time domain waveforms at vc=0.8V both for PSS and Transient Analysis.

supermoment wrote on Aug 7^th, 2010, 1:40am:

However, which result is reliable?

Result of PSS is reliable.

Tighten accuracy of Transient Analysis, e.g. "reltol".

See http://www.designers-guide.org/Forum/YaBB.pl?num=1237627988/5#5

Title: Re: dfferent freq results in pss vs transient
Post by casual on Aug 7^th, 2010, 4:04am

i have tried 2 methods to confirm that PSS run is more accurate

1) change reltol from 1e-3 to 1-5
2) sim conservative run

the diagram shows the conservative run overlapping with pss result.

thx

Title: Re: different freq results in pss vs transient
Post by pancho_hideboo on Aug 7^th, 2010, 4:08am

supermoment wrote on Aug 7^th, 2010, 4:04am:

the diagram shows the conservative run overlapping with pss result.

I think you can understand you have to pay attention to accuracy of Transient Analysis of Cadence Spectre if you use it for VCO analysis.

Most serious issue is such irregular large deviation from correct value.

I often encounter irregular large deviation from correct value when I use Cadence Spectre.
You could not notice such large deviation, if you simulate only one point such as vc=0.8V.

Even if I use Golden Standard HSPICE, I can observe deviation from correct value if I don't tighten accuracy such as "delmax".
But I never encounter such extraordinal deviation when I use Golden Standard HSPICE.

Again see http://www.designers-guide.org/Forum/YaBB.pl?num=1237627988/5#5

I always set "maxstep" for VCO analysis if I use Cadence Spectre.

Title: Re: dfferent freq results in pss vs transient
Post by sheldon on Aug 7^th, 2010, 8:24pm

Casual,

Can you explain more about your measurement methodology?
1) How long do you allow for the VCO to settle to steady-state?
2) How do you measure period?
Do you measure for one period to calculate the frequency?
Do you average over several periods to calculate the frequency?
Do you use the frequency meter?

The results for both tools should be consistent. However, it is more
difficult to perform the measurement using transient analysis.

Best Regards,

Sheldon

Title: Re: dfferent freq results in pss vs transient
Post by vp1953 on Aug 8^th, 2010, 10:35am

Hi Sheldon,

Why would be period be different with and without averaging for transient analysis (assuming that noise sources are not enabled)?

Title: Re: dfferent freq results in pss vs transient
Post by vp1953 on Aug 8^th, 2010, 11:57am

Hi Sheldon,

Come to think of it, the presence of higher tones in the spectrum can create a difference between a single period and averaged period - is this what you had in mind?

Title: Re: dfferent freq results in pss vs transient
Post by casual on Aug 9^th, 2010, 8:59pm

Seldon,

The first method I use is to measure frequency of the clk after a long settling time about 30ns in the transient analysis since I still new in PSS. You could see from the instantaneous freq waveform that it has settle down to around 4.88GHz. So one freq and average freq will be almost the same (small variation).

I do not know what the freq meter about? is it written in verilogA?

from the study here, found that sweep PSS is the best method to extract vco gain behaviour.
However, in the close loop simulation, it is still running in transient mode. So it is great to check in transient mode (my point of view) and adjust the accuracy setting.

Title: Re: dfferent freq results in pss vs transient
Post by sheldon on Aug 10^th, 2010, 6:22pm

Casual,

With adaptive time stepping, the simulator will not necessarily stop at
the instance in time required for the frequency measurement and as a
result, there will be interpolation error. By averaging the effect of
interpolation error is reduced.

The freq_meter is an element that was created for the noise aware
PLL flow and is in the pllMMLib, (requires Spectre RF license). The
component is used to measure the frequency at the output of a VCO
or PLL in the time domain. It also enables time domain measurement
of phase noise, Direct Plot --> main form --> tran --> PLL Noise PSD.

Best Regards,

Sheldon

Title: Re: dfferent freq results in pss vs transient
Post by casual on Aug 10^th, 2010, 9:56pm

sadly, i do not have noise aware pllflow for mmsim61 ic5.1.41

Quote:

time domain measurement
of phase noise, Direct Plot --> main form --> tran --> PLL Noise PSD.

Could i do this in manual way with the calculator?

Title: Re: dfferent freq results in pss vs transient
Post by sheldon on Aug 11^th, 2010, 5:05am

Casual,

No, this functionality is not supported in the calculator. I suspect that
the approach is similar to the approach described in Ken's paper on
PLL modeling and simulation.

http://www.designers-guide.org/Analysis/PLLnoise+jitter.pdf

Best Regards,

Sheldon

Title: Re: dfferent freq results in pss vs transient
Post by Ken Kundert on Aug 11^th, 2010, 7:51am

You could use a verilog-a module to accurately measure the frequency. You can probably use the one on this site (http://www.designers-guide.org/VerilogAMS/functional-blocks/periods.tgz). You can plot the internal variable 'freq' to show frequency versus time.

-Ken

Title: Re: dfferent freq results in pss vs transient
Post by casual on Aug 12^th, 2010, 1:41am

thank you very much.