Designer’s Guide Community

Pages: 1 2

.pss Simulation Results Vary in Different runs in Cadence (SpectreRF) (Read 3785 times)

Maryam Junior Member Offline Posts: 15	Re: .pss Simulation Results Vary in Different runs in Cadence (SpectreRF) Reply #15 - Sep 06^th, 2008, 5:23pm By the way, in general, the pss simulation of my circuit does not always converge. I keep simulating it several times and change the integration methods (mostly default, gear2, gear2only and sometimes trap and traponly) until I get convergence.
Back to top	IP Logged

pancho_hideboo

Senior Fellow

Offline

Posts: 1424
Real Homeless

Re: �.pss Simulation Results Vary in Different runs in Cadence (SpectreRF)
Reply #16 - Sep 6^th, 2008, 10:05pm

Maryam wrote on Sep 6^th, 2008, 5:19pm:

Yes, the results were different in these two cases,
in terms of the duty cycle of transistors being in triode and the amplitude of the voltages.

I can't believe that remarkable discrepancies exist although convergence succeed with reasonable ConvNorm.
Show me result's graph which show discrepancy with each ConvNorm.

Maryam wrote on Sep 6^th, 2008, 5:19pm:

I simulated a transient with moderate settings; the results were fine.

Make your issues correct and clear.

(1) Transient Analysis of errpreset=liberal --->> unreasonable results

(2) Transient Analysis of errpreset=moderate --->> reasonable results

(3) Transient Analysis of errpreset=conservative --->> reasonable results

(4) PSS Analysis of errpreset=conservative --->> often give unreasonable results

From your description I understand that transient(moderate) analysis give reasonable results while pss analysis often give unreasonable results.
Is this correct ?

How about PSS(moderate) with all default setting ?
Compare two cases, transient(moderate) and pss(moderate).
Is there remarkable difference between two ?

Maryam wrote on Sep 6^th, 2008, 5:23pm:

By the way, in general, the pss simulation of my circuit does not always converge.
I keep simulating it several times and change the integration methods (mostly default, gear2, gear2only and sometimes trap and traponly)
until I get convergence.

Try to use Harmonic Balance PSS not Shooting Newton PSS.

Kita━━━━━━(ﾟ∀ﾟ)━━━━━━ !!!!!
http://www7.plala.or.jp/ungeromeppa/flash/kita.html
http://www.youtube.com/watch?v=mjIxGh55bMM&feature=related

IP Logged

Maryam

Junior Member

Offline

Posts: 15

Re: .pss Simulation Results Vary in Different runs in Cadence (SpectreRF)
Reply #17 - Sep 6^th, 2008, 11:14pm

(1) Transient Analysis of errpreset=liberal --->> totally unreasonable results (different operating regions for the transistors)

(2) Transient Analysis of errpreset=moderate --->> reasonable results

(3) Transient Analysis of errpreset=conservative --->> reasonable results

Although it took a long time for the two transients to settle so I didn't compare the exact steady state magnitude of cases 2 and 3, but the results reasonably agreed.

(4) PSS Analysis of errpreset=conservative --->> the results don't agree. Usually they were close, in some cases they were different by about 0.2V difference in amplitude, and in a few cases, I got totally different results.

By changing the design parameters, I face different situations, so I cannot generate the same big discrepancy now since I have changed the design parameters.

"From your description I understand that transient(moderate) analysis give reasonable results while pss analysis often give unreasonable results.
Is this correct ?"

Yes. However, since the transient analysis takes a very long time to simulate and settle, I don't run it that often. So, I cannot say that I always get the exact same result from moderate and conservative transient. But, so far I didn't get unreasoonable results from transient if it converged.

"Try to use Harmonic Balance PSS not Shooting Newton PSS."

I thought Spectre uses Shooting method and ADS uses harmonic balance.
Can I do harmonic balance in Spectre?

Thank you very much.

« Last Edit: Sep 7^th, 2008, 3:04am by Maryam »

IP Logged

pancho_hideboo

Senior Fellow

Offline

Posts: 1424
Real Homeless

Re: �.pss Simulation Results Vary in Different runs in Cadence (SpectreRF)
Reply #18 - Sep 6^th, 2008, 11:20pm

Maryam wrote on Sep 6^th, 2008, 11:14pm:

(4) PSS Analysis of errpreset=conservative --->> the results don't agree.
Usually they are close, in some cases they are different by about 0.2V difference in amplitude,
and in a few cases, I got totally different results.
By changing the design parameters, I face different situations, so I cannot generate the same big discrepancy now
since I have changed the design parameters.

When you see large diffences, does PSS converge with small ConvNorm truely ?
I think you are missing any caution or message in Spectre's logfile.

Anyway your circuits is difficult in convergence with Shooting Newton PSS.
Try to use Harmonic Balance PSS.

If a sinusoidal input signal is large enough near or over gain compression point,
Shooting Newton PSS often doesn't converge at all on the contrary to Cadence advertisement of Shooting Newton PSS.

I've never used SpectreRF mainly for RF circuit.
http://www.designers-guide.org/Forum/YaBB.pl?num=1162921214/0#0
http://www.designers-guide.org/Forum/YaBB.pl?num=1170364605/2#2
http://www.designers-guide.org/Forum/YaBB.pl?num=1194765058/1#1
http://www.designers-guide.org/Forum/YaBB.pl?num=1208334972

I have many simulators working on Cadence ADE, so I don't have to insist on Cadence SpectreRF at all.
If you have other simulators such as ADS, GoldenGate, HspiceRF, SmartSpiceRF, EldoRF, etc., try other simulators.
Maybe you will get far good simulation performance than Cadence Spectre.
http://www.designers-guide.org/Forum/YaBB.pl?num=1210062036

Maryam wrote on Sep 6^th, 2008, 11:14pm:

I thought Spectre uses Shooting method and ADS uses harmonic balance.
Can I do harmonic balance in Spectre?

Yes, Cadence Spectre have Harmonic Balance Analysis since Version 6.0USR1.

See the following.
http://www.designers-guide.org/Forum/YaBB.pl?num=1125929808
http://www.designers-guide.org/Forum/YaBB.pl?num=1205223090

« Last Edit: Sep 7^th, 2008, 1:50am by pancho_hideboo »

Kita━━━━━━(ﾟ∀ﾟ)━━━━━━ !!!!!
http://www7.plala.or.jp/ungeromeppa/flash/kita.html
http://www.youtube.com/watch?v=mjIxGh55bMM&feature=related

IP Logged

Maryam

Junior Member

Offline

Posts: 15

Re: .pss Simulation Results Vary in Different runs in Cadence (SpectreRF)
Reply #19 - Sep 7^th, 2008, 3:02am

"When you see large diffences, does PSS converge with small ConvNorm truely ?"

The two cases of which I attached log file, had different results, while at the last step, conv norm wassmaller than 1.

As far as I know, for circuits with large amplitudes and switching behavior of the transistors, actually HB has convergence difficulty (the extreme case is having logic gates).

1) liberal pss (default setting) did not converge at all.
2) moderate pss (default setting) converged and the discrepancy between the results was larger than conservative pss .
3) conservative pss (default setting) converged and had inconsistent results.

4) liberal transient converged. I could get consistent results with default settings this time while before it did not give reasonable results!!
5) moderate transient (default setting) gave consistent results.
6) conservative transient (default setting) gave consistent results.

The results of cases 5 and 6 (moderate transient and conservative transient) agreed.

IP Logged

pancho_hideboo

Senior Fellow

Offline

Posts: 1424
Real Homeless

Re: �.pss Simulation Results Vary in Different runs in Cadence (SpectreRF)
Reply #20 - Sep 7^th, 2008, 5:36am

Maryam wrote on Sep 7^th, 2008, 3:02am:

As far as I know, for circuits with large amplitudes and switching behavior of the transistors, actually HB has convergence difficulty (the extreme case is having logic gates).

No. What you know is just same as Cadence advertisement.
And current HB analysis of Agilent ADS or GoldenGate is very robust.

If an input signal or output signal are pulse-shaped wave having sharp edge, shooting newton is superior than HB analysis.

But if an large sinusoidal input signal is added to amplifier input, for example, for evaluation of gain compression of amplifier, shooting newton can't converge at all while HB analysis can converge.
I can't evaluate gain compression characteristics using Shooting Newton PSS of Cadence Spectre.

For common classical RF evaluation cases where sinusoidal signal drive are used, Shooting Newton PSS of Cadence Spectre is inferior than HB analysis.
And I always encounter convergence problem of Shooting Newton PSS of Cadence Spectre.

SpectreRF can work for simple circuits which behave gently and is not robust at all for actual circuits on the contrary to Cadence Advertisement.

Maryam wrote on Sep 7^th, 2008, 3:02am:

1) liberal pss (default setting) did not converge at all.
2) moderate pss (default setting) converged and the discrepancy between the results was larger than conservative pss .
3) conservative pss (default setting) converged and had inconsistent results.

In your case, shooting newton PSS with any errpreset can't be reliable.

Anyway try HB analysis of Cadence Spectre or other RF simulator.

« Last Edit: Sep 7^th, 2008, 4:36pm by pancho_hideboo »

Kita━━━━━━(ﾟ∀ﾟ)━━━━━━ !!!!!
http://www7.plala.or.jp/ungeromeppa/flash/kita.html
http://www.youtube.com/watch?v=mjIxGh55bMM&feature=related

IP Logged

Pages: 1 2

‹ Previous topic | Next topic ›