The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> Evaluation Of Cadence Flexible Balance
https://designers-guide.org/forum/YaBB.pl?num=1170364605

Message started by toolewh on Feb 1^st, 2007, 1:16pm

Title: Evaluation Of Cadence Flexible Balance
Post by toolewh on Feb 1^st, 2007, 1:16pm

I have recently just started an evaluation of Cadence Flexible Balance (FB) and am somewhat perplexed by it. I have considereable experience with other Harmonic Balance simulators such as RFDE and GoldenGate. However, I was very surprised to see Cadence PSS or QPSS simulation, with the Flex balance button checked, run a time domain simulation before launching the Flex Balance analysis. Why does it do this? Shouldn't it launch the frequency domain simulation right away like the other HB simulators do. This seems to defeat the purpose of simulating in the frequency domain especially if I have 2 or more closely spaced tones. Does anyone have experience with Cadence FB? How do the simulation times compare to say RFDE and/or GoldenGate. How is the convergence of Cadence FB?

Title: Re: Evaluation Of Cadence Flexible Balance
Post by sheldon on Feb 1^st, 2007, 5:25pm

Toolwh,

Flexible balance inherits the default simulation setup from the
time domain, Shooting-Newton analysis. If you want to run pure
Harmonic Balance analysis, then set tstab=0. It may be useful
to review the discussion Harmonic Balance analysis section of
the SpectreRF Simulation Option Theory manual.

Best Regards,

Sheldon

Title: Re: Evaluation Of Cadence Flexible Balance
Post by pancho_hideboo on Feb 2^nd, 2007, 2:00am

Hi.

SpectreRF's FB(HB) Analysis speed is practical level for small block such as LNA, Mixer.
Maybe it is slow compared to ADS's HB especially in case of multitone FB(QPSS).
However it is accceptable level.

But for large or high nonlinear circuit, SpectreRF's FB(HB) Analysis can't give reasonable results.

With nonzero tstab, I expect SpectreRF's FB(HB) Analysis to work like as transient assistted HB analysis in ADS.
But it is no helpful same as tstab in shooting QPSS.

Now ADS's HB is very robust.

Title: Re: Evaluation Of Cadence Flexible Balance
Post by toolewh on Feb 2^nd, 2007, 6:22am

Thanks Sheldon

I had not set a value for tstab but, as the guide specifies, it defaults to "one cycle of the signal period". Setting tstab to 0, as you pointed out, forces it to run a pure HB sim. I had read the relevent section before but missed the bit about tstab. I am still not clear what it does with the transient when tstab is not set to 0. I assume it uses it as an intial guess for the HB solution, similar to what "Transient Assisted Harmonic Balance" (TAHB) does in RFDE and GoldenGate. However, in RDFE and GoldenGate TAHB, it runs the trasient only for the first value in a swept sim, not every swept value. I also don't like the fact that for a two tone HB PSS you still have the set the beat frequency, and, if the tone are close together, you have to set a large number of harmonics, which significantly slows things down. It appears that QPSS is the better sim to run, which, is what they suggest in the guide.

Bill

Title: Re: Evaluation Of Cadence Flexible Balance
Post by sheldon on Feb 2^nd, 2007, 7:34am

Bill,

Yeah, the comment is definitely in the fine print as it were!
Also thought the comments also said that the DC operating point
is used when tstab=0.

I think that you are correct, in that HB QPSS is appropriate for
two-tone simulations. If you use HB PSS for two tone simulations,
then you need a lot of tones which can impact performance. There
is an option that will allow you to re-use the initial condition and it
might take care of the re-running the initial transient issue during
sweeps. As I remember the defaults is to re-generate the initial
conditions. It might be useful to talk to Cadence and get some help
working through these issues.

Best Regards,

Sheldon

Title: Re: Evaluation Of Cadence Flexible Balance
Post by toolewh on Feb 2^nd, 2007, 7:53am

Thanks again Sheldon.

I am working with my local Cadence apps engineer on these issues but things can be slow with them. I also wanted to find out what and how real RF designers are using the tool and what their impressions are. My first impression is that the HB option in cadence coupled with QPSS is a significant improvement over the "old" PSS sims. Is the HB engine robust enough to handle large circuits for say a receiver chain (or a good portion of it) and how does it compare to RFDE or GoldenGate is another question. I need to bash away at it and see where the tool breaks. Let the fun begin.

Bill

Title: Re: Evaluation Of Cadence Flexible Balance
Post by Andrew Beckett on Feb 7^th, 2007, 4:07am

Harmonic Balance in SpectreRF is robust with large circuits (I've simulated a complete receiver chain using it, including blocks with parasitics included from parasitic extraction - e.g. ). However, the thing to remember is that harmonic balance is best with weakly non-linear circuits - if you're dealing with larger amounts of nonlinearity then you can still use shooting methods - which are a big strength of SpectreRF. Shooting methods are generally a robust way of handling strongly nonlinear circuit problems.

So the real benefit is that you have a choice - you can pick the right engine depending on the situation.

Having a transient assist on each point in the sweep makes sense - since as you increase power (say), the steady state changes considerably, and so the initial stab you got at low power is not going to be a good guess for higher powers.

Regards,

Andrew.

Title: Re: Evaluation Of Cadence Flexible Balance
Post by toolewh on Feb 7^th, 2007, 5:04am

Thanks Andrew.

It is encouraging to hear that the Cadence HB is robust enough to simulate a receiver chain. I am getting that sense as I continue to play with the simulator. Your comment about using HB for weakly nonlinear circuits is always the case no matter what HB simulator you use. I have had reasonable success using RFDE and GoldenGate for circuit well beyond 1dB compression point but you need a lot of harmonics and transient assist helps a fair bit. For those types of sims it is probably best to use the shooting method but it can be done in HB with some care. As you said, with Cadence, it is nice to have the choice to choose the appropriate engine.

As far as transient assist HB (TAHB) goes, I would agree that the initial tstab at low powers will not be valid at higher powers but most other HB simulators will use the tranient solution the first point only in the sweep and then use the previous HB solution for the next point and so on. They usually have the option to run TAHB for each point but in 95% of any cases I simulated, it was not needed. It would be nice if Cadence had the option. I have asked my local Cadence Apps engineer about this and he is checking into it.

Thanks again.

Bill

Title: Re: Evaluation Of Cadence Flexible Balance
Post by Andrew Beckett on Feb 8^th, 2007, 10:57am

Bill,

Actually you can just do the transient assist on the first point of the sweep. If you set the pss or qpss parameter "restart=no" then it will re-use the previous solution during the sweep as a starting point.

So, you can do:

No transient assist - use tstab=0
Transient assist on initial sweep point - use restart=no
Transient assist on every sweep point - use restart=yes (the default)

Regards,

Andrew.

Title: Re: Evaluation Of Cadence Flexible Balance
Post by toolewh on Feb 8^th, 2007, 11:03am

Andrew

Perfect, it worked like a charm. Thanks for the info.

Bill

Title: Re: Evaluation Of Cadence Flexible Balance
Post by nandy on Jul 25^th, 2008, 5:48am

How do I set tstab=0? In the pss analysis window, Im asked to enter "Additional time for stabilization". This is over and above the tstab decided by spectre.

I want to set the absolute value of tstab according to my choice in order to reduce simulation time, because spectre makes a poor estimate and the simulation runs for too long

Title: Re: Evaluation Of Cadence Flexible Balance
Post by sheldon on Jul 25^th, 2008, 10:08am

Nandy,

Could a little more information about what you are trying to
do? I believe that the default tstab is controlled by the designer
when the set the fundamental period for PSS analysis. If the
transient is taking too long, it indicates that there is an issue
with the fundamental period you have specified.

Best Regards,

Sheldon

Title: Re: Evaluation Of Cadence Flexible Balance
Post by Ken Kundert on Jul 25^th, 2008, 11:02am

Remember that SpectreRF will run a transient analysis from time t=0 until the independent sources become periodic before starting the PSS analysis (after all, the circuit will only have a periodic steady-state response once the sources become periodic). Perhaps you should examine all of your sources and make sure they all become periodic or constant valued in a reasonable amount of time.

-Ken

Title: Re: Evaluation Of Cadence Flexible Balance
Post by nandy on Jul 25^th, 2008, 11:34am

Hi

Im running pss as a forerunner to pnoise of a 10G LC oscillator I have designed. I notice from the stored transient response that the swing becomes uniform after only 10ns, but pss runs transient for 280ns (I set the additional stabilization time to be 0). I observed that the amount of time that pss intends to run is established as soon as the simulation starts, which means pss isnt doing any dynamic checks to see if response has settled. It decides upon the run time and sticks to it no matter what. I tried changing the methods by which tstab is set, like trap, traponly, euler etc. but all of them run for the same amount of time.

Title: Re: Evaluation Of Cadence Flexible Balance
Post by nandy on Jul 25^th, 2008, 11:36am

O, and for the simulations I had set the estimated frequency to be 9.5GHz, and pss locked it at 9.7G

Title: Re: Evaluation Of Cadence Flexible Balance
Post by Ken Kundert on Jul 25^th, 2008, 4:08pm

Have you checked your independent sources?

-Ken

Title: Re: Evaluation Of Cadence Flexible Balance
Post by nandy on Jul 27^th, 2008, 4:37am

Yes I have. All the independent sources are dc voltages or currents. Anyway, pss-oscillator analysis doesnt allow for any time varying independent source.

Title: Re: Evaluation Of Cadence Flexible Balance
Post by pancho_hideboo on Jul 27^th, 2008, 7:05am

nandy wrote on Jul 27^th, 2008, 4:37am:

Yes I have. All the independent sources are dc voltages or currents.

Show me your netlist regarding
all sources
options
pss statement

And show me logfile.

nandy wrote on Jul 27^th, 2008, 4:37am:

Anyway, pss-oscillator analysis doesnt allow for any time varying independent source.

What do you mean ?
We can use time varying independent source such as step signal sources or a damped sinusoid voltage sources in Cadence Spectre Autonomous PSS.
This is also true for Agilent ADS and GoldenGate Autonomous HB analysis where transient assisted HB analysis is invoked.