The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> RF Simulators >> Ill Convergence of PSS when Hysteresis Buffer is included
https://designers-guide.org/forum/YaBB.pl?num=1282752641

Message started by pancho_hideboo on Aug 25^th, 2010, 9:10am

Title: Ill Convergence of PSS when Hysteresis Buffer is included
Post by pancho_hideboo on Aug 25^th, 2010, 9:10am

When target circuit include Hysteresis Buffer such as Schmitt Trigger Buffer,
it is very difficult to get convergence of periodical steady state analysis based on HB or Shooting-Newton approaches.

This issue is not concerned with specific vendor's simulator engine.

Is there anyone who have same experience ?
How do you overcome this issue ?

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by RFICDUDE on Aug 26^th, 2010, 4:45pm

Hi Pancho,

I suspect, although I do not know for sure, that hysteresis is a form of the dreaded "hidden state" problem that plagues steady-state simulators. It would seem to be true that hysteresis certainly depends on a past value of a state-variable. If this is true, then it may be the case that this problem cannot be solved even with transient assistance (again I really don't know for sure).

I would hope that Ken Kundert could provide some insight here?

The Dude

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by pancho_hideboo on Aug 27^th, 2010, 8:29am

RFICDUDE wrote on Aug 26^th, 2010, 4:45pm:

I suspect, although I do not know for sure, that hysteresis is a form of the dreaded "hidden state" problem that plagues steady-state simulators.

I don't think so.

If we realize "Hysteresis" by Verilog-A, we will encounter "hidden state" problem.
But my circuits are real transistor level circuits, so there is no "hidden state" problem.

If we use "@cross()" to describe "Hysteresis" in Verilog-A, such module can't work in PSS analysis based on HB approach,
although it can work in PSS analysis based on Shooting-Newton approach.

RFICDUDE wrote on Aug 26^th, 2010, 4:45pm:

It would seem to be true that hysteresis certainly depends on a past value of a state-variable.

It is true for real transistor level circuits.

"Charges of Capacitor" are "State-Variables".

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by vp1953 on Aug 27^th, 2010, 4:20pm

Hi Pancho,

Does increasing the number of harmonics (substantially) make any difference? I am thinking something like a schmitt trigger with abrupt output transition edges (almost like a square wave) may need much more harmonics to adequately cover the spectrum

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by pancho_hideboo on Aug 27^th, 2010, 10:29pm

See http://www.designers-guide.org/Forum/YaBB.pl?num=1237890768/10#10

vp1953 wrote on Aug 27^th, 2010, 4:20pm:

Does increasing the number of harmonics (substantially) make any difference?

I usually use 31 as Harmonics in HB.
I also tried 101 Harmonics in HB. But convegence can't be improved or succeeded.

Harmonics of 31 is generally enough for master large signal analysis, although it can not be enough for slave small signal analysis.

I think ill-convergence is due to a deadzone of Hysteresis Buffer.
There could be an oscillative behavior if an input signal is small around a deadzone.

pancho_hideboo wrote on Aug 25^th, 2010, 9:10am:

When target circuit include Hysteresis Buffer such as Schmitt Trigger Buffer,
it is very difficult to get convergence of periodical steady state analysis based on HB or Shooting-Newton approaches.

Again see start descriptions of this thread.

I tried following engines.
- HB with Frequecy Domain Preconditioner in Agilent GoldenGate
- HB with Time Domain Preconditioner in Agilent GoldenGate
- TB(Time-Balance) in Agilent GoldenGate
- TS(Time-Shooting) in Agilent GoldenGate
- TB Assisted HB in Agilent GoldenGate
- TS Assisted HB in Agilent GoldenGate
- Transient Assisted HB with Frequecy Domain Preconditioner in Agilent GoldenGate
- Transient Assisted HB with Time Domain Preconditioner in Agilent GoldenGate

- PSS based on HB in Cadence Spectre
- PSS based on Shooting-Newton in Cadence Spectre

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by vp1953 on Aug 30^th, 2010, 3:28pm

Hi Pancho,

Just thinking out loud, maybe you have already tried this. In PSS with Cadence Spectre (with either of the methods), setting the beat frequency (using manual adjustment) to something that is substantially lower than the frequency of the input signal and at the same time setting the number of harmonics so that it covers 31x of the input signal frequency.

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by RFICDUDE on Aug 31^st, 2010, 2:57am

Are you simulating just the trigger circuit by itself (i.e. are there other circuits also driven by the periodic source)?

Is the circuit behavioral or transistor level?
I only ask because I could play around with a behavioral circuit to see the circuit can be tricked into converging by adding a similar driven circuit without hysteresis included.
It is doubtful that my idea will work, but it is something to try.

I think (again I don't know for sure) that the hysteresis creates an ill condition for determining the periodicity of the solution because it perturbs the zero crossings when it makes a decision. You may have to think of ways (tricks) to make the output of the circuit periodic on average (I am not sure if this can easily be done).

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by pancho_hideboo on Aug 31^st, 2010, 4:37am

vp1953 wrote on Aug 30^th, 2010, 3:28pm:

In PSS with Cadence Spectre (with either of the methods),
setting the beat frequency (using manual adjustment) to something that is substantially lower than the frequency of the input signal and at the same time setting the number of harmonics so that it covers 31x of the input signal frequency.

It's never appropriate and never practical and never helpful, if we consider a common divisor frequency between input signal and oscillative signal.

If I dare to try alternative analyses, appropriate candidates are :
- Autonomous Multitone Steady State Analysis such as Autonomous Multitone HB(Autonomous-QPSS)
- One Driven Tone Envelope Analysis

There is no "beat" in analysis statement for PSS of Cadence Spectre.
Don't use a term of "beat", if you mean PSS of Cadence Spectre.

See "spectre -h pss", "beat frequency" of PSS setting UI is not beat frequency.
It is a "fundamental frequency" not beat frequency.

See http://www.designers-guide.org/Forum/YaBB.pl?num=1268969030/7#7

RFICDUDE wrote on Aug 31^st, 2010, 2:57am:

Are you simulating just the trigger circuit by itself (i.e. are there other circuits also driven by the periodic source)?

Hysteresis buffer is no more than one small part of circuits in my simulation.

RFICDUDE wrote on Aug 31^st, 2010, 2:57am:

Is the circuit behavioral or transistor level?

Full transistor level.

pancho_hideboo wrote on Aug 27^th, 2010, 8:29am:

RFICDUDE wrote on Aug 26^th, 2010, 4:45pm:

I suspect, although I do not know for sure, that hysteresis is a form of the dreaded "hidden state" problem that plagues steady-state simulators.

I don't think so.
If we realize "Hysteresis" by Verilog-A, we will encounter "hidden state" problem.
But my circuits are real transistor level circuits, so there is no "hidden state" problem.

RFICDUDE wrote on Aug 31^st, 2010, 2:57am:

I only ask because I could play around with a behavioral circuit to see the circuit can be tricked into converging by adding a similar driven circuit without hysteresis included.

If I remove or make disable Hysteresis buffer, simulation can converge without any difficulty.

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by Ken Kundert on Sep 1^st, 2010, 5:43am

There is nothing about hysteresis that in inherently problematic for PSS analysis. I suspect that there is something else that is causing the problem. If you simulate the circuit in Spectre, try using transient analysis with strobing and set the strobe period equal to the period of the expected fundamental tone. Then examine all the strobed signals and make sure they all appear to become constant valued. If they do not then your circuit does not have a periodic solution with that period. Check the supply currents first, they are most likely to show the problem.

-Ken

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by pancho_hideboo on Sep 1^st, 2010, 6:00am

Ken Kundert wrote on Sep 1^st, 2010, 5:43am:

There is nothing about hysteresis that in inherently problematic for PSS analysis.

I don't think so.

pancho_hideboo wrote on Aug 27^th, 2010, 10:29pm:

I think ill-convergence is due to a deadzone of Hysteresis Buffer.
There could be an oscillative behavior if an input signal is small around a deadzone.

Ken Kundert wrote on Sep 1^st, 2010, 5:43am:

Check the supply currents first, they are most likely to show the problem.

There is no problem in supply currents in my test case which include hysteresis buffer.

An oscillation could occur between Vdd and Drain of MOSFETs.
See the following case, although it is not hysteresis buffer.
http://www.designers-guide.org/Forum/YaBB.pl?num=1221012939

Even if I can observe oscillation in supply current for Transient Analysis,
driven HB Analysis can be converged without any difficulty in the above case which does not include any hysteresis buffer.
Also see the followings.
http://www.designers-guide.org/Forum/YaBB.pl?num=1243078889
http://www.designers-guide.org/Forum/YaBB.pl?num=1234428781/7#7

But ideal voltage source having zero impedance is used in my test case which include hysteresis buffer.
So there is no oscillation between Vdd and Drain of MOSFETs.

And convergence can be improved if I loose accuracy setting, for example from moderate to liberal in Shooting-Newton PSS of Cadence Spectre.

Title: Re: Ill Convergence of PSS when Hysteresis Buffer is included
Post by Ken Kundert on Sep 3^rd, 2010, 6:45pm

Again, there is nothing about hysteresis itself that is inherently problematic, as long as you generate a good starting point, which would be done naturally if you use an initial transient interval. That does not mean that there is something else about your hysteretic buffer that is causing the problem. I don't know what you mean by a dead zone, but I cannot see why a deadzone in general would be problematic. If the buffer were unstable, then that would be problematic because it would result in a non-periodic solution. That is why I recommended the strobing. With strobing you will be able to easily see if your circuit has a periodic solution for the period you chose.

-Ken