A transient simulation of the step response is the most reliable way of determining stability. It should always be done to verify the stability of the circuit, even if the circuit was designed with small-signal analyses. The disadvantage of the transient simulation is that it usually takes much longer than a small-signal analysis and that it is more difficult to extract a number like the phase margin from the results.

Frank Wiedmann wrote on Oct 18^th, 2009, 12:51am:

    Thanks for your suggestion. I still have some question about the stability judging by the transient simulation. For opamp simulation, input is a square wave using vpulse source in spectreRF, and if no ring in the rise time at the output. It is ok! For RF circuit, usually, there is an input match network including capacity for match and dc separation in amplifier and with an output match network after it. In this case, should I still use vpulse source in series with a 50 resistance（vpulse source has no output resistance）, and add the vpulse in the input match and see the wave after the output match? If there is no ring, it is ok?
     Another question is about the stability at the time of power on, should I simulation it to confirm the stability? If I should simulation the state of power on, what input signals source should I choose?

You should always simulate the circuit configuration(s) that you will have in your application, including matching networks and parasitic components from interconnections. This applies for all types of simulations, not only stability simulations.

The same is true for simulations of power-up. Find out the output signal of the power source (and the power network, inlcuding parasitics) at power-up (rise time, possibly overshoot or ringing) and use it in your simulation. If you are designing a catalog part, use the worst-case values (shortest rise time etc.) from the datasheet. Also don't forget to check that you don't exceed the maximum rated values of your components during power-up (and power-down).

I would guess so. If you are concerned about stability at certain points inside your circuit, you might also want to probe the corresponding internal nodes and check for ringing there. Depending on the structure of the circuit, not all internal ringing may be visible at the output.

Your original question was about large-signal stability. Just applying a square wave and looking for ringing is insufficient. In doing so you are likely just replicating a small-signal stability test. If a circuit is unstable in a small-signal sense, then it generally becomes very obvious in transient, because the circuit breaks into oscillation. However, a circuit that is stable in a small-signal sense and unstable under large signals will not just break into oscillation in transient. You need to provide a signal that will get it into is unstable region. For example, a cohort designed a track and hold that had a feedback loop that contained two op amps. Under small signal conditions, it worked fine. The second op amp was configured as a unity gain buffer that had a bandwidth well beyond the unity gain frequency of the loop. But once that opamp entered slew-rate limiting, its effective bandwidth dropped by several orders of magnitude making the loop unstable. The only way he could see this in simulation was by providing the right large stimulus that put that opamp into slew limiting. So in general, there is no one particular test you can run that if your circuit passes you can say with certainty that your circuit is large-signal stable. Instead, you have to know your circuit and know what states it can get into and test each of them.

-Ken

Ken Kundert wrote on Oct 19^th, 2009, 12:39pm:

Dear  Ken ,thank you for your advise! thank you for you example for my understanding of large signal unstable!

What you need to do is just to simulate the large signal [s] parameter (LSSP), then from the large signal [s], you can easily calculate stability factor and so on.