June 02, 2001
Question: Is a conditionally stable loop considered acceptable?
Original Question: I have a question regarding conditional stability. I keep having problems with low frequency zero phase crossings in discontinuous current-mode-controlled flybacks. What is considered an acceptable loop? I am familiar with converting the conditionally stable bode plots to a Nyquist plot, but the term conditionally stable only leaves me feeling conditionally certain in my design. Is this common or am I doing something fundamentally wrong? J. P. 06/02/2001
Answer: Strange. A flyback operating in the discontinuous-current mode should have a single dominant pole that rolls it off with a slope of one. The only thing that would cause it to behave this way that I know of would be poles and zeros added into the control amplifier to give this unwanted response. One possible explanation is that the flyback operates in both the continuous and discontinuous mode and was stabilized for the continuous mode, which is a more complex problem since it is a second-order system with a right-half-plane zero. I'm not sure about this, but I seem to remember that there is a drop in gain as well as a change in dynamics when you transition from the continuous-mode to the discontinuous mode. This drop in gain, combined with the poles and zeroes needed to stabilize operation in the continuous-current mode, may be the cause of the problem.
An first glance, there is nothing wrong with a conditional stable system where the gain remains greater than zero dB while the phase shift dips beyond 180 degrees and comes back before the zero dB crossing. When you do a worse-case design of a buck converter with voltage-mode control, conditional stability can often happen.
At second glance, you do not want a conditionally stable loop. The reason is that there are several things that can saturate an amplifier in the feedback loop. This often happens during transients when the control loop commands maximum or minimum duty cycle. When an amplifier in the feedback loop saturates, the effect is often a reduction in gain of the loop and this can cause the gain to drop to zero dB when the phase shift is 180 degrees and the system becomes unstable and may not recover. The first couple of buck converters I ever designed were conditionally stable, and by dumb luck they were never a problem in the field. Once I understood the problem, which I have seen happen to others many times (their power supplies went unstable in the field under certain operating conditions due to conditional stability), I made sure that none of my loops were ever conditionally stable again.
Posted by Jerrold Foutz at June 2, 2001 04:07 PM