Help! My power supply unit is unstable – part 2

In the first installment of this series, I stated that there are many reasons for switched-mode power supply (SMPS) instability, only one of which is that the control loop has insufficient gain or phase margin. In this installment, I will offer some tips about identifying and curing subharmonic oscillations in peak current mode (PCM)-controlled SMPS systems and talk briefly about input-filter oscillations. 

Subharmonic oscillations

There is a well-known, inherent instability in continuous conduction mode (CCM) PCM control loops when they operate at duty cycles greater than 50%, as shown in Figure 1. Discontinuous conduction mode (DCM), transition mode (TM), average current mode (ACM) and voltage mode-controlled (VMC) systems are not susceptible to this type of instability. But be careful – because DCM, TM, ACM and VMC systems often use PCM control when they operate in current limit.

Figure 1: Subharmonic oscillations

Diagnosis and solution

Subharmonic oscillations appear as large changes of duty cycle from cycle to cycle. They usually persist because the average duty cycle remains greater than 50%, but they can appear transiently if a load change causes the controller to run at a more than 50% duty cycle for a few cycles. It’s also worth noting that without slope compensation, current perturbations take longer and longer to die out as the duty cycle increases towards 50%. Here is a short list of the behaviors you might see.

  • Does the problem disappear at duty cycles less than 50%? If so, the solution is to correct the amount of slope compensation.
  • Increase the inductor value so that you need a slower slope compensation ramp.
  • Reduce the loop bandwidth – loop bandwidths that are more than about one-fourth of the switching frequency can become unstable if subharmonic oscillations become established.
  • It may be possible in some cases to change the transformer turns ratio or the operating range of the SMPS so that it never exceeds a 50% duty cycle.

Input-filter oscillations

Most power supplies present a constant power load to their inputs and therefore have a negative incremental input resistance. This means that the input current will decrease as the input voltage increases. In an offline power factor correction (PFC) stage, the current control loop forces the system to emulate a positive resistance at line frequencies so that the input current follows the sinusoidal shape of the input voltage. But the negative input resistance behavior is present at frequencies beyond the control loop crossover.

DC/DC and offline AC/DC converters will normally have some form of input filter like that shown in Figure 2. This filter is necessary to meet conducted electromagnetic interference (EMI) requirements but it can oscillate under some circumstances if not designed correctly. This topic has been widely discussed in literature, but the summary rule is simple enough: the output impedance of the filter must be less than the input impedance of the converter at all frequencies.

Figure 2: Typical AC/DC converter input filter, with the SMPS input impedance in green, the undamped filter output impedance in red and the damped filter output impedance in blue

Diagnosis and solution

The simplest way to identify an input-filter oscillation is to remove the input filter by short-circuiting the input-filter inductors. The filter will normally oscillate at one or other of the resonances of the filter. These resonances are normally in the range between 1kHz to 10kHz depending on the filter design. Curing input-filter oscillations requires modifying the input filter to reduce its output impedance while maintaining its effectiveness. Here are the two things to try:

  • Change the L and C values to reduce the impedance.
  • Add damping resistors to reduce the filter impedance at its resonant frequencies (also called the Q factor). You can compare the red (undamped) and blue (damped) traces in Figure 2.

So far, I have discussed classic feedback loop instability, subharmonic oscillations and input-filter oscillations. In the next installment, I’ll look at oscillations caused by remote sensing connections.