I will start by saying that I am a digital engineer, not a power supply engineer, yet am writing about stability testing a switching power supply. There are no doubt some reading this with more experience than I, so feel free to comment below.
So, why do we need to tune and stability-test a switching power supply? Basically, we would like our power supply to respond to changing loads quickly enough to avoid dropping out of our acceptable output specification, but not so quickly that it oscillates or rings. Using simulation and math tools available from the power supply controller vendor, we can make a good first stab at compensation network values for the supply, but then we must test a physical device, and make whatever adjustments are necessary based on our final design.
The design I am testing is based on the Linear Tech LTC3731 three-phase synchronous buck controller. An example circuit from the data sheet is shown below. I am shooting for similar values, but this diagram shows the components of interest. A more detailed diagram of the control loop can be found in the data sheet. Linear Tech, as well as other vendors, also offer demo boards that can be modified to suit your tastes and allow you to prototype your design before you build your own.
How do we test for stability?
The first thing we want to do before we get very far into evaluating our prototype circuit is test the response of the power supply with a network analyzer. In the design phase, it's a good idea to put a 50Ω resistor in the voltage divider (near the 9.09kΩ), so that we can attach a transformer to allow the network analyzer to inject signal into the control loop to verify performance. The network analyzer can inject a sweep into the controller to perturb the loop, and measure the gain and phase response of the power supply. This will allow us to measure the 0dB frequency of the power supply, and the phase margin at that point.
If we have low phase margin, or even negative phase margin, our power supply will be underdamped and may oscillate. We should shoot for a phase margin at 0dB of better than 45°. And if our 0dB point crosses at a frequency that is too low, even though we have plenty of phase margin, our power supply may be too slow in responding to transient load changes. The goal is to adjust compensation components to achieve the desired phase and frequency response.
For my first attempt, I used some compensation values derived from the vendor spreadsheet and simulation files, modified the prototyping board, and checked it with the network analyzer. I used an Agilent 4395A Network Analyzer and Bode box to connect to the prototype board. Below is the Bode plot for full load (I also tested at other load conditions).
My main points of interest are the phase margin, at about 44.6°, and the 0dB crossing of about 25kHz. The phase margin is a bit low for a reliable design over temperature and component variation, and with such a low 0dB crossover, I'm concerned that my transient response will be poor. We can get an idea of transient performance by hitting it with a load step, measuring the voltage deflection on the output and checking for oscillations. In the capture below, the power supply is hit with a load step of 30 percent to 100 percent at a slew-rate of 0.5A/μs. The equipment used was:
- Tektronix MSO 4104 (1GHz, 5 GSa/s)
- Tektronix TCP0150 Current Probe (150A)
- Sorensen Programmable Load
As you can see, the load itself (magenta) overshoots a bit on the up slope, but we can still get an idea of how the power supply responds. The output (blue) deflects down by 30mV, which is out of my comfort zone. I will attempt to improve the phase-margin, and push the 0dB frequency up; two goals that are at odds with each other.
This took multiple iterations of moving poles and zeros with the compensation components to get it tuned just the way I wanted it, but basically I experimented with:
- Increasing the resistance in the RC of the compensation network to increase the 0dB crossing frequency, and
- Decreasing the capacitance of the compensation capacitor, and installing the feed-forward capacitor to achieve some phase-boost.
My final results are in the Bode plot below.
My 0dB crossing frequency increased from 25kHz to 49.5kHz, which should yield a faster response, and I was able to achieve a healthy 69.6° of phase margin. This should yield a much more responsive and stable power supply. As can be seen from the load-step capture below, the output is much more responsive and well damped during the load-step, only deflecting about 8mV.
The next step will be to take the lessons learned from the prototype board, and incorporate them into the final design (and then retest, of course).
How does this process compare to your experiences?