Advertisement

Blog

Active Load for Power Supply Testing

I was just reading a LinkedIn message regarding methods that could be used to load-test a power supply.

The original poster wanted a method to do some active load testing inexpensively. He wanted to avoid buying any expensive test equipment (active load box, network analyzer). This seemed like a common problem. I needed to do this on one of my basement lab projects and (since I have never won the lottery) I wanted to do my testing on the cheap. Mostly, I needed to know if the control loop in my supply had sufficient phase margin that it wasn't going to ring like the local neighborhood kids on one of their candy-selling marathons whenever the load conditions changed.

For the particular specs with my supply (5V at 2A), I knew that if I loaded it at around 10 percent of full load, then switched to full load, and then switched back, I could simply observe the output voltage to see what happened. I built the following circuit on a scrap of perf-board using a medium power NPN transistor. I used a TIP41B because I had some, but those are not recommended for new designs. Almost any decent NPN power transistor or N-channel power FET will work in this application.

The schematic of my active load device.

The schematic of my active load device.

I set the '555 up as a free-running square wave generator (well, within a few percent of being a 50 percent duty cycle square wave). Frequency is just below 100Hz. The actual frequency doesn't matter as much as the rapid rise and fall of the waveform. This is the way the circuit was used: I connected the 2.49Ω resistor (LOAD SINK) to my power supply being tested. I also had an additional 27Ω resistor connected to the power supply. When the power transistor is on, load current was about 2A; when off, around 200mA. I monitored the voltage at the LOAD SINK terminal with my scope. The SYNC terminal is used if I needed a separate sync input to the scope for more accurate triggering. More on that in a minute.

The circuitry kludged up for my active load device.

The circuitry kludged up for my active load device.

What I was looking for is excessive ringing in the output voltage as the high-current draw snapped on and off. If the voltage just overshot slightly and then quickly settled to a final value, then my loop compensation was probably just right. If the phase margin was insufficient, the ringing would last a while. That SYNC input comes in handy if the output is ringing excessively. It's tough to set a good, consistent trigger point on such a waveform. The SYNC output provides a clean, square waveform with no ringing.

I couldn't produce quantitative evaluations with this gadget (I could not specify precise values of phase margin), but it worked well enough to do a quick evaluation of power supply prototypes.

6 comments on “Active Load for Power Supply Testing

  1. Vincent Rheaume
    February 18, 2013

    Regarding the circuit of fig. 1: power NPN often have rather low gain (hfe of 15-75 for the TIP41, according to the datasheet), and your circuit assumes that it is in saturation, not in linear mode, so that the load current is mostly limited by the load resistor, not the BJT (5V/2R49 = 2A)… If I were to rebuild your circuit, I'd use a Darlington configuration and/or a lower base resistor! Or, as you actually mentioned, pretty much any power NMOS nowadays. Hope you don't mind me nitpicking, it's in my nature! 🙂 

    On the subject of active load testing, if other readers haven't read it already, I recommend reading Jim Williams' posthumously-published article on EDN, “Design a 100A active load to test power supplies” ( http://www.edn.com/design/analog/4368416/Design-a-100A-active-load-to-test-power-supplies ). It shows a more complex circuit for rather advanced needs, but in my opinion it still fits the “cheap” criterion (just a handful of amplifiers, transistors, diodes and resistors after all!). 

    That being said– Brad, thank you for another technical blog post, always interesting to read! 

  2. Brad Albing
    February 18, 2013

    Thanks – I do appreciate the comments and don't mind the nitpicking. You're right about beta being a concern with the NPN that I used. Again – just something I threw together. The transistor did saturate sifficiently well – but surely, if I built 100 of them, some pro'ly would not work at the current levels where I operated. A Darlington would be better in some ways, worse in others – kind of high saturation voltage, especially for my low-voltage supply testing. So if I did it again, I'd use a medium power FET with a low gate threshold voltage.

  3. jbike
    February 21, 2013

    If you want to take your approach one step further try this. Modify your current sink (the load) to change sinusoidally. This really isn't too difficult since the transistor has such a high output impedance in the linear region, where a large change of collector/drain voltage is necessary to instantiate even a small change in drain/collector current.  Mosfets are better, but your transistor should work just fine. We can use an opamp, a mosfet,  and a function generator to generate the sinusoidal load. Now look at both the load current (measured across a small emitter resistor which is needed anyways), and the power supply voltage on your scope. You will see a phase shift between each waveform. The time that the waveforms skew out of the period of the sinusoidal stimulus indicates your phase. For example: 3us/(20us period) = x/360 => 56 degree shift. Remember, this phase is on top of 180, since feedback is returned to the inverting input in any amplifier, including your power supply. If you hit 360 degrees when you still have gain, the supply will oscillate.  Now you can increase the frequency to zero in on the phase margin (where the gain is 1). The truly beautiful advantage of this technique is that it works just as well when the supply is actually oscillating. In this case, just use the oscillation as the stimulus. No circuits are required, just the load! Network analyzers are useless or at least very difficult to interpret when the circuit is oscillating. In any case, the sinusoidal stimulus allows you to use each of your scope probes to walk through the circuit and measure the phase shift across circuit blocks inside the supply, to zero in on devices with more than their alotted phase shift.

  4. ZekeR0
    February 21, 2013

    Hi Brad,

    I found myself evaluating the dynamic response of a power converter with a super-fast 1MHz control loop. To be meaningful, the load step must be much faster than the control bandwidth — ideally about 10x faster — and all the active loads in the lab wouldn't step any faster than 1µs. So I jerry rigged a fast load box: four 2N2222's in parallel, each with a 1kΩ base resistor connected to the function generator. The rise/fall times were around 100 to 200ns, which was good enough for me. As for specifying the amount of current, I just dialed the high and low voltage levels on the function generator while observing the probed current on the oscilloscope. This made the active load good for 10mA, 1.5A, and anywhere in between. Not the most precise thing in the world, and the current would change somewhat as the transistors warmed up, but I worked my way around it. I used my load box to test supplies from 1V to 3.3V without complaint, and keeping thermals in mind it could be used for much more. For those who can't afford current probes, the current could easily be measured by placing a 100mΩ resistor in the active load's return path and measuring the voltage. As long as ground loops are considered, the current measurement will be good to a few percent. Not bad for something so simple.

    “I couldn't produce quantitative evaluations with this gadget (I could not specify precise values of phase margin), but it worked well enough to do a quick evaluation of power supply prototypes.”

    Using a load step box gives a time-domain evaluation of the circuit's output impedance. With closed-loop control, every parameter in the loop will have (1+T) in its denominator, where T is the loop gain. If Zo is taken as the open-loop output impedance, the closed-loop output impedance will be Zo/(1+T). Instability occurs when the denominator (1+T) goes to zero, i.e., T goes to -1 and the parameters go to infinity. In the frequency domain, this will be observed as resonant peaking; a high-Q resonance indicates instability. Evaluating the ringing after a step disturbance is a legitimate way of figuring out the Q-factor of the resonance, and therefore, of determining a control loop's stability. There's a reasonably straightforward relationship between the stability margin, Q-factor and the step response, and several textbooks have nice graphics depicting it. Just because you can't “read” the phase margin off the oscilloscope doesn't make it any less credible than a Bode or Nyquist plot obtained from a vector network analyzer.

    It can be argued that the step response is in fact better  than the frequency response for evaluating stability. Set aside the fact that phase margin doesn't by itself give a good measure of stability (remember that phase margin can be good at the same time that gain margin is bad, and vice versa). Set aside the fact that breaking the loop to measure loop gain can disturb its behavior and requires some point in the loop to have certain impedance characteristics (some will advocate two-port output impedance measurements for this reason). Set aside the possibility of an unstable inner loop inside the IC, which a plot of the outer loop's gain will not show. Set aside the fact that appropriate frequency-domain tools are often prohibitively expensive. All this talk of the frequency domain is subject to a very important simplification: Linearity. Small-signal. As soon as you consider a circuit's large-signal, slew-limited, saturating, nonlinear behavior, you move away from the nice analytical tools that linearity gave you, including Bode plots. The truth is, it's possible for a circuit to be small-signal stable but large-signal unstable. Or have no small-signal overshoot, but have immense large-signal overshoot. Or be small-signal stable at one operating point and small-signal stable at another operating point, but small-signal unstable while slewing between the two. Good luck finding this out with a Bode plot.

    In my opinion, doing a series of load steps with various amplitudes and at various operating points is your best bet for a true, final evaluation of power supply stability. Be sure to exercise it thoughout its voltage and current ranges, and through transitions in operating mode such as PFM and PWM. The frequency domain has its advantages: dynamic range (due to logarithms), and the ability to identify poles and zeros. As a design tool, knowing where your poles and zeros are sure is handy. But if you're not trying to get super fancy with your power supplies, and as long as you carefully and thoroughly test them, doing fancy frequency-domain measurements usually doesn't tell you much about the circuit's stability that you couldn't already tell from the step response. If you want to find out if something's tough and robust, kick it.

    -Zeke

  5. Brad Albing
    February 21, 2013

    Hi Zeke – that is an excellent summary of the advantages and pitfalls of the methods we commonly use to test supplies. Also very good summary regarding inner and outer loops and operating points. Thanks so much for your good words.

  6. Brad Albing
    February 26, 2013

    That's a pretty clever idea for measuring phase-shift – I never thought of that. I will try that!

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.