A power supply that I was examining used an opto-coupler in its feedback path for control of its output voltage. This is a commonplace and well established design approach, but the feedback loop’s unconditional loop stability was not altogether a certainty.

I decided to measure the opto-coupler’s transfer function in two ways as seen below. I knew that the device’s datasheet presented this characteristic, but I wanted a first hand look for myself (Figure 1).

Figure 1 Opto-coupler transfer function test set-ups.

The 1.33k and 150 ohm resistors just happened to be conveniently on hand so I used them. A 1k and a 100 ohm pair would have sufficed just as well, but I just didn’t feel like rummaging at that moment.

The two test results did seem to confirm the datasheet presentation (Figure 2).

Figure 2 Opto-coupler transfer function test results.

The upper trace of Figure 2 goes with the upper sketch of Figure 1, the lower trace with the lower sketch. It was reassuring to note the symmetry of the two test results.  However, upon some mental reflection, I realized that there was a loop gain booby-trap of sorts of which to be aware.

Using the lower trace of Figure 2 and measuring the transfer function slopes, we see the following (Figure 3):

Figure 3 Close-up examination of the transfer function.

The “linear” nature of the opto-coupler is not to be taken too literally. The device’s transfer function doesn’t switch or anything like that, but the first derivative of the output versus the input, which is to say the slope of the output versus the input, varies versus where you place the device’s operating Q-point. The variability in the device I examined was nearly 19 dB.

That much gain variation could have a rompin’-stompin’ effect on a feedback loop’s overall transfer function, possibly pushing a marginally stable feedback loop into conditional instability.

The cautionary note of all this is to be sure to check the true slope of whichever opto-coupler you’ve chosen for yourself and where you choose to set its Q-point.

John Dunn is an electronics consultant, and a graduate of The Polytechnic Institute of Brooklyn (BSEE) and of New York University (MSEE).

A better approach would be to use a truly linear opto coupler like the IL300 which operates with two matched photodiodes and one LED. One photodiode is used as an output on the isolated side while the second photodiode is used to sense the current. This essentially operates as a servo and keeps the gain nearly constant, within 5% or so.

That is just passing the same stability problem off to the local servo loop. But it is probably a lot easier to manage there, rather than the entire power supply loop.

Generally, a high frequency pole is required on each amp in the servo to compensate for the zero created by the capacitance of the photdiode/transistor. These pole are usually >250kHz for good stability I’ve never had any issues with using this type of servo-optocoupler in any power supply if you follow this advice.

This type of opto circuit is also nearly immune to the effects of aging or ionizing radiation on the light output efficiency of the LED. as long as the dynamic range of the drive current is sufficient.

There is a fascinating comment on LinkedIn at the Mixed Signal Electronics/PCB Design group from Thomas Mathews, MSEE, PE 2nd degree connection 2nd Entrepreneur .

“We had some automotive customers who refused to use opto-isolation. They said that, over years, the CTR (current transfer ratio) degrades. I think this effect is much exaggerated but, it resulted in many customers switching to ADI’s magnetic isolators and Texas Instruments’ capacitive isolators.”

Nice! Sounds like the Automotive boys who objected to Optos need to read this and verify the Q point of the failures!. I designed some stuff for Nissan, who has a big standard practices and design rule document. It was an obvious response to all the major issues they had over the past 40 years and the corresponding restrictions to combat them. All issue were due to poor design and manufacturing choices, and of course most of the concerns and rules were antiquated. I laughed at the numerous sections about how to properly implement circuits using Germanium semiconductors! I also wonder if the current generation of engineers could devise and make the simple test circuits you show, since the Optos have no debug port or USB connection.

Unfortunately the real booby-trap maybe not covering how much both of those gain section are affected by temperature, as is the transition point; not to mention that overall the very wide spread from device to device (even of the same part #). A cheap opto used in power supply feedback will almost always be used with a voltage sensing reference chip, like a TL431, within a localized feedback loop, to not only set a stable opto operating point, but to greatly linearize the overall transfer function, owing to a large amount of local negative feedback. The gain and slope examples for the opto alone, do not equate directly into a typical power supplies feedback loop or stability calculations; but rather only the very small residual non-linearity does, after the very large negative feedback of the TL431 has made the opto behave more like a near perfect device.

A voltage control chip like the TL431 will serve well to set an opto-isolated feedback loop’s q-point predictably and accurately, but any non-linearity of the current transfer ratio (CTR) in the opto-coupler will not be overcome. The slope of the CTR will remain as a scale factor multiplier in the feedback loop’s gain equation.

If the loop’s pole-zero constellation is well chosen, this may not matter in the same way that variability of an op-amp’s open loop gain value might not matter either. That op-amp might have an open-loop gain of 20,000 or 100,000 for example and still serve the feedback loop adequately.

Getting back to the opto-coupler however, if the pole-zero constellation is not quite where it should be, or if there is some critical reason for such a particular constellation, feedback loop stability could get put in some jeopardy as a CTR slope varies.

You must Sign in or Register to post a comment.