EEVblog #957 - How To Measure DC-DC Converter Efficiency

[EEVblog]

Dave shows you how to setup, measure and generate an Efficiency vs Load Current characteristic curve for a DC-DC converter, just like you see in the datasheets.
Several traps for young players are discussed, along with plotting Logarithmic axis charts in a spreadsheet.
Electronic Load Kit: http://amzn.to/2i4cUdF
BM235 Multimeter: http://amzn.to/2hJp78T
Rigol PSU: http://amzn.to/2gZWkhw

[hans]

Yep, measuring stuff always has lots of traps buried in there

A thing I wanted to underline in measurements: take the example when it is sinking 80mA at 6:26.
I will be ignorant for now and just take the voltage, current & power measurements from the PSU and load.

In: 5.00*0.234 = 1.17W
Out: 11.80*0.080 = 0.944W
So that's 80.6%, nice.

As shown in the video the BK precision doesn't have great resolution. 80mA could have been 79.5 or 80.5, and just to show:
11.80 * 0.0795 = 0.938W -> 80.2%
11.80 * 0.0805 = 0.950W -> 81.2%

Oops, that's +/- 0.5% right there.

In fact; this is not entirely correct neither. The BK only gives us 2 significant digits for current, and so our measurement can only be 2 significant digits.
So with that measurement we can only conclude it's 81%, but not better. Luckily the BK also only shows 0.94W, because all other digits behind that are bogus. 0.94W is not the same as  0.940W or 0.94000W

This is a little trap and in practice I think often ignored, aside probably from the voltnuts who already know this. Of course you aren't also going to round off numbers and look at some bouncy graphs, but it's always nice to know what degree of precision measurements were taken.

But the introduction of the SMU outlines that the BK Precision electronic load isn't a perfect fit for this measurement. That load really shines if you have a few amps to measure because then you do get 4 significant digits.

[alxpo]

The idea to plot power losses and make any conclusions with logarithmic X axis and linear Y axis was not very good. The plot of power loss looks like exponent because of logarithmic X axis only. I guess in log-log coordinates the Power loss vs. Output current plot will be more or less straight.

[Brumby]

I'm quite happy with the axes Dave used.  They are quite appropriate in showing the performance of the unit.

Over the specified operating range, there is a gentle rise in the power dissipation - and when it goes over spec, the power changes are quite dramatic.  Shows me what to expect when pushed too far.

[Fungus]

I'm quite happy with the axes Dave used.  They are quite appropriate in showing the performance of the unit.

Yep. This is about looking for minimums and maximums, not the shape of the curve.

If it wasn't logarithmic a graph from microamps to amps would be about ten miles wide.

[GreggD]

I am about 99% sure this video is to get people ready to understand the coming video of probes up a batteries butt while it is wearing a red jacket.

[ggchab]

I like Dave's tutorials and this one is no exception.
Any technical reason why we never see the SPD3303X power supply in experiments ? 

[TheAmmoniacal]

I like Dave's tutorials and this one is no exception.
Any technical reason why we never see the SPD3303X power supply in experiments ? 

The Rigol is simply better, so why use the Siglent? Curious what happened to it though.

[mikerj]

That's a heck of a voltage drop on the input wires considering they are so short.  Special Chinese copper plated steel wire?

[Kleinstein]

The rather large resistance of the wires at the input side could cause additional errors. The current drawn is usually not pure DC, but often with rather high ripple. The input voltage is not constant either and most of the current is drawn, when the input voltage is below average. So this measures the input power to high. So for a good test use low resistance wires and maybe an extra capacitor.

One could see that, by having an extra capacitor right at the input of the DC/DC converter - this should reduce the current needed.

[twice11]

That's a heck of a voltage drop on the input wires considering they are so short.  Special Chinese copper plated steel wire?

I guess the breadboard contact resistance might be significant, too. That's why Dave said that he would solder the wires if it were a professional characterization of the power brick. Especially if he did the measurement for the product datasheet: The loss in the contact resistance between the breadboard and the power brick is now accounted to the power brick, and it would be nearly eliminated on a soldered connection, so measured efficiency would be better. You definitely want to advertise the highest possible efficiency.

[twice11]

Dave, you keep saying (and writing) "9V power brick", but the output voltage shown is 12V, as is the label on the brick...

[EEVblog]

Any technical reason why we never see the SPD3303X power supply in experiments ? 

Yes, I don't have one. They took it back.

[EEVblog]

Dave, you keep saying (and writing) "9V power brick", but the output voltage shown is 12V, as is the label on the brick...

I know, I was going to test the 9V brick I had but used the 12V one. My mind was just confused.

[Dave]

I guess in log-log coordinates the Power loss vs. Output current plot will be more or less straight.

Not even close.
You can clearly see that the curve doesn't approach zero on the low end, so how could it possibly be a straight line in the log-log scale?

[HKJ]

As Dave says the manual method works fine for a single curve.
For more curves and/or power supplies automation is definitely the way to go, I would not get anything done without automation. It requires equipment that support computer connected and some software that can read and control all the equipment. The last part is very limiting.

[Fungus]

I found the little yellow breadboard quite an appealing addition. It reminds me of the ones I got from an Indiegogo campaign for $9 Arduino clones. I think they were a perk if the campaign went well. I keep wondering if that is where Dave got them from too.

http://www.ebay.com/sch/i.html?_nkw=mini+yellow+breadboard

I find myself using those small ones more and more these days. I usually have half a dozen things on the go and trying to organize huge breadboards is a pain. With those I can just join a few together to whatever size I need.

[jmaja]

How well do multimeters work in this case? The current can have short peaks etc. Especially when the chip has some system to lower quiescent current (e.g. stops switching for a while) and there is only a rather small ceramic capacitor. Are all the multimeters able to show the accurate average current needed for calculating power?

[Kleinstein]

The usual DMMs are quite good in averaging. They usually do than on a multiple of 16.6 / 20ms, often 100 ms and multiples (so they work well with 50 and 60 Hz).  So only pulsed operation at less than about 10 Hz might be a slight problem.

The bigger problem it that average power is not the same as average current times average voltage, because voltage ripple is highly correlated with current ripple.

[rch]

Shouldn't we be measuring RMS current?   I know expensive meters do RMS voltage, but do they automatically do RMS current on the current range?

[twice11]

Shouldn't we be measuring RMS current?   I know expensive meters do RMS voltage, but do they automatically do RMS current on the current range?

A typical multimeter first does current-to-voltage conversion (i.e. a shunt resistor), followed by voltage measurement (including AC-to-DC conversion) in the current range. So there is no reason to assume they use a different AC-to-DC conversion for current measurement than they use for actual voltage measurement.

[HKJ]

Shouldn't we be measuring RMS current?   I know expensive meters do RMS voltage, but do they automatically do RMS current on the current range?

On AC they do RMS, not necessary on DC.
Average is good if you want to measure current consumption. RMS can give wrong results for that, because it measures the current in the wire and that can include some current running back (A switcher with not too much decoupling).

[jmaja]

I took a look at the schematics of my old Fluke 8060A True RMS multimeter. For AC range the voltage (from voltage divider of V input or voltage across the shunt resistor on current input) is fed through a capacitor to a RMS converter. The output of the RMS converter is then fed trough a RC filter (220k, 0.1u) to ADC input. On DC scale the voltage is fed directly to the same RC filter.

The RC filter will average the voltage and thus RMS value can't be measured on DC scale.