2
\$\begingroup\$

Note that this is not a question for circuit critique (and lots of things could be said about this circuit) but this circuit is only the base for the discussion of further points.

Here it is (an NE555 based PWM driving a buck): (.asc file can be found here)

I want to measure the efficiency of such converters and similar circuits, along with the contribution of single parts to the dissipated power (in order to pinpoint components that could be optimized to dissipate less).

Now to measure the efficiency, I should simply do \$ \eta=\frac{P_\mathrm{out}}{P_\mathrm{in}}\$. And for DC and simple loads, it is rather easy to calculate the dissipated power as \$P=V_\mathrm{drop}\cdot{}I\$. For a sine wave and ohmic load, its also not hard, you seem to either do \$P=rms(V_\mathrm{drop})\cdot{}rms(I)\$ or when you have a tool like LTspice you can make it plot \$V_\mathrm{drop}\cdot{}I\$ for sufficiently small timesteps and then just take the average of it.

But what do we do for more complex/unusual waveforms? Take for example LTspices plot of \$V_\mathrm{drop}\cdot{}I\$ for L1:

Although not surprising from the fact that the voltage drop reverses over the inductance, the plotted negative power dissipation confuses me. So this leads me to the first real question:

What does this negative Power mean, intuitively, and how does it relate to the power dissipated at heat (i.e. how to use it to calculate that power)?

Now this circuit has a bit ripple at the load (diode+resistor together) but that shouldn't hurt calculation of \$P_\mathrm{out}\$. But now lets look at the plot of what I want to calculate \$P_\mathrm{in}\$ from:

This is not anywhere near a sine curve, so my question for this is:

Is taking the average of this (and similar) plot (assuming sufficiently small step sizes) still a good approximation of the total disspiated energy (heat)?

For completeness, here is the output of the .measure statements:

avwin: AVG(-v(in)*i(b1))=1.74212 FROM 0.002 TO 0.009
avx=0.007 FROM 0.002 TO 0.009
avwout: AVG(v(inductor_diode,diode_resistor)*i(d1)+v(diode_resistor)*i(r2))=1.27344 FROM 0.002 TO 0.009
avdloss: AVG( v(from_mosfet)*i(d4))=-0.0478017 FROM 0.002 TO 0.009
avlloss: AVG(v(from_mosfet,inductor_diode)*i(l1))=0.0946656 FROM 0.002 TO 0.009
avlloss_a: AVG(abs(v(from_mosfet,inductor_diode)*i(l1)))=1.14626 FROM 0.002 TO 0.009
avcloss: AVG(v(inductor_diode)*i(c1))=0.00312184 FROM 0.002 TO 0.009
aveff: (avwout/avwin)*100=73.0969
xloss: avwin - avdloss - avlloss - avcloss - avwout=0.418698

This would indicate an efficiency of ~73% which does seem reasonable, but looking at the individual losses I have no idea since all have plots with negative values, and as per above question, I am confused as to how to interpret them or use them to calculate heat dissipated.

\$\endgroup\$
7
  • \$\begingroup\$ Inductors and caps can both store and release energy; when they release it you have negative power with the passive sign convention that SPICEs (all of them) use. See related question about transistor with negative power: electronics.stackexchange.com/questions/149315/… (due to capacitance). \$\endgroup\$ Nov 21, 2015 at 0:30
  • \$\begingroup\$ And you'll probably be interested in linear.com/solutions/1100 if you haven't found it already. They give an example of measuring power in a converter (albeit for the purpose of PFC). \$\endgroup\$ Nov 21, 2015 at 0:42
  • \$\begingroup\$ Which LTspice pot model did you use in that schematic? Google returns a bit too many hits on this... everyone has its own favorite ltspice pot model it seems... it would take me a bit long to find out which one you have in there. Ah, never mind I found it: ltwiki.org/files/LTspiceIV/lib/sym/Pot&Sw/potentiometer.asy \$\endgroup\$ Nov 21, 2015 at 0:58
  • \$\begingroup\$ Also they've removed BC107's from their Standard.bjt and 1N34A from their Standard.dio since you wrote this. Also the micro-symbol (for the 280uH) got effed-up by pastebin or something... \$\endgroup\$ Nov 21, 2015 at 1:11
  • \$\begingroup\$ Also to calculate efficiency for SMPS there is a much simpler way in LTspice: their built-in efficiency report; see slide 54 in here. However it doesn't work for your circuit; complains that it cannot find steady state. I'm not sure what is that you're trying to do with that behavioral source, but I suspect it may be a part of the problem/reason why the steady state can't be determined needed for the cookbook efficiency report. Alas even after changing it to a constant 12V source still no go \$\endgroup\$ Nov 21, 2015 at 1:38

2 Answers 2

3
\$\begingroup\$

You are doing most of those measurements correctly, in particular the input and output power look ok, but you have sign errors for dissipation in certain components. You have to be careful which way LTspice considers the current by hovering the mouse over the component. For example, for the swiching diode D4 you have it the wrong way; LTspice measures the current going up, so the dissipation calculation should reverse the sign of this current or the sign of the voltage. That way you'd get a (correct) positive average dissipation on the diode instead of:

avdloss: AVG( v(from_mosfet)*i(d4))=-0.0477985 FROM 0.002 TO 0.009

For a full breakdown, you'd also need to look at the [switching] losses thorough the MOSFET, which you don't seem to do, and perhaps at the driver BJT pair as well.

Short-term power dissipation for transistor or diode can indeed be negative because these components have capacitances that do matter in switching applications... but if the average power over a long period is negative, you probably got the sign wrong. Actually one way to check the way power signs should be is to alt-click a component. This plots its power with the correct signs using passive sign convention [PSC].

enter image description here

For example, below is your D4 diode, alt-clicked; it automatically got a "-" sign because the way the current is oriented relative to the non-zero voltage potential. Although at the switching frequency power swings both ways (reverse recovery time!), on average over a longer period power through it is clearly positive (dissipative) with PSC.

enter image description here

Also you can make LTspice integrate and/or average right from the waveform viewer, but ctrl-cliking the waveform name. If you zoom on the waveform before ctrl-clicking, you only get it for the visible interval. This is for your diode:

enter image description here

Doing the same (in the same interval) for the MOSFET I get 263.58mW. For Q1 I get 1.0816mW, for Q2 1.0047mW, etc. I can even get it for the NE555 as 45.49mW. YMMV how accurate this last one is.

This is obviously somewhat tedious to do by hand for all components. Alas, LTspice doesn't actually let you use its built-in efficiency report function except when using LT's own smps controllers... I suspect this is mostly because the steady state determination, which is a prerequisite for that report, seems to need special knobs in the controller's spice model... LTspice can't even figure out that a simple circuit made of a voltage source in series with a current source has steady state! This after putting all the recommended (load) labels and so forth on it.

Also, I've used the plot-power-then-average-from-graph method to get the overall in and out power (again from 2.1ms to 10ms); I got avg in 1.7385W and avg out 1.275W this way, which is 73.33% efficiency, so confirming your MEAS results in that respect.

As a sanity check, I've compared the power losses measured as a difference between in and out vs individual components, and it does check out.

enter image description here

So yeah, you can do this efficiency report/breakdown for non-LT smps controllers in LTspice... but it takes a bit more work. Oh, and in case some passerby wonders at the big losses in L1:

enter image description here

(The C1 cap has ultra-low ESR.)

\$\endgroup\$
2
  • \$\begingroup\$ I tried to look at the LT1171 model (for which I know the "steady" parameter to tran works). Alas LT is anything but open with their spice models. Besides the pin reference, which is text, but doesn't contain anything obviously helpful in this respect, the model is all in a binary ".sub" format for which there's no documentation whatsoever. \$\endgroup\$ Nov 21, 2015 at 7:26
  • \$\begingroup\$ Yeah, even the help says so "Steady state detection is written into the SMPS macromodels. Typically they are written to look for zero error amp output current averaged over a clock cycle. The algorithm takes the error amp's output compliance range into consideration." Doesn't say how write a macromodel so that happens.... \$\endgroup\$ Nov 21, 2015 at 10:50
2
\$\begingroup\$

Your supply is a battery and whatever shape the current taken from the supply is, it is the average current that tells you (along with the supply voltage) what you circuit power drain is. So, feed the supply thru a small resistor and large capacitor and plot transient current draw before the resistor (it should be fairly smooth). Use the math that LTSpice probably permits to produce a new transient graph which is voltage x current.

This gives you average power into the circuit.

You'll probably also find that LTSpice permits the voltage across a component to be multiplied by the current thru the component and truly calculates power.

\$\endgroup\$
2
  • 1
    \$\begingroup\$ "You'll probably also find that LTSpice permits the voltage across a component to be multiplied by the current thru the component and truly calculates power." exactly that is what I am doing for the green graph. And as stated in my question, I am confused by that power, as to what it means that it goes negative, which makes me think that just taking the average over the whole graph might not be right. Also my voltage source is B source that I tried to make behave rather bad, not a batter, although filtering should work there too. Does your suggestion mean that using the avg(V*I) is wrong? \$\endgroup\$
    – PlasmaHH
    Dec 4, 2014 at 12:48
  • \$\begingroup\$ Average power is the right method. Cyclically the waveform for power can go negative when reactive components are in the mix. Average power is true power given off as heat. \$\endgroup\$
    – Andy aka
    Dec 4, 2014 at 13:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.