another goofy boost converter architecture

Use filters.

I've found out the hard way that transmission lines on your schematic force Tmaxstep = Tdelay

With PSpice you can do .PRINT (not available in LTspice) with an appropriate increment, then run the results thru Excel. ...Jim Thompson

If you don't mind some criticism... your time constants weren't long compared to the signal shape.

From your sampling points I don't know how that data had any meaning. ...Jim Thompson

With a switcher, that wouldn't be a huge problem because you want the transmission delay to be an integral number of cycles, and obviously the time step has to be shorter than that!

Cheers

Phil Hobbs

On Tue, 09 Jul 2013 12:02:21 -0700, Jim Thompson wrote:

It's useful to me. I have used tiny output caps in most of the sims, to save time, but if I use bigger caps after things are tuned, the efficiency calc lags become insignificant. The actual charging times will be fractions of a second, although I may never sim that... I don't have enough hard drive for the .RAW file.

The efficiency really matters to me in the "recharge" time, when the caps charge from about 38 to 48. My graph is good for estimating that.

The filters are there to smooth the graphs to make them usable. Too fast and the graphs are fuzzy or wild; too slow and the efficiency calc lags realtime. The ones I used were OK for me, but I did a later version with 2nd order current shunts and no post-calc filtering, which is a little better, I guess. See below

One could pick off the raw currents and add serious Bessel or transitional filters off to the side. A Bessel impulse response is a nice windowing function, with a little time lag of course.

Version 4 SHEET 1 3540 2452 WIRE 1216 1056 1168 1056 WIRE 1264 1056 1216 1056 WIRE 1376 1056 1344 1056 WIRE 1520 1056 1456 1056 WIRE 1808 1056 1520 1056 WIRE 1920 1056 1808 1056 WIRE 2112 1056 2000 1056 WIRE 2240 1056 2192 1056 WIRE 2304 1056 2240 1056 WIRE 2432 1056 2384 1056 WIRE 2560 1056 2496 1056 WIRE 2624 1056 2560 1056 WIRE 2768 1056 2704 1056 WIRE 2912 1056 2848 1056 WIRE 2960 1056 2912 1056 WIRE 3008 1056 2960 1056 WIRE 2112 1072 2112 1056 WIRE 2192 1072 2192 1056 WIRE 1168 1152 1168 1056 WIRE 1312 1152 1168 1152 WIRE 1520 1152 1520 1056 WIRE 1520 1152 1376 1152 WIRE 2560 1152 2560 1056 WIRE 2720 1152 2560 1152 WIRE 2912 1152 2912 1056 WIRE 2912 1152 2784 1152 WIRE 2912 1168 2912 1152 WIRE 1968 1184 1936 1184 WIRE 2112 1184 2112 1152 WIRE 2112 1184 2048 1184 WIRE 1808 1232 1808 1056 WIRE 1936 1232 1936 1184 WIRE 1168 1248 1168 1152 WIRE 2912 1280 2912 1232 WIRE 1520 1312 1520 1152 WIRE 1728 1312 1520 1312 WIRE 2064 1312 2016 1312 WIRE 2112 1312 2112 1184 WIRE 2112 1312 2064 1312 WIRE 2240 1312 2112 1312 WIRE 2352 1312 2304 1312 WIRE 2352 1360 2352 1312 WIRE 1168 1392 1168 1328 WIRE 2192 1392 2192 1152 WIRE 2192 1392 2016 1392 WIRE 1296 1472 1168 1472 WIRE 1504 1472 1376 1472 WIRE 1584 1472 1504 1472 WIRE 1728 1472 1584 1472 WIRE 2064 1472 2016 1472 WIRE 2192 1472 2144 1472 WIRE 1168 1504 1168 1472 WIRE 1504 1536 1504 1472 WIRE 1168 1632 1168 1584 WIRE 1504 1632 1504 1600 WIRE 1808 1648 1808 1552 WIRE 1936 1648 1936 1552 WIRE 2192 1648 2192 1472 WIRE 2800 1808 2768 1808 WIRE 2832 1808 2800 1808 WIRE 2768 1856 2768 1808 WIRE 1296 1872 1168 1872 WIRE 1504 1872 1376 1872 WIRE 1568 1872 1504 1872 WIRE 1616 1872 1568 1872 WIRE 1856 1872 1760 1872 WIRE 1936 1872 1856 1872 WIRE 2016 1872 1936 1872 WIRE 2144 1872 2096 1872 WIRE 2176 1872 2144 1872 WIRE 1168 1888 1168 1872 WIRE 1504 1920 1504 1872 WIRE 1936 1920 1936 1872 WIRE 1168 2016 1168 1968 WIRE 1504 2016 1504 1984 WIRE 2768 2016 2768 1936 WIRE 1936 2032 1936 2000 FLAG 2912 1280 0 FLAG 2192 1648 0 FLAG 1936 1648 0 FLAG 1808 1648 0 FLAG 1168 1392 0 FLAG 1216 1056 9V FLAG 2960 1056 VP FLAG 2768 2016 0 FLAG 2800 1808 EFF FLAG 2064 1312 SW FLAG 2240 1056 DIO FLAG 2352 1360 0 FLAG 2144 1872 VP FLAG 1584 1472 RUN FLAG 1504 1632 0 FLAG 1936 2032 0 FLAG 1568 1872 DONE FLAG 1168 1632 0 FLAG 1856 1872 VD FLAG 1504 2016 0 FLAG 1168 2016 0 SYMBOL ind2 2096 1056 R0 WINDOW 0 -10 38 Right 2 WINDOW 3 -12 66 Right 2 SYMATTR InstName L1 SYMATTR Value 2µ SYMBOL ind2 2208 1168 R180 WINDOW 0 -16 70 Right 2 WINDOW 3 -12 39 Right 2 SYMATTR InstName L2 SYMATTR Value 18µ SYMBOL RES 2064 1168 R90 WINDOW 0 -10 106 VBottom 2 WINDOW 3 -38 56 VTop 2 SYMATTR InstName R1 SYMATTR Value 49.9K SYMBOL cap 2896 1168 R0 WINDOW 0 52 27 Left 2 WINDOW 3 50 64 Left 2 SYMATTR InstName C1 SYMATTR Value 10µ SYMATTR SpiceLine Rpar=.1Meg SYMBOL schottky 2432 1072 R270 WINDOW 0 -38 43 VTop 2 WINDOW 3 -49 46 VBottom 2 SYMATTR InstName D1 SYMATTR Value BAT46WJ SYMATTR Description Diode SYMATTR Type diode SYMBOL res 2048 1488 R270 WINDOW 0 -44 35 VTop 2 WINDOW 3 -17 75 VBottom 2 SYMATTR InstName R2 SYMATTR Value 2K SYMBOL VOLTAGE 1168 1232 R0 WINDOW 0 58 37 Left 2 WINDOW 3 59 73 Left 2 SYMATTR InstName V1 SYMATTR Value 9 SYMBOL res 1360 1072 R270 WINDOW 0 -37 34 VTop 2 WINDOW 3 -9 80 VBottom 2 SYMATTR InstName R5 SYMATTR Value 1m SYMBOL cap 1376 1136 R90 WINDOW 0 76 63 VBottom 2 WINDOW 3 49 8 VTop 2 SYMATTR InstName C3 SYMATTR Value 25m SYMBOL PowerProducts\\LT3420-1 1872 1392 R0 SYMATTR InstName U2 SYMBOL bv 2768 1840 R0 WINDOW 3 -558 176 Left 2 WINDOW 0 -92 115 Left 2 SYMATTR Value V= LIMIT (0, 100 * I(R7) * V(VP) / ( I(R5) * 9 ), 100) SYMATTR InstName B1 SYMBOL res 2608 1072 R270 WINDOW 0 -41 38 VTop 2 WINDOW 3 -14 85 VBottom 2 SYMATTR InstName R7 SYMATTR Value 1m SYMBOL cap 2784 1136 R90 WINDOW 0 74 68 VBottom 2 WINDOW 3 47 5 VTop 2 SYMATTR InstName C5 SYMATTR Value 25m SYMBOL ind 1248 1072 R270 WINDOW 0 -35 27 VTop 2 WINDOW 3 -8 83 VBottom 2 SYMATTR InstName L4 SYMATTR Value 25n SYMBOL ind 2752 1072 R270 WINDOW 0 -33 25 VTop 2 WINDOW 3 -3 76 VBottom 2 SYMATTR InstName L5 SYMATTR Value 25n SYMBOL zener 2304 1296 R90 WINDOW 0 72 33 VBottom 2 WINDOW 3 80 32 VTop 2 SYMATTR InstName D2 SYMATTR Value ZNR48 SYMATTR Description Diode SYMATTR Type diode SYMBOL res 2288 1072 R270 WINDOW 0 -33 54 VTop 2 WINDOW 3 -44 56 VBottom 2 SYMATTR InstName R6 SYMATTR Value 0.480 SYMBOL res 1904 1072 R270 WINDOW 0 -33 33 VTop 2 WINDOW 3 -6 87 VBottom 2 SYMATTR InstName R8 SYMATTR Value 0.24 SYMBOL res 2112 1856 R90 WINDOW 0 -48 63 VBottom 2 WINDOW 3 -37 57 VTop 2 SYMATTR InstName R11 SYMATTR Value 46.75K SYMBOL res 1392 1456 R90 WINDOW 0 -54 53 VBottom 2 WINDOW 3 -44 53 VTop 2 SYMATTR InstName R13 SYMATTR Value 1K SYMBOL cap 1488 1536 R0 WINDOW 0 67 15 Left 2 WINDOW 3 65 49 Left 2 SYMATTR InstName C2 SYMATTR Value 1n SYMBOL res 1920 1904 R0 WINDOW 0 66 38 Left 2 WINDOW 3 55 76 Left 2 SYMATTR InstName R9 SYMATTR Value 1.25K SYMBOL bv 1168 1488 R0 WINDOW 0 33 115 Left 2 WINDOW 3 20 193 Left 2 SYMATTR InstName B2 SYMATTR Value V = LIMIT( 0, 5, 5000 * (1.25 - V(VD) ) ) SYMBOL res 1392 1856 R90 WINDOW 0 -15 75 VBottom 2 WINDOW 3 -39 23 VTop 2 SYMATTR InstName R3 SYMATTR Value 1K SYMBOL cap 1488 1920 R0 WINDOW 0 67 38 Left 2 WINDOW 3 65 75 Left 2 SYMATTR InstName C4 SYMATTR Value 1n SYMBOL bv 1168 1872 R0 WINDOW 0 35 122 Left 2 WINDOW 3 23 -95 Left 2 SYMATTR InstName B3 SYMATTR Value V = LIMIT( 1, 10, 5000 * ( V(VD) - 1.20 ) ) TEXT 2576 1432 Left 2 !.tran 0 10m 0 20n TEXT 2248 1240 Left 2 !K1 L1 L2 0.95 TEXT 2144 1192 Left 2 ;1:3 TEXT 2224 1200 Left 2 ;LPR4012-202DML TEXT 1224 1728 Left 2 ;COMPARATORS: USE LM393 TEXT 2576 1320 Left 2 ;LT3420-1 WITH HELP TEXT 2592 1376 Left 2 ;JL July 9, 2013 TEXT 1248 1544 Left 2 ;RUN COMPARATOR TEXT 1232 1928 Left 2 ;DONE COMPARATOR TEXT 1712 1728 Left 2 !.model ZNR48 D(IS=1.54e-17 N=1 XTI=1 Rs=.1 Cjo=10p TT=10n BV=48 IBV=25m NBV=35 Vpk=48 mfg=OnSemi type=Zener)

TX lines store a lot of information. I wonder where LT Spice stashes it.

No.

The parts count would be exactly the same. The dynamics would be simpler.

You're getting whiney again. Say something on-topic.

My original post uses filters to display realtime efficiency. Jim used integrators to get total energy efficiency since startup. Then it was suggested to take a delta-t finite difference of the integrator output to get back to realtime efficiency.

An integrator is a 1st order lowpass. A filter can be higher order, which is a better compromise between plot noise and time response.

A finite difference is a differentiator, which, following an integrator, retrieves a lot of the noise that the integrator was supposed to remove.

This is a switching regulator, so some sort of filtering or averaging is needed. The input power peak and the output power peak don't happen at the same time, so the divide won't work.

Unless you use filters.

A finite difference of an indefinite integral is a definite integral, which is a constant times the moving average. Taking the finite difference of a lowpass filter gives you something a bit smoother in general.

Cheers

Phil Hobbs

But the energy change in the capacitor would be...

Thus my puzzlement as to what he was actually measuring.

I did plot _energy_ out versus _energy_ in, versus time. There's probably some way to process that into some "spot" equivalent, though I doubt it's usefulness.

And Larkin did indicate a need for computing the 38V back up to 48V scenario. The Watt-seconds required is available right off my graph. ...Jim Thompson

V*I is instantaneous power, any way you slice it.

d/dt(CV**2/2) = V*(C*dV/dt) = V*I.

The only issue is how to average over some small integer number of cycles, when the cycles aren't actually periodic and the conditions aren't necessarily the same before and after.

You just have to account for other stored energy, at least down to the accuracy you care about. Personally I wouldn't be that worried about a bypass cap, because it's guaranteed to average to zero. Trying to make the perfect SPICE model of a built-up circuit with nonlinear magnetics, thermal transients, device to device variations, yada yada, is pretty much a waste of time anyway, AFAICT.

The difference between your method and John's is a lot like a DVM versus a scope. Sometimes you want 0.1% accuracy with a 5-Hz update rate, sometimes you'd much rather have 3% accuracy in real time, so you can see what's going on.

A Gaussian or Bessel lowpass filter is a good choice for that, because the smooth monotonic tails of the impulse response smear out that cycle-to-cycle jitter. It's hard to eyeball small changes in the slope of the integral curve.

Taking a finite difference of an indefinite integral is another reasonable choice, but it will be a bit noisier in this case because the impulse response has cliffs on the ends, so with jagged waveforms it matters quite a bit whether it's a bit wider or a bit narrower than a cycle.

Even an RC is plenty good enough for eyeballing while the sim runs, which is John's original point. Postprocessing will give you the right answer DVM-fashion, but you don't want to have to wait till the run is over to see if some small change was an improvement or not. Once you're happy with the results, you can do the i-dotting and t-crossing to get a final simulated answer.

Cheers

Phil Hobbs

The integral+fd is just a rectangular-window sliding average, which can get nasty in a switcher. I don't know how to do that in LT Spice as a plottable waveform. Well, it can be done with delay lines as you have noted, but those delay lines have to store mountains of data.

Offline post-processing (of gigabyte-sized RAW files) and graphing isn't very appealing to me either. I want my sims to feel like using an oscilloscope on a real circuit.

Filtering is fast and simple, and is equivalent to a sliding average with a better windowing function. I wonder why some people don't like it.

IIR filters are great, even for digital filters. I've annoyed some number of digital filtering experts by designing simple models of dual-integrator state-variable filters while they were struggling with boxcar or butterfly filters with insane gains and coefficients.

I've told you a zillion times: Pout/Pin, just as you might measure eficiency of any reasonably filtered dc/dc converter.

My filtered input shunt (especially the RLC version) is equivalent to a pi filter, which I will certainly have on the real boost converter, without the startup surge. The only difference is where the cold side of the cap is connected.