# Efficiency question of # of LEDs for 6 volt 4 cell Joule Thief Flashlight circuit

This is a high voltage Joule Thief circuit simulation I developed in LTSpice. I am sure that it qualifies as a true Joule Thief because, in the picture below, we are zoomed into the "dead" voltage area where light is still being produced, the area from 0.8-volts-per-cell (4x0.8=3.2v) and 0.5-volts-per-cell (4x0.5=2.0v), which is quite dead, according to usual battery expectations and individual cell voltage. Yet we're still putting out about 10mA RMS at each LED. I have tried to make a more efficient Joule Thief, and here is my question.

Higher voltage seems to make a more efficient Joule Thief, up to a point. You will see in the screen shot, efficiency calculations for powering strings of LEDs with different numbers of LEDs in them. What's really going on here, and how can we improve the efficiency even more?

FOR 3 LEDs, (D1-D3 and R1=15.000K) 613.12mW / 714.21mW = 85.85 % Efficiency

FOR 4 LEDs, (D1-D4 and R1=14.135K) (SHOWN) 613.98mW / 714.19mW = 85.97 % Efficiency

FOR 5 LEDs, (D1-D5 and R1=13.410K) 607.05mW / 714.26mW = 84.99 % Efficiency

For this 6-volt 4-Cell Joule Thief... Why does efficiency go up from 3 to 4 LEDs? Why does efficiency start going down at 5 LEDs, and how do we make the most efficient Joule Thief?

If it seems like I'm asking too many questions, I'm actually not sure what question to ask, because I don't know what I'm missing.

Also, (and I'll move this to a different question on this site) there seem to be really two concepts of efficiency here, an "instantaneous" efficiency, which changes per voltage and resistor settings, and an overal battery-life-efficiency or energy-usage-efficiency, which would put a figure on how efficient a store-bought resistor regulated flashlight would be compared to a Joule Thief, or even a constant current regulated flashlight. The closest I could come up with a way to test this kind of efficiency is to charge up some supercapacitors to the same voltage point and let the flashlight use the energy however it wanted, to see how long they last. So what are the industry terms used for these kinds of efficiency?

Here is a screen shot that describes the nature of my question:

and here is the LTSpice simulation file (*.asc) so that you can play with it:

Version 4
SHEET 1 1364 680
WIRE -160 -64 -272 -64
WIRE 64 -64 -160 -64
WIRE -160 -32 -160 -64
WIRE 64 -32 64 -64
WIRE -272 0 -272 -64
WIRE -240 0 -272 0
WIRE -224 0 -240 0
WIRE -272 64 -272 0
WIRE -160 64 -160 48
WIRE 272 64 256 64
WIRE 288 64 272 64
WIRE 64 80 64 48
WIRE 112 80 64 80
WIRE 224 80 176 80
WIRE 256 80 256 64
WIRE 256 80 224 80
WIRE 304 80 256 80
WIRE 400 80 368 80
WIRE 496 80 464 80
WIRE 592 80 560 80
WIRE 864 80 656 80
WIRE 64 112 64 80
WIRE -160 160 -160 144
WIRE 0 160 -160 160
WIRE 224 160 224 80
WIRE -272 240 -272 144
WIRE 64 240 64 208
WIRE 64 240 -272 240
WIRE 224 240 224 224
WIRE 224 240 64 240
WIRE 864 240 864 80
WIRE 864 240 224 240
WIRE -272 256 -272 240
FLAG -272 256 0
FLAG 272 64 D1
FLAG -240 0 V1
SYMBOL ind2 -176 -48 R0
WINDOW 0 44 73 Left 2
WINDOW 3 45 46 Left 2
SYMATTR InstName L2
SYMATTR Value 220µH
SYMATTR Type ind
SYMATTR SpiceLine Rser=0.001 Rpar=0 Cpar=0
SYMBOL ind2 80 64 R180
WINDOW 0 67 40 Left 2
WINDOW 3 38 66 Left 2
SYMATTR InstName L1
SYMATTR Value 220µH
SYMATTR Type ind
SYMATTR SpiceLine Rser=0.001 Rpar=0 Cpar=0
SYMBOL voltage -272 48 R0
WINDOW 3 -64 132 Left 2
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
WINDOW 0 34 58 Left 2
SYMATTR Value PULSE(0 6.2 0 5ms 10ms 5ms)
SYMATTR InstName V1
SYMBOL npn 0 112 R0
SYMATTR InstName Q1
SYMATTR Value 2SCR533P
SYMBOL res -176 160 M180
WINDOW 0 36 76 Left 2
WINDOW 3 36 40 Left 2
SYMATTR InstName R1
SYMATTR Value 14.135K
SYMBOL LED 304 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 77 40 VTop 2
SYMATTR InstName D1
SYMATTR Value NSPW500BS
SYMBOL schottky 112 64 M90
WINDOW 0 0 32 VBottom 2
WINDOW 3 32 32 VTop 2
SYMATTR InstName D99
SYMATTR Value 1N5817
SYMATTR Description Diode
SYMATTR Type diode
SYMBOL cap 208 160 R0
SYMATTR InstName C4
SYMATTR Value 22nF
SYMATTR SpiceLine V=50 Irms=0 Rser=0.008 Lser=0
SYMBOL LED 400 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 106 36 VTop 2
SYMATTR InstName D2
SYMATTR Value NSPW500BS
SYMBOL LED 496 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 77 40 VTop 2
SYMATTR InstName D3
SYMATTR Value NSPW500BS
SYMBOL LED 592 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 106 36 VTop 2
SYMATTR InstName D4
SYMATTR Value NSPW500BS
TEXT -120 -48 Left 2 !K1 L1 L2 1.00
TEXT -256 224 Left 2 !.tran 20ms startup


and here is the LTSpice "Plot" file (*.plt) to set up the graphing:

[Transient Analysis]
{
Npanes: 1
{
traces: 4 {524290,2,"V(D1)*I(D1)"} {524291,2,"V(V1)*I(V1)"} {268959748,0,"V(v1)"} {303038469,1,"I(D1)"}
X: ('m',0,0,0.002,0.02)
Y[0]: (' ',1,0,0.6,6.6)
Y[1]: ('m',0,-0.02,0.02,0.18)
Volts: (' ',0,0,1,0,0.6,6.6)
Amps: ('m',0,0,0,-0.02,0.02,0.18)
Units: "W" (' ',0,0,1,-2,0.5,3.5)
Log: 0 0 0
}
}


=============

# EDIT #1

To show that this is not just "noise" and is found elsewhere, here is a shot that (to me) shows this happens with a Buck converter chip (and, wow, shows 98 percent efficiencies are attainable) -- (the graph on the right). Even more graphs are in the datasheet, for different-sized inductors.

Furthermore, here is a picture of a 6 LED version, with the constant 6.2 volts I used in making my test figures. The figures for the 6-LED version are 594.96mW / 714.17mW = 83.31 % Efficiency. As I hope you can see, the efficiency starts to plummet the more LED's you add. Of course, at 6 LED's, we're having to pump up the voltage from 6.2 volts to 19.455 volts RMS, so I sort of expect it, but I don't really know what I'm fighting. At 10 LED's, (R1=12.575K) we fall to 573.36mW / 714.17mW = 80.28 % Efficiency.

Finally, here's an updated LTSpice file (*.asc) to go along with the above picture:

Version 4
SHEET 1 1364 680
WIRE -160 -64 -272 -64
WIRE 64 -64 -160 -64
WIRE -160 -32 -160 -64
WIRE 64 -32 64 -64
WIRE -272 0 -272 -64
WIRE -240 0 -272 0
WIRE -224 0 -240 0
WIRE -272 64 -272 0
WIRE -160 64 -160 48
WIRE 272 64 256 64
WIRE 288 64 272 64
WIRE 64 80 64 48
WIRE 112 80 64 80
WIRE 224 80 176 80
WIRE 256 80 256 64
WIRE 256 80 224 80
WIRE 304 80 256 80
WIRE 400 80 368 80
WIRE 496 80 464 80
WIRE 592 80 560 80
WIRE 688 80 656 80
WIRE 784 80 752 80
WIRE 928 80 848 80
WIRE 64 112 64 80
WIRE -160 160 -160 144
WIRE 0 160 -160 160
WIRE 224 160 224 80
WIRE -272 240 -272 144
WIRE 64 240 64 208
WIRE 64 240 -272 240
WIRE 224 240 224 224
WIRE 224 240 64 240
WIRE 928 240 928 80
WIRE 928 240 224 240
WIRE -272 256 -272 240
FLAG -272 256 0
FLAG 272 64 D1
FLAG -240 0 V1
SYMBOL ind2 -176 -48 R0
WINDOW 0 44 73 Left 2
WINDOW 3 45 46 Left 2
SYMATTR InstName L2
SYMATTR Value 220µH
SYMATTR Type ind
SYMATTR SpiceLine Rser=0.001 Rpar=0 Cpar=0
SYMBOL ind2 80 64 R180
WINDOW 0 67 40 Left 2
WINDOW 3 38 66 Left 2
SYMATTR InstName L1
SYMATTR Value 220µH
SYMATTR Type ind
SYMATTR SpiceLine Rser=0.001 Rpar=0 Cpar=0
SYMBOL voltage -272 48 R0
WINDOW 3 -7 54 Left 2
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
WINDOW 0 34 58 Left 2
SYMATTR Value 6.2
SYMATTR InstName V1
SYMBOL npn 0 112 R0
SYMATTR InstName Q1
SYMATTR Value 2SCR533P
SYMBOL res -176 160 M180
WINDOW 0 36 76 Left 2
WINDOW 3 36 40 Left 2
SYMATTR InstName R1
SYMATTR Value 13.336K
SYMBOL LED 304 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 77 40 VTop 2
SYMATTR InstName D1
SYMATTR Value NSPW500BS
SYMBOL schottky 112 64 M90
WINDOW 0 0 32 VBottom 2
WINDOW 3 32 32 VTop 2
SYMATTR InstName D99
SYMATTR Value 1N5817
SYMATTR Description Diode
SYMATTR Type diode
SYMBOL cap 208 160 R0
SYMATTR InstName C4
SYMATTR Value 22nF
SYMATTR SpiceLine V=50 Irms=0 Rser=0.008 Lser=0
SYMBOL LED 400 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 106 36 VTop 2
SYMATTR InstName D2
SYMATTR Value NSPW500BS
SYMBOL LED 496 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 77 40 VTop 2
SYMATTR InstName D3
SYMATTR Value NSPW500BS
SYMBOL LED 592 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 106 36 VTop 2
SYMATTR InstName D4
SYMATTR Value NSPW500BS
SYMBOL LED 688 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 77 40 VTop 2
SYMATTR InstName D5
SYMATTR Value NSPW500BS
SYMBOL LED 784 64 M90
WINDOW 0 -5 30 VBottom 2
WINDOW 3 106 36 VTop 2
SYMATTR InstName D6
SYMATTR Value NSPW500BS
TEXT -120 -48 Left 2 !K1 L1 L2 1.00
TEXT -256 224 Left 2 !.tran 0 4.2ms 0.2ms startup
TEXT -824 -40 Left 2 ;FOR 3 LED's,  (D1-D3 and R1=15.000K)\n613.12mW / 714.21mW = 85.85 % Efficiency\n \nFOR 4 LED's,  (D1-D4 and R1=14.135K)\n613.98mW / 714.19mW = 85.97 % Efficiency\n \nFOR 5 LED's,  (D1-D5 and R1=13.410K)\n607.05mW / 714.26mW = 84.99 % Efficiency\n \nFOR 6 LED's,  (D1-D6 and R1=13.336K)\n594.96mW / 714.17mW = 83.31 % Efficiency
TEXT -240 256 Left 2 ;For each case, R1 has been adjusted to draw in the same amount of power at V1 (714mW).\nEfficiency = V(D1)*I(D1)  /  V(V1)*I(V1)   electronics-stackexchange-efficient-6v-joule-thief-query-1
TEXT 104 -72 Left 2 ;L1-L2 is a hypothetical perfect coupled inductor\n(ESR=1mOhm, 100% coupling).\nD1=65.975mA RMS, 37.424mA AVG, 172mA PK for 4 LED's.\nD99 & C4 prevent negative currents in diode string.


Thanks ahead of time to all who can shed light on a truth that seems to be "increased voltage means increased efficiency" and more importantly, WHY this seems to be true, and also what "brick wall" it seems to hit.

• What were the input voltages for the 3, 4, and 5 LED efficiency runs? Why are you ramping the input voltage up and down rather than holding it steady? – Bruce Abbott Mar 21 '18 at 19:00
• You are using a pulsed input. Why? A Joule Thief doesn't. Also, I would like to see LTspice do the integration for power on both the source (DC as it is normally operated and not pulsed) and the LED stack-up. What I see instead, so far, isn't representative of reality and I'm not sure how anyone can address the "more efficient" question. – jonk Mar 21 '18 at 19:00
• I averaged the input and output power for 4 LEDs with V1 = 6V and efficiency worked out to 85.5%, so your figures look about right. Not sure that the tiny variations between LED numbers are significant though. – Bruce Abbott Mar 21 '18 at 19:05
• Even though my first LTSpice file contained a ramping voltage source, I added that just before I posted the file (sorry). I really did all of my testing with a straight 6.2 volts. Please see the edited version of my question, as I added a screen shot from a buck converter datasheet that shows the same effect, and another case (6-LED's), along with an updated LTSpice file, at the bottom. Thanks all. – MicroservicesOnDDD Mar 21 '18 at 20:30
• The more LEDs, the higher the output voltage, and the larger the reverse base emitter voltage which can break the transistor. And for my money, 85.85%, 85.97% and 85.-01% are the same number, at least for the large variation in the number of LEDs. – Neil_UK Mar 21 '18 at 22:06

a truth that seems to be "increased voltage means increased efficiency... and also what "brick wall" it seems to hit"

So higher voltage increases efficiency - until it doesn't?

What generally happens is that as output voltage increases the duty cycle also has to increase in order to 'transform' the power to a higher voltage at lower current. This causes increased losses in some components and reduced losses in others. Which ones (if any) 'win' depends on their individual characteristics, but in general efficiency tends to reduce as the voltage is boosted higher.

However the efficiency can also go down when the voltage boost is small, due to the increased output current required to deliver the same power. This appears to be what you are seeing - efficiency increasing slightly from 3 to 4 LEDs, but reducing above that.

If you go below 3 LEDs things start to get weird. With 2 LEDs the efficiency drops dramatically. With 1 LED you can get over 94% - but only by increasing R1 until oscillation stops completely, pushing DC current through the LED. With zero LEDs...

OK that's silly, but it shows that broad generalizations such as 'increased voltage increases efficiency until it hits a brick wall' are no substitute for a proper analysis. Between 3 and 5 LEDs your efficiency curve is practically flat, and hinting towards gradually reducing efficiency at higher output voltage. So in this particular case your generalization is not just useless, it's wrong.

Also note that this is only a simulation, using component models which don't have all the characteristics of a real circuit. In a real Joule thief the efficiency curve could be quite different.

• What about the Buck converter efficiency graph I added above? Doesn't it show that higher voltages are more efficient, or am I just dreaming? And what is affected by the I-squared-R losses -- the transistor? The schottky? The cell internal resistance? For the same power, if you double the voltage, don't you halve the current? This has to have an effect as well with LED driver circuits, or I really am imagining things. But I'm convinced there's some greater factor stealing what would be efficiency gains at higher voltages (but I can't prove it). – MicroservicesOnDDD Mar 23 '18 at 6:24
• "What about the Buck converter efficiency graph I added above? Doesn't it show that higher voltages are more efficient" - No, it shows that the lower the input/output ratio the better the efficiency. This the same a boost converter, only the output voltage is lower than the input voltage rather than higher. – Bruce Abbott Mar 23 '18 at 8:31
• "what is affected by the I-squared-R losses -- the transistor? The schottky? The cell internal resistance? " - yes to all (though your simulation did not address cell internal resistance). "For the same power, if you double the voltage, don't you halve the current?" - only at the output. Input current is the same, but the duty cycle changes so rms currents (and losses) change. – Bruce Abbott Mar 23 '18 at 9:23
• I don't respect your answer much, as it seems to be overly critical, and you were not patient enough to find out if a brusque answer was called for. #1. I'm developing a flashlight for the third world, where people don't have enough to eat, and might be able to find a dead battery at the dump, that is, if they can read. #2. I'm a Senior Developer of 15 years, #3. I've done more than 1400 LTSpice simulations to try to understand, #4. With pulse-skipping, and other evidence, I still think the generalization is true. I just don't think I've supported it well. Any ideas? – MicroservicesOnDDD Mar 30 '18 at 7:26
• If I come across as brusque or overly critical it's because I hate broad generalizations, particularly when they are only marginally true (or even flat out wrong). You should too, because clinging to them is not helping you to achieve your goals. – Bruce Abbott Mar 31 '18 at 0:20