D MacAlpine said:
PWM works for incandescent lamps and LEDs. Apparently it is the best way to dim an incan because it runs the lamp at spec, improving efficiency over underdriving it.
PWM is the worst possible way to get efficiency out of the LED.
Actual measurements:
Improvement in LED efficiency by dimming using constant current vs. PWM:
If you'd like the numbers:
Kram said:
Just to weigh in on this, most of the "flickering" people complain about is only visible when the source (i.e., the light) is moving relative to the observer. Then you can see that the LED is on part of the time and off at other times. The PWM method can be a much more efficient way to control an LED's apparent intensity than a resistor. Most of the PWM schemes now in use operate at frequencies that are not that terribly high above the threshold of your eye's persistence.
Interestingly, most of the circuits used for single cell lights boost the voltage by building up a magnetic field in a coil and a charge in a capacitor and then discharging it through the LED. But if you use an oscilloscope to look at the voltage, it is actually a train of pulses. Generally, though, the frequency using this method is much higher than that normally associated with PWM control, so you can't really see it no matter how fast you move the light.
Mark
On your first point, my eyes are faster than a lot of folks, and when I am hiking at night, many of the PWM lights actually bother me.
On your second point, that is a falacy, and is very wrong. Some companies use too small of a capacitor, and there is ripple in the waveform, but it doesn't actually turn off. There is a design I saw where they skipped the capacitor entirely, and it isn't a regulated supply anymore, but more of a modified PWM.
An example is the MagLite LED. Zero volts is the black line at the bottom of the screen, and every solid line going up the screen is 0.5 Volts. As you can see, there is ripple in the waveform, but it doesn't actually shut off:
I've got the actual light output waveform somewhere but I can't find it at the moment.
Other switchers have adequate capacitance, and have 0.002 Volts of ripple, and 0.05 Volts that are narrow spikes when it switches. As I recall, the voltage at the time was 4 volts, and the current was 1 Amp. The amount of DC ripple works out to 0.05% if you ignore the narrow spikes, and even the narrow spikes are only 1% of the voltage.
.
Bertrik said:
I have limited practical experience with PWM but I think that a slower flash rate will not give a much higher runtime. There may be a slight effect due to switching losses, but at several kHz or so I cannot imagine that the switching loss already plays a significant role (some switching power supplies go up into the MHz range!). The most important factor that determines the runtime is the duty cycle, e.g. a light at 25% duty cycle should have an approximate 4 times longer runtime than a light at 100% duty cycle.
If a switching supply is designed well, to keep it's losses under 5% (and 2% is practical), there is another factor that plays a big role.
Batteries have resistance in them, and losses occur due to the resistance in the cell.
This is a Current^2 * Resistance loss factor.
With a PWM, you are hitting the battery with the full load current. Which a switcher, you are reducing the load current on the cell, especially if an adequate capacitor is used at the input of the switcher. This capacitor, and the inductor cause a lower continous load current on the battery.
This reduces the losses in the battery by a Current^2 factor. There is an old post I did back in 2004 here, where I covered this in more detail, and it adds up.
So, you have a reduction of losses in the LED, which can be up to 228% more efficient than PWM at 2% PWM duty cycles, as compared to a constant current power supply (such as a linear or switcher).
Add on to that, the reduction of losses in the battery, due to using a switching power supply, and your gain can be very substantial.
Furthermore, folks often forget that a Luxeon has roughly 700pf of die (the part that makes the light) capacitance, combined with the ESD diode that is under the die. When just using PWM only, there are losses at the time the PWM "switch" is thrown, which amount to 50% of the power required to charge up that 700pf capacitor.
The loss here, is about the same loss on the gate drive, to drive the upper and lower MOSFETs in a switching power supply, example part:
http://www.fairchildsemi.com/ds/FD/FDN335N.pdf