PWMing of White LED LCD back light

On Sat, 23 Jul 2005 02:43:37 +0200, Sylvain Munaut wroth:

The LC filter probably isn't necessary. In fact, it could create a problem. Depending on the PWM frequency, the LC component values along with their parasitic components, and at some on/off ratios, you could get a higher output voltage across the LED than the 5 volts you are switching.

The safest thing to do is to simply PWM the supply voltage without any filtering.

Jim

It is very normal procedure to dim LED backlights with a PWM. Don't filter. Just switch the FET at a few kHz.

Meindert

First, you don't need the NPN. If your MCU is fed also with 5 V, the PWM output can directly command the PMOS. You don't even need a resistor. You might want to include it, in series with the gate of the PMOS, if you were concerned about radiation (EMI). The PMOS must have a gate threshold voltage lower than 5 V (as it was also in your circuit). Just take into account that the LEDs will be on while your MCU output is 0. No problem at all. However, if you still want to complement it, most MCUs with PWM allow you to invert the output inside the MCU itself.

Right. Ignoring interaction with the scan rate of the LCD, any pulse repetition frequency (PRF) of the PWM signal higher than about 1 kHz is ok. I have been designing circuits for lighting products based on high-power LEDs (1.2 W/LED), and using 3.9 kHz as the PRF, and no problem. The frequency is high enough not to be noticed by our eyes (even if you move your sight with respect to the source of light), and low enough to be able to use very small MOSFETs (with very low total gate charge). However, I would encourage you to take advantage of more duty ratio steps. I was using 256. At the low-light side of the range, the (linear) steps appear to be larger to our sight.

No, no, don't worry. If you guarantee that there is no 't' for which v(t) or i(t) is going to exceed the ratings, the LED does not care about the shape of the excitation. Most LEDs in this world are being excited with rectangular waves.

No, forget about it. In fact, delivering the average voltage to the LED would be worse in terms of chromaticity. It is better to send him the full rectangular wave. I can explain this better, but just as a summary: the instantaneous current flowing through an LED determines not only the luminous flux, but also the color of the light.

Yes.

No. It is worse if you use it.

PRF between 1 kHz and 5 kHz is probably best. You may need to choose some especific frequency related to the LED scan rate, not to have a bad interaction.

Ok, first thank you all for you answers.

See some comment below about Mochuelo answers.

No, the MCU is a Au1100 (32bit MIPS) and has 3.3V IO and when in sleep or early boot phase it's output might be Tristated or low and the backlight must be off.

In fact I said 64 as a worst-case value. It depends on the divider. Basically I have a 6Mhz clock that I can divide up to 4096 by setting a divider. Then I can choose until which step it's off or on. So by setting the frequency at 1.5 kHz, I would have up to 4096 steps to choose from.

Great, good to know, even easier then ;p

Sylvain

Ok, if the MCU IO is 3.3 V and the backlight input is 0 to 5 V, it makes sense using an extra transistor. However, if you have access to the positive and the negative node of the backlight subcircuit (i.e., if it is not internally grounded), I would use a single NMOS (with a gate threshold voltage lower than 3.3 V) and a high-ohmic resistor from gate to GND. Source to GND; drain to backlight negative; backlight positive to 5 V. The resistor would guarantee that the gate is low when your MCU is in sleep mode (regardless on whether the PWM output is high-z or low).

In this case, even 100% PWM would be OK because of the limiting resistors. If you are driving LEDs directly, you do need to be careful about the peak current. The specs I found on the Nichia website for a white LED on

Paul E. Schoen, President P S Technology, Inc.