120 Hz Flicker in LM3445 LED Driver

Benny:
LM3445 AN-1935 demo board is a valley fill buck. It has very little 120Hz LED current ripple compared with other buck designs. There ia a few ways to furtrher reduce the 120Hz ripple in the LED current. 1. Increase output cap C12. 2. Increase inductor L2. 3. Increase switching frequency.
Dvaid Zhang

David:

Increasing the output cap, C12, from 1uF to 6.8uF and then to 68uF shows very little effect on the 120 Hz ripple magnitude (2%).

The switching frequency is set at 400 kHz. How will increasing this frequency remove the 120 Hz variation?

The inductor value is 150 uH. Increasing its value will reduce the current ripple through the inductor at 400 kHz. (See Eqn. 22 on page 23 of the LM3445 data sheet.) How will that affect the 120 Hz ripple?

As observed on an oscilloscope, the average voltage across the LED increases slightly as the 120 Hz, full-wave-rectified input voltage peaks and then starts to decrease. Then, the LED voltage decreases to a stable lower level until the next Vin peak. As I said, this change is about 2%. That means that the average LED current must also increase slightly at the same time. The LED current is being controlled, cycle-by-cycle at 400 kHz, by the LM3445 Isns pin. Apparently, when Vin peaks, pulling Vbuck upward along with it, something is causing the Isns function to allow an increase in cycle-by-cycle peak inductor current (and, therefore, peak LED current ) for the duration of the Vbuck disturbance. Otherwise, the average LED current, as seen on the 'scope at 120 Hz, would not change.

It seems that there is some kind of coupling going on wherein the LM3445 is reacting to the peak of Vin and is making a change in the threshold of the Isns function, thereby allowing the peak and average inductor current (and LED current) to be higher.

The 2% variation in LED voltage occurs with a forward-phase-controlled dimmer at 100% supplying the AC input. Without the dimmer, the flicker effect is a bit less -- maybe 1.5%. But it is still noticeable to my iPhone video camera.

How can I remove this interaction between the behavior of Vbuck and the current through the LED? The valley-fill circuit is apparently independent of the dim-detect circuit of the LM3445. Valley-fill simply ( and dumbly) supplies an operating voltage level for the buck regulator. But the 400 kHz, cycle-by-cycle, peak inductor current seems to respond to what is going on in the valley-fill circuit. Why? and how?

Thanks,

Benny

Benny:
Please check the voltage on the FLTR2 pin. If there is a 120Hz ripple on that pin, that can cause the 120Hz ripple on the LED current.
David Zhang

David,

FLTR2 pin shows a DC level plus high-frequency variation; no 120Hz variation.

Could the 120 Hz variation across the LED modulate the current from the PNP transistor which charges Coff and sets the OFF time of the LM3445?

If that were so, then the OFF time would vary with the 120 Hz "bumps" that I see in the LED voltage. Could that variation in OFF time translate to a change in the average LED current (which is what I see when I have the scope set to display 120 Hz time span?

In AN-2034, page 6, I see that the PNP current source is tied to Vcc, not placed across the LED. In that application, there is no valley fill circuit and the 120 Hz variation across the LED is probably large, hence the need to set OFF time from a steadier source (Vcc).

I am trying to commercialize my LM3445 design as an LED driver. So, time is wasting.

Can you recommend a power supply or LED driver design engineer (perhaps a consultant) that I can hire to solve this problem? Someone who has some experience with the LM3445, perhaps?

Thanks,

Benny

Benny:
The off time should be fairly constant since the current charging the Coff is a constant current. Can you show me the LED current scope shot? Sorry I can't help you finding a power supply engineer for your project.
David Zhang

Benny Smith said:

I have measured the amplitude of the voltage variation across the LED and it amounts to just 2% of the total (mostly DC) voltage across the LED.  But that 2% is definitely noticeable by any video camera.

LED’s are acting like a Zener diode. 2% of voltage change will have a huge current change trough the LED’s and so a large power –light emitting- change.

I think it would be possible to change the design from ‘Constant LED Voltage’ to ‘Constant LED Current’ using the Coff input.

I don’t have an EVM for this device otherwise I would make a test.

Leo,

Thanks for your reply and insight.

I assumed that the design shown in the LM3445 data sheet is a "constant-current" realization, since the current-sense (Isns) pin responds to the voltage across the sense resistor, which is directly proportional to the peak inductor current.

Benny

Benny Smith said:

which is directly proportional to the peak inductor current

That’s correct, but the voltage across the inductor (Vbuck-Vled) changes and so the power stored in the inductor.

David,

A scope shot of the LED voltage. I have not rigged up a technique to measure the LED current yet.

Notice that the LED-voltage average value is around 41V but the variation is about 900mV P-P. The variation seems to have abrupt transitions. I would not expect that effect from the relatively gentle changes in the valley-fill voltage. But the period of the variation of the LED waveform is definitely ~8ms (120 Hz). There are numerous ups and downs within each 8 ms period.

Any ideas?

Benny

David,

I forgot to attach the LED scope screen shot.  Here it is.

Benny:
You need to get a current probe to measure the LED current. If you can not get a current probe, insert a resistor in the LED string and measure the voltage difference across the resistor.
David Zhang

I’m wonder if you were able to solve the 120Hz flicker problem?
And if, how?