DIY LED Controler/Driver

[theatrus]

Looks nice. 1kHz is fine for human flicker resistance, but I've pushed 50kHz on the right drivers which helps a ton for video and photography, especially as the dimming doesn't seem to trigger the anti-flicker algorithms in some cameras.

The whine as you noticed is likely the 1kHz leaking back through something on the power supply. Staggering / dithering should mitigate it for sure.

 

[Sral]

Looks nice. 1kHz is fine for human flicker resistance, but I've pushed 50kHz on the right drivers which helps a ton for video and photography, especially as the dimming doesn't seem to trigger the anti-flicker algorithms in some cameras.

The whine as you noticed is likely the 1kHz leaking back through something on the power supply. Staggering / dithering should mitigate it for sure.

Well, sadly the PCA9685 caps out at 1.5kHz. If I want more I would need something faster, like the integrated 20+ MHz timers on the "faster" microControllers (PI Pico and ESP32 should do the trick, if I'm not mistaken). What signal source were you using for your 50 kHz?

The FOD3182 I'm using should be capable of something close to 50kHz with maybe some inaccuracies at very low intensities, since their distortions are in the range of <40ns. I would think that my jumper wires and connectors are a much bigger hindrance towards these frequencies

I agree about the whine, handling a 92W load @ 24V (e.g. 3.8A) with 1.5kHz at 50% duty cycle is a serious strain on my 100W power supply (Meanwell LPV-100-24). That's why I was considering buffer capacitors on my 24V line, but didn’t install them due to concerns that the AC current might heat and blow them up or the added capacity might trip the over current protection of the MeanwellPowerSupply.

 

[Sral]

Just for reference, here is the brightness-controller that the manufacturer included with the lamp:

The coil for the wireless remote blocks a few things, but you can see one MOSFET underneath it on the right, which dims all channels by itself. So pretty identical, my only guess is that it probably uses a higher PWM frequency. Sadly there are no markings on the IC that generates the PWM.

 

[Sral]

Okay, I revisited that bypass-capacitor I was planning. I bought a few of these Cornel-Dubilier 1mF 35V capacitors and originally I planned to go overboard and add up to 6 of these (one per channel).

When one considers that this bypass capacitor (together with the capcitors included inside the power supply) is supposed to smooth the load current almost to a constant DC current one can approximate the current flows over time. FOr this to be true the capacitance needs to be large enough compared to the nominal resistance of the load and the applied frequency:

capacity >= 10/(resistance_nom * frequency)

In my case:

resistance_nom ~ 24V / 3.83A ~ 6.266 Ohm
frequency = 1500 Hz

capacity > 10/6.266/1500 ~ 1 mF

When that is fulfilled the calculation becomes a bit easier:

I_nom = 92W/24V ~ 3.83A # lamp current under full power
duty = 0.0 ... 1.0 # PWM duty cycle
I_average ~ I_nom * duty

I_discharge ~ I_average * (1-duty)
I_charge ~ I_average * duty

Which one can understand principally as well: when the duty cycle is low, the capacitor has a long time charge back up, so the charge current is low. Vice versa, when the duty cycle is large the charge current increases, so I_charge is somewhat proportional to the duty cycle.
Conversely, when the duty cycle is low, the discharge current supplied to the load in the short active time is very large, since the capacitor needs to make up the difference between the average current from the power supply and the much larger nominal current, that the load draws. Vice versa, when the duty cycle increases, the discharge current becomes smaller, since more and more current comes from the power supply until at a duty cycle of 100% the power supply takes all the load and the capacitor is idle.

In this picture the ripple current, described by its RMS (RootMeanSquare) becomes:

I_RMS^2 = 1/T_period *  [ I_discharge^2 * T_discharge + I_charge^2 * T_charge]
I_RMS^2 = 1/T_period *  [ I_discharge^2 * duty * T_period + I_charge^2 * (1-duty) * T_period)]
I_RMS^2 = T_period/T_period *  [ (I_average*(1-duty))^2 * duty  + (I_average*duty)^2 * (1-duty))]
[...]
I_RMS = I_nominal * sqrt(duty - duty^2)

For my case of I_nominal = 3.83A this becomes:

As one expects, the most load on the capacitor happens around a PWM duty cycle of 50%. When one considers the datasheet, the maximum RMS ripple current for my capacitors are 2770mA @ 100kHz and +105°C. I don't know yet how this relates to my conditions of 1.5 kHz at room temperature, but I'm assuming that I'm more or less safe.

To make sure, I took the Dissipation factor of my 35V capacitors, which is tan(delta) ~ 0.12 and calculated the Equivalent Series Resistance (ESR) of the capacitor and from that the electrical power on the capacitor using the maximum ripple current of about 2A:

ESR ~ tan(delta) / (2*Pi*frequency * capacity)
P_el ~ ESR * I_RMS^2

Which for me equates to about:

ESR ~ 0.12/(2*3,14159*1500 Hz * 0.001 F) ~ 0.013 Ohm
P_el ~ 0.013 Ohm * (2 A)^2 ~ 50mW

That sounds absolutely manageable.

My much smaller diodes for example had a thermal resistance of about 300 K/W. I'm guessing from this paper, which gives a thermal conductivity of about 10-20 W/m^2/K for the surface for still air and the dimensions of my capacitor, that I should have something like 50 K/W, which sounds reasonable, since the capacitors are much larger than the tiny diodes. This means my capacitors shoud heat up by about 2,5K (in the worst case maybe something like 10K).

 

[Sral]

Long story short: I think I can try including a single 1mF capacitor in my unit.

First, it should be able to handle the ripple currents already by itself. Additionally, it never sees the whole ripple current, since it is also supported by the capacitors inside the Meanwell LPV-100-24 power supply.

What do you think ?