PC high-current PWM fan controller issues

[ratdude747]

Reviving this as I've seen some inspiration that may allow for a retry of this design:



My suspicion (which was never verified, as I moved right around the time I posted last) was that the issue related to the LC output configuration not having anywhere to drop ripple voltage. Which makes sense, as a "normal" linear supply filter (no inline resistor before the Cap) like I was using creates an output equivalent to the Vpeak of the input rectified waveform, not the RMS or midpoint. Freshman level stuff, and I flunked it  . What is needed is a decoupling capacitor (to convert the PWM signal to an analog DC value) and then an RC stage (aka the linked capacitance multiplier, ideally using the same model MOSFET as the first stage) to filter out the ripple, no massive inductor needed. This will probably throw a wrench in the works for the PCB layout (double the # of mosfets with heatsinks, which I may try to run without), but that's besides the point since the old design was junk and not working.

I've been bogged down with work and homeowner chores, and may not get to this any time soon... But with any luck I'll have some time over the next few weeks to breadboard a circuit and give it a test spin.

Am I possibly on the right track?

[Ian.M]

The easy option would be to use a LM2596 or similar buck converter module.   They are readily and cheaply available on EBAY.   To add PWM output voltage control, simply disconnect the bottom of the voltage set divider chain from ground and inject a control voltage derived from the low pass filtered PWM signal + an OPAMP.   The reference voltage is about 1.2V,   Inject that voltage and the output will go to zero, Inject 0V and it will go to the voltage set by the divider ratio.  Inbetween you get linear control from Vref up to the max voltage set.  You'll need a small 5V regulator feeding a dual rail-to-rail I/O OPAMP, one section as a comparator to invert and buffer the PWM and limit it before filtering so the result isn't dependent on the motherboard logic levels, and to get it the right way round (100%=0V) for the LM2596 and the other to buffer the filtered analog output voltage to provide a low impedance drive to the bottom of the LM2596 voltage set divider chain.

[xani]

Did you actually scope the input, output and drive voltage, and the waveform coming from mobo ? Or at least simulate it in LTSpice.

From what I understand motherboards will output 0-100% duty cycle PWM. That is NOT what a buck converter wants as a drive signal, or rather the relation between duty cycle and output voltage/current is not linear. IIRC mobo will produce 20-100% duty cycle, which means that even on slowest setting duty cycle might be high enough to drive your fan at high speed

I've built a similar project out of junkbox parts as I wanted to use PC fan in my ARM cluster, and just used micro to drive it (that used tacho and ADC for feedback). But it maxes speed at around ~30% duty cycle, mostly because I oversized inductor (or rather didn't had any other on hand).


My guess is your circuit "works" but mobo does not go to the very low duty cycle required to actually make your fans spin slower.

Doing it the "analog" way is very much possible, but considering that you could just get literally any micro that has an ADC and a timer to do it, it probably wouldn't be cheaper, or easier.

[oPossum]

I built and characterized some 4 wire to 3 wire converters. They are described is this thread: https://www.eevblog.com/forum/projects/pc-motherboard-pwm-4-pins-to-3-pins-digital-converter/msg24112/#msg24112

[xani]

And have you tried same method (feeding PWM on from 0 to 100%) of characterizing on your project ?

Honestly it looks like it should work just fine, just that with no any real negative feedback the output voltage will depend both on actual PWM signal frequency, and current consumption of the fan (which will then vary on spinup)