[ell-i-developers] About the LED driver SMPS approach and topology

  • From: Pekka Nikander <pekka.nikander@xxxxxx>
  • To: ell-i-developers@xxxxxxxxxxxxx
  • Date: Thu, 5 Jun 2014 10:25:26 +0300

> Interesting thing is that LED efficiency seems to go down as instant
> current goes up, it might be worthwhile to have a SMPS which sets the
> voltage in "general area" of bit more than forward voltage of LEDs and
> then use PWM-dimming across a resistor.

My understanding is that that the best way is to use a high-frequency (500 kHz 
- 4 MHz) PWM/PFM controlled soft-switching SMPS, followed with a low-pass 
filter that essentially produces directly constant current.  Then you use the 
PWM/PFM control to adjust the current, and use an ADC channel to measure the 
current.  You don't need to care about the voltage, as it will become 

My understanding is that implementing a soft-switching buck is not that hard as 
such, but will require a few iterations due to the PCB parasitic inductances 
(and maybe also capacitances) starting to play a role.  At 4 MHz the coil 
inductance may be as low as 1 uH or just a few hundred nH, and the PCB 
parasitic inductances may be several tens of uH.  The really hard part is then 
achieving real ZVS for the FETs, as that may require a few tens ns of accuracy, 
but than can be achieved either using an analog circuit or an FPGA.

If we consider a soft-switching SMPS too hard for us for now, then my 
understanding is that a constant-current mode SMPS working in the 50-200 kHz 
frequency is the second best.  That we can do already, it is relatively easy if 
we don't care too much about EMI.  We still get pretty constant, adjustable 
current, leading to minimal loss in the LED itself.  However, the switching 
losses will be higher and the coil/transformer and capacitors clearly larger.  
The coil will be in the order of a few hundred uH, and the parasitics are 
more-or-less insignificant.

> As an added bonus it's easier to have higher resolution PWM control when
> the frequency can be in range of 100 Hz rather than in 100 kHz-range.

I don't think so.  I think they are both equally easy.  However, when you go 
over some 200 kHz (i.e. less than 250 CPU cycles/PWM cycle), then the accuracy 
stats to suffer too much, especially in the low end...  However, in the low end 
you can probably use PFM instead, depending on the details of the SMPS topology.

I think we Otso should take a face-to-face session on this at some point.


Attachment: smime.p7s
Description: S/MIME cryptographic signature

Other related posts:

  • » [ell-i-developers] About the LED driver SMPS approach and topology - Pekka Nikander