Hi David and all, tx for all the input. I just got 'switching power=20 supply design" by pressman et al and am digging into it. It is looking like the driver for the PFET in a buck type configuration=20 makes all this more troublesome than it is worth doing. To get the=20 necessary hard turn on/turn off requires yet more parts, and by the time=20 you're done with a DIY buck you could have done it for less $ and fewer=20 parts by using a $1.25 dc/dc. My situation is that I'm about to start 2 related LED designs, one with=20 44 drivers per unit and the other with 24, to be made in hundreds and=20 hopefully thousands qty. I'll have a microcontroller and fpga if needed=20 available, and in order to get the very high PWM rates and lots of=20 dimming steps a custom repurpose of dc/dc chips (be it constant voltage=20 or constant current designed for LEDs) is required (or at least very=20 desirable), so most of the pieces needed to make the whole buck control=20 are already needed, dc/dc chip or not. So I'm spending some time=20 (luxurious) to fiddle around while the funding arrives. Anyway. It is looking like spending the $1.25 for a dc/dc chip to avoid=20 having to make a driver for the PFET switch is worth doing all by=20 itself. An alternative 'totem pole' with 2 or 3 bipolars plus some=20 resistors winds up being close cost-wise and now there is a pile more=20 parts on every pcb... Thanks again everyone J David VanHorn wrote: > You've almost reinvented the hysteretic switcher. > > Something to think on, for LEDs you don't really need a constant current > supply, with its high output impedance and tight regulation. You mostly > need a "steady" current supply. Current ripple is tolerable. The human e= ye > can't perceive it. > > > On Wed, Jun 5, 2013 at 5:21 PM, Isaac Marino Bavaresco < > isaacbavaresco@yahoo.com.br> wrote: > >> I had understood that you would put a resistor in series with the >> transistor. After looking at your schematic I realized that you are >> relying only in the transistor's internal resistance. >> >> If you replace in my post the word "resistor" with "transistor's >> internal resistance" it will make more sense. >> >> Actually you are dissipating the excess energy in the transistor's and >> capacitor's internal resistance. >> And you will have REALLY LARGE current spikes. >> >> >> Isaac >> >> >> Em 5/6/2013 19:58, Isaac Marino Bavaresco escreveu: >>> Without an inductor your regulator will be linear, it doesn't matter >>> whether it is frequency switching the transistor or not. >>> If you do the math, you will see that the extra energy will be >>> dissipated in the resistor and transistor and will be the same as in th= e >>> standard linear regulator. >>> All that you have is a linear regulator with a lot of ripple voltage an= d >>> current. >>> >>> Now, if you use an inductor then it may become a completely different >>> beast. The inductance acts as an "inertial flywheel" for the current, >>> limiting it without converting it in heat as in the resistor and >>> transistor (of course it is not an ideal inductor, it has a tiny >>> resistance and thus will waste some power). >>> The inductor limits the raise and fall of the current, storing energy a= s >>> B field while the current is increasing and supplying current while the >>> transistor is off. >>> >>> >>> Isaac >>> >>> >>> Em 5/6/2013 17:30, Jesse Lackey escreveu: >>>> Hi all, I'm looking at dc/dcs for running 400mA - 1Amp highpower LEDs >>>> with some special requirements, and in thinking about various schemes,= I >>>> wonder how (or if) something this simple would work. >>>> >>>> Setting aside the minor additional complexity of making a >>>> constant-current vs. constant-voltage dc/dc, what if there were a >>>> circuit like: >>>> >>>> +12V-->PFET-->Cap-->Output(say 5V target). >>>> The output would have a voltage divider (i.e. feedback) going to a >>>> comparator, which has a reference voltage for its other input. The >>>> comparator would turn on the PFET if the feedback voltage were below t= he >>>> reference voltage, and turn off the PFET if not. >>>> >>>> Schematic... >>>> >>>> >>>> Now I haven't even tried a simulation (which I rarely do, I don't do >>>> much analog that isn't largely 'cookbook' with tweaks) much less built >>>> it, so I have no 'real data' on how well or if it works. >>>> >>>> But it sure seems like it would operate more or less correctly. As C1= 's >>>> value gets larger, it takes more time to charge (thru PFET) and >>>> discharge (thru load), and as the comparator hysteresis gets smaller, >>>> the PFET switching on-time gets shorter, so it switches more often (le= ss >>>> efficient) but you have less output ripple. >>>> >>>> If the comparator were an opamp, driving the PFET gate directly, you'd >>>> have a DIY linear regulator. By using a comparator with hysteresis, y= ou >>>> get a dc/dc. I think. >>>> >>>> For my application, I will know the load (a string of LEDs being >>>> high-speed PWM dimmed), the approximate needed output voltage to light >>>> them at the designed current, and will be doing a little trickery to >>>> handle all that. I can tolerate a moderate (humm... maybe 10%? TBD) >>>> output voltage ripple. >>>> >>>> This seems somehow too simple to work reasonably, but I'm not sure why= .. >>>> C1 would have to handle the ripple current and the inrush current (= as >>>> would the PFET - something that would need a bit of additional design >>>> work, can be handled a few ways), R3 has to be low enough value to tur= n >>>> off the PFET quickly when the NFET turns off, but high enough value to >>>> not be too wasteful when the PFET is on. >>>> >>>> I like that I can set the output ripple by choosing C1's value and the >>>> comparator hysteresis. >>>> >>>> Thoughts? >>>> What's going to fail first if run 24/7 for years? >>>> Similar, polished, known-working designs anywhere? >>>> >>>> Thanks all! >>>> J >> >> -- >> http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive >> View/change your membership options at >> http://mailman.mit.edu/mailman/listinfo/piclist >> --=20 http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist .