OK, the improved Schmitt Trigger helped. Still not right, but the transistor is cooler (but hot). I can see a definate duty cycle improvement looking at VR2,3. I've decided to skip R10, and replace the transistor with a TIP107, a Darlington PNP. I've also bought some 3A Shottky's (1N5821) to try out.

*************************************

Success! It is finally converting. 1.89A in, 2.15A out, the new transistor is better, still getting hot, but I have it well heatsinked, and everything is good.

I don't like the Q2 arrangement one bit, but this aspect of the project is complete. I'll be following up with another diagram eventually.

Lessons learned:

The Schmitt Trigger is critical. You need as wide as swing as you can get by with. The op amp does not go rail to rail, you have to compensate for that somehow.

Apparently a Schottky is pretty important. I still think a conventional diode will work, but I knew the diode might be under rated. I'm not sure if this is a rule, but you need to match the diode current to the output current.

Final Schematic

Switching Regulator Parts List
U1 – JRC4565 (1458 style dual op amp w/ better specs)
VR1 – 10V zener diode, 5W
VR2, 3 – 3mm Red LEDs
VR4 – 10V zener diode, 1W
Q1 – TIP107 PNP Darlington power transistor
Q2 – 2N2222 NPN transistor
D1 – Red LED
D2 – Green LED
CR4 – 1N5821 3A 30V Schottky
R1, 11 – 1 KΩ ¼W resistor
R2 – 2KΩ Variable resistor
R3, 5, 6, 7 – 10KΩ ¼W resistor
R4 – 7.5KΩ ¼W resistor
R8 – 4.7KΩ ¼W resistor
R9 – 150Ω 1W resistor
R10 – 360Ω ¼W
C1, 2 – 470µF 35V Capacitor
L1 – 350µH

 

I have to tip my hat to a chap who designs a switcher from generic parts, just for the hell of it.

Bill is quite right, the TIP107 darlington has its own base-emitter resistors (they're there to improve the turn-off time), and so using this there is no longer an issue with Q1 base floating when Q2 is off. What I didn't spot before though is that Q2 can float because of VR2 and VR3 - that b-e junction definitely needs a pulldown resistor or Q2 will be very noise-prone. I'm not sure what VR2 and VR3 are for, but it will work well without them.

I was originally thinking that maybe Q1 was turning on spontaneously due to noise pickup and leakage currents and this was what was causing it to get hot, but it seems this isn't the case - saved by the low gain of a power PNP! I can tell where the heat is coming from now though, after looking at the TIP107 datasheet: It's simply the high Vce(sat). A TIP107 will drop about 1.1 V at 2 A, so it will dissipate about 2.2 W. This dissipationfigure can be dropped drastically by using a P-channel MOSFET (clamp the g-s junction with a zener to keep Vgs within spec). Really efficient buck regulators use a N-channel MOSFET (lower Rds(on) than a P-channel) with a charge pump to float the gate voltage above the input supply voltage.

You're right, the power diode CR4 can be non-schottky, but the latter are preferred for their greater efficiency (lower Vd) and fast response. Watch out for the power dissipation in this diode - check it's not getting too hot, as that's a classsic gotcha. Ultra-efficient switchers use an actively switched MOSFET instead of a power diode to reduce the dissipation further.

To improve the noise immunity it wouldn't do any harm to decouple pins 3 and 6 of the op-amp. The switching will refer lots of noise back to the input supply rail, and this will be fed back to the references. VR4 will be intrinsically a little noisy too - it's always a good idea to decouple zener references.

I do hope I'm not coming across as an interfering busybody here, as it's been kind of neat watching this design unfurl. I'm taking a cue from Bill's wonderful tagline - "Good enough is enemy of the best". Not a bad philosophy to live by.

 

No problem, I like talking shop. It's the reason I posted the sucker. I used VR2 and VR3 as quick and dirty voltage droppers, since I knew the op amp made a lousy comparitor. Even though the op amp didn't get near the power supply rail the LEDs soaked up excess voltage. Actually it wouldn't work at all without at least one, or a low voltage zener (which is how I'm treating them). The fact it was a decent diagnostic was a bonus.

Noise isn't as big an issue I think, since switchers basically are noisy to begin with.

I'm going to do a redesign, I really didn't like Q2 and R9, mostly R9, it generates its own share of heat (the enemy).

If you check my AAC blog out you'll find I do quite a bit of writing for this site, mostly for beginners. If I make a design I'm happy with it will end up being made into an article for Volume VI of the AAC book. My blog points to my other articles.

If you want to learn, teach.

 

If you want to learn, teach.

Very true. Bill's fine attitude is common amongst the best engineers, and the jealous hoarding of knowledge is common amongst the worst.

Now you've got a darlington for Q1, R9 could go up in value and save a bit of power. R9 could go even higher with a MOSFET for Q1.

I've just had a look at the op-amp datasheet, and I realise the purpose of VR2 and VR3 - it's because the op-amp will only swing down to about 2 V above ground, and Q2 won't be turned off without either dropping at least 2V of the output, or using a rail-rail output op-amp. Q2 still needs a b-e pulldown resistor with those diodes in the base feed or it will be very prone to turning on in response to EMI - I could relate some horror stories...

Omitting decoupling caps on the voltage references on the op-amp inputs won't just give extra noise on the output, it could also seriously compromise the stability of the circuit, as two unwanted feedback loops have been introduced. Imagine that the turning on of Q1 dips the 12V rail briefly (to 10 V, say) - all of a sudden the reference voltage at U1 pin 6 will dip proportionally, and the comparator will react accordingly. Best case, the circuit will have poorer regulation and increased output noise; worst case it will become unstable. The second inadvertent feedback loop to U1 pin 3 will be less critical as the noise will be attenuated somewhat by zener VR4, but it's still advisable to decouple that too. And the supply rail of U1 also, as the electrolytic C2 can't supply current fast enough for the op-amp output that's whacking from min to max as fast as it can.

I look forward to the blog - it's very good of you to take the time to do this sort of thing.

 

There will be another design, on this thread. The initial 5V power supply still looks viable, and would be usefull for some game machines. Having a kid tends to give me project ideas.

Another project idea is a variable switcher, something that can go between 1V to 20V, say about 6A, as a bench supply. The old ½A variable PS I built after college is a little long in the tooth.

If you have any ideas feel free to suggest em, don't promise I'll use it but if I like it I'm not too proud to use other peoples concepts.

I'll probably build an electronic load first, I really needed one to test this out properly, and they are simple enough.

The drawing package (if it can be called that) is also available from my blog, under the name of PaintCAD. It is a bunch of templates meant to be used with a graphics package such as MS Paint.

I may yet try a MOSFET, I have an embarrising admission to make, they didn't teach them in my college, they didn't exist yet, so I'm not comfortable using them. I worked on equipment that had them, one in house process machine had to have them replaced every year or more (it had 6 parallel process, each with two) because while they could handle the load the pins were too small on TO220 to properly carry the current. Their solution was to make brass strips that went down the length of the pins, I woud reuse the brass strips and solder a new MOSFET using a hotplate to raise the temperature to improve solder flow and a soldering iron. Maybe my last design could be a good canidate, since Q2 should provide plenty of drive, and could handle the surge gate current.