Buck converter with discrete BJT gate driver has problems?

Hello,

The attached Buck converter (LED driver) has a P mosfet driven by a discrete BJT gate driver.
[V(in) = 12V]

The transistor Q1 is being driven into saturation, but still manages to switch the PFET off quickly (the turn OFF gate voltage plateau is only 82ns).

How can this be?, why is BJT Q1 not taking ages to turn off?, ...after all it was saturated

Also, the NPN Q2 suffers a periodic reverse voltage of -9V on its Vbe. Will this be OK?, ..since its only a spike, and not a persistent reverse voltage of -9V on the Vbe.

Here is datasheet for MMBT3904 showing max allowable Vbe on MMBT3904 is -6V
https://www.electro-tech-online.com/custompdfs/2013/05/2N3904.pdf

(All waveforms below)
Schematic; PFET gate voltage; Q2 Vbe

By the way , the LTspice simulation is also here...

Why are you not driving the MOSFET directly from the ltc1693 driver?

What's the purpose of D26?

D26 is reverse polarity protection.
Sometimes the Vin is 24V and we couldnt use the LTC1693-3 to drive the fet.
(to be honest, in the actual circuit the LTC1693 is actually a 6 pin PIC microcontroller with an internal comparator.....but i didnt have it as a spice model so used the external comparator and gate driver instead....the purpose of the post is about the BJT driver, so its ok for me to tell this "lie")

PIC used is PIC10F200
https://www.electro-tech-online.com/custompdfs/2013/05/41239a.pdf

...The workings are exactly the same as above, its just that the actual circuit uses the comparator inside the PIC10F200 to drive Q1

I forget, how do you convert the text file to something spice works with?

you just name it with a .asc ending.

I think Q3 turning on pulls Q2 off.
If you add a 2K between R3 and R8 it will reduce the Vbeo to about 4 volts with a small loss in turn on/off time.

Thanks ronv, the 2K gets the Vbe back within limits...that is , until the Vin goes up to 24V, and then Vbe goes back above 6V, but incrreasing to 4K takes it back below 6v again, so youre a winner.
(by the way, this schem is of a real cct on a real product which got put in production because the company were fed up of being ripped off by design consultancies, so did this themselves, "warts and all")

I don't know if it's a concern in this circuit but you can reduce the ≈1.6μs turn-off storage delay for Q1 by connecting a capacitor of about 1nF in parallel with the base resistor R1. That gives a turn-off delay of about 0.5μs.

Thanks , i see what you mean.
Though in fact, doing that has the effect of increasing the switching frequency which increases switching losses.
-I've found that the switching frequency can be brought back down though by putting a rc filter in the feedback path, so maybe its better with the cap.....no, just tried it, the switching losses are worse with the inF across the 1K.

Incidentally, if you note the schematic, you can see the LEDs are in parallel.....thats because this company's staff aren't capable of doing a boost or buckboost converters and putting the leds in series......they appraoched electronic design consultancies, but were quoted prices of £30,000 plus to do this product with a buckboost and series leds.......so they said, blow that, we'll just do it ourselves with parallel leds & the simple driver seen in the top post......each led is rated to carry the full current, so if a led should hog all the current, then its rated to carry that anyway........seems expensive way of doing it....but cheap when you think they didnt have to pay the £30000 design fees.
...of course , the parallel leds should be matched in vf too...so theyll be expensive leds, but again, they avoid the £30000 design fees.
The fees charged by design consultancies are enormous, another product, a simple boost converter led driver with 12vin and driving 24 leds (three string s of 8 with 10r in each string) at 150mA in each led also cost them £30000 to have designed................thats half a days work to design something like that (plus thermal chamber time)...and they charge £30000!!!!

Why not use an LT3741 or an LT3755. This will achieve what you are trying to do with short/open circuit protection and inbuilt current limiting. And you can simulate both in LTSpice

In case you want to try a faster transistor for Q1, a 2N2369 is a much faster switch than the 2N3904.

Note that you can reduce the operating frequency of the system by adding several tens of millivolts of hysteresis (say 50mV) to the comparator. That will increase the ripple current through the LEDs but will not be noticeable to the eye.

Thanks Simon, i tryed to click "+" to thank you, but was holding a knife and fork and missed with the mouse and clicked "-" by accident...sorry, it wont let me re-do it.