EEVblog Electronics Community Forum
Electronics => Beginners => Topic started by: jazzclassics on May 11, 2013, 06:48:07 pm
-
hello,
Does anybody know how the below Negative voltage regulator works? Can it output 50V 5A to a 10ohm load?
Thanks a lot.
-
How it works:
- Q2 and Q1, sharing D8, form two rough current sources at about 9mA and 4mA ((VF(D8) - VBE(Q2))/R3) and (VF(D8) - VBE(Q1))/R7).
- I(Q2) feeds U1, an adjustable voltage reference set to -3V. I(Q1) pulls the gate of M2 to the negative rail.
- Q3 and Q4 are a differential amplifier, working as a (very) rough op amp. This matches the voltage on Q4's base (the feedback voltage from R1/R2) to the -3V reference, by driving the MOSFET M1.
Now, I had a couple problems with it:
1) I had to lower the Q1 current source from 4mA to 1mA (change R7 to about 2.7-3.3k), otherwise Q4 would be unable to pull M2's gate to the right voltage.
2) The feedback resistors are wrong for 50V; they're actually set to give 96V! Change R2/R1 to 3k/47k to get 50V.
As for your second question - yes, but the MOSFET will be dissipating 50W. You will need one huge heatsink!
This is a weird circuit. Where did you find it?
-
Another note - I see no substantial error by replacing the first current source (Q2) with a simple resistor. That lets you eliminate both Q2 and R3, changing R8 to about 6.8k and connecting it straight to the -60V rail. Also, the amount of current there (9mA) is wastefully high and results in one hot transistor (or resistor, in this case). I'm assuming that's what R8 is there for - sharing some of the heat. Just change the current to about 4mA (R3 becomes 680, or if using just one resistor, about 18k).
And another couple observations:
- It appeared to be borderline unstable, but I didn't investigate that. In the simulation I poked at it with, I didn't bother with things like the series resistance of C3, which would have an effect on that. Make sure it will be stable, especially with no load.
- Using TL431 as a voltage reference with that Q3/Q4 "op amp" is like dressing your dog up in diamond jewelry. Use a five cent Zener.
- C3 is unnecessarily large. Cut it by ten...
- 2N3906 has a maximum voltage of -40V. Use something like BC556 instead.
- The transient response is pathetic, mostly due to the Q3/Q4 op amp. Use a real op amp instead. They're quite cheap these days...
-
This is a weird circuit. Where did you find it?
Hi, Thank you very much!
I have found it in some old archived PSU design files on my backup HDD, while I was looking for some PSU for my diy semi-symmetrical amp.
I think that I might have downloaded the file from the www.elecfans.com (http://www.elecfans.com), but even I cannot find the original post now.
-
Another note - I see no substantial error by replacing the first current source (Q2) with a simple resistor. That lets you eliminate both Q2 and R3, changing R8 to about 6.8k and connecting it straight to the -60V rail. Also, the amount of current there (9mA) is wastefully high and results in one hot transistor (or resistor, in this case). I'm assuming that's what R8 is there for - sharing some of the heat. Just change the current to about 4mA (R3 becomes 680, or if using just one resistor, about 18k).
And another couple observations:
- It appeared to be borderline unstable, but I didn't investigate that. In the simulation I poked at it with, I didn't bother with things like the series resistance of C3, which would have an effect on that. Make sure it will be stable, especially with no load.
- Using TL431 as a voltage reference with that Q3/Q4 "op amp" is like dressing your dog up in diamond jewelry. Use a five cent Zener.
- C3 is unnecessarily large. Cut it by ten...
- 2N3906 has a maximum voltage of -40V. Use something like BC556 instead.
- The transient response is pathetic, mostly due to the Q3/Q4 op amp. Use a real op amp instead. They're quite cheap these days...
Hi, c4757p, Thank you very much for enlightenning me for how the circuit works. I tried all your changes and they all worked like a charm. :-+
I think maybe the original designer wants to make a good precision high voltage psu. I guess that the purpose of using discrete is that the parts are easier to source than high voltage op-amps. If just use TL431 to drive the mosfet, can you still get output voltage > 36v?
Why you think it might be unstable? How to simulate and create the unstable situation? I have tried simulating the circuit with no load but it looks having no stability issues.
The problem I found with this circuit is the output voltage changes with the DC input voltage as well as the load, ie. the output voltage is not stable enough.
If I change the q3/q4 to a real op-amp, what op-amp should I use?
-
Another note - I see no substantial error by replacing the first current source (Q2) with a simple resistor. That lets you eliminate both Q2 and R3, changing R8 to about 6.8k and connecting it straight to the -60V rail. Also, the amount of current there (9mA) is wastefully high and results in one hot transistor (or resistor, in this case). I'm assuming that's what R8 is there for - sharing some of the heat. Just change the current to about 4mA (R3 becomes 680, or if using just one resistor, about 18k).
It's possible the model doesn't get the TL431 absolutely accurate here, which is why you see no big error. I didn't simulate it myself, but try putting a big load on it, and running the DC input stepped.
The current source through Q2 would seem to me to be designed to improve the PSRR of this regulator.
If the current is stable across the resistor R5, where it develops the 1.25V reference for the TL431, then the 3V drop across the TL431 will be equally as stable as that current source. From the datasheet, a 1.4mV change in the reference voltage of the TL431 will result in -1V change in the voltage dropped across it. Since the TL431 is the main reference for the diff-amp formed by Q3/Q4, you want it to be as stable as possible, Also, knowing this, you want to make R11 as small as possible to improve it's immunity to current fluctuations, which result in voltage changes across it. 5K may be enough given that they use a constant current source, but you may want to make it smaller if you remove Q2. Personally I'd keep Q2 as it is there to improve PSRR.
If you remove Q2 and R3, (no more constant current source) then the current in that resistor string to ground (R10,R11,R8,Q2,R3) will change with input voltage. This will cause a changing voltage in R11 (the TL431 reference voltage) which causes the main reference voltage 3.0V to change. You and I could both do the math to find out the PSRR with and without Q2, but I think it will be better with Q2 ;)
-
- Using TL431 as a voltage reference with that Q3/Q4 "op amp" is like dressing your dog up in diamond jewelry. Use a five cent Zener.
Yep. I agree. My previous comment talked about the need for the constant current source to regulate the zener. Given that, it's even more awful that they use a TL341 with a nice current source only to use a crap 2 transistor diff-amp (which doesn't have any constant current source!)
- The transient response is pathetic, mostly due to the Q3/Q4 op amp. Use a real op amp instead. They're quite cheap these days...
yep. but don't use a 5-cent zener with a real op-amp, stick to the TL431, its pretty good here, once you have a decent op-amp.
-
I think maybe the original designer wants to make a good precision high voltage psu. I guess that the purpose of using discrete is that the parts are easier to source than high voltage op-amps. If just use TL431 to drive the mosfet, can you still get output voltage > 36v?
(Precision! Bahahahahah... :-DD High voltage! Bahahahah..... :-DD)
As long as the current draw is relatively constant, this circuit can easily be made to float on a resistor à la LM337. If it's not, it'll float on a Zener just as happily. Just be careful about what you're referencing voltages against.
Why you think it might be unstable? How to simulate and create the unstable situation? I have tried simulating the circuit with no load but it looks having no stability issues.
Yeah, that's why I said "borderline". Simulate it with a pulsing load and you'll see some pretty unsightly ringing on the edges; I'm not sure I trust that they'll be as well damped in real life.
The problem I found with this circuit is the output voltage changes with the DC input voltage as well as the load, ie. the output voltage is not stable enough.
Yes, it's quite poor. Most of that has to do with Q3 and Q4. You could also try replacing R5 with a third constant current source - that might improve it a bit.
If I change the q3/q4 to a real op-amp, what op-amp should I use?
LM358 is popular in power supply applications. TL072, another favorite of mine, has a higher voltage rating, so you won't have to waste as much power in that "floating resistor" I mentioned earlier. You're right, an op amp that will take the full 60V might be a bit tricky to find.
There's no reason not to use a discrete diff amp here, just not this one. This one sucks ass.
codeboy2k - yes, I agree, that TL431 ought to have a CC load in a better circuit, and the PSRR won't be very good (it wasn't). I meant "substantial error" as relative to the error already exhibited by the rest of this circuit, which easily washed out any stability in the TL431. I found this circuit to have positively awful PSRR even with an LTspice "perfect" voltage reference stuck up Q3.