Linear Pass transistor regulator - my design

[keantoken]

Hi.

I've been tweaking this design for quite a while now, and want to run it by the DIY community for feedback or suggestions. I have not built it yet, for lack of parts.

This isn't really intended for anything serious, I just started designing to see how well I could do. I figured that in the end I would have a nice lab supply with minimal ripple. Output is around 33VDC but can be changed by using a different zener diode. Adjustable Zener circuits work but not as well as using a single zener.

Rather than a capacitance multiplier, I decided I would go more the route of a specialized (super-sensitive) error amplifier keeping the output voltage steady.

If you want higher current you will need to lower the value of R3.

Q4 and Q6 act as a VERY sensitive feedback pair, which will react when current travels through the Zener diode. The 22 ohm resistor was a 1N34A germanium diode but... they don't exist anymore. The resistor works just as well here I suppose. Operating current through the 22 ohm is at about 20mA, but a .5W resistor is best because failure will likely result in oscillation.

C2 keeps the circuit from oscillating, though it seems from simulation that 100nF is over-conservative.

There is a hitch that I have not been able to work out, which is that the output is most steady after 0.2A, so ideally a class A or AB amp would have a standby current consumption of at least this much. Decreasing input voltage does not affect this. Any suggestions on how to fix this would be helpful.

I know that in the schematic has the input coming from a voltage source and not from a mains transformer. This is because I am lazy, not because I know nothing about safety. In an actual built unit, there would be fuses as well.

Simulated, ripple is about 1-2 mV at 2A, I haven't tested further than that. At 1A, ripple is about 400uV.

- keantoken

[janneman]

Hi,

To get a real good feeling for the performance of the circuit, you should also test the ripple as a function of load current frequency. The ripple can be very low at 1kHz, but what at 5kHz, 10kHz, 20kHz?
Also, how does the output voltage change with changing input voltage? And what is the output ripple against the frequency of the input ripple frequency?

Jan Didden

[Mooly]

Hi,
It might be OK in simulation but a few things stand out. Q4 what happens when it conducts- it looks like it can pull unlimited base current through Q6. R2 and Q3 ? not sure what you are trying to achieve there Connecting zeners across potentially "low impedance" points with high current capacity is asking for trouble, any spike will zap it.
Keep working on it, do you intend to build it

Edit, I would be nervous with this -- there is the potential for Q2/3/5/7 to put a virtually dead short over the unreg supply through all those B-E junctions. It would take only a brief "spike" to spell disaster. Q4 and Q6 "appear" to form a thyristor. Once Q4 is triggered into conduction Q6 will sustain it's base current whether or not D5 is passing current.

[keantoken]

If Q4 conducts even a small bit, Q6 attempts to turn it hard on. Ideally, this would only happen if the unit started to oscillate since this would pull a lot on Q5s base and turn the whole thing off. However, I see the risk. On a previous version, this would turn on fully in the right conditions and draw around 2A. Those transistors would be history .

Once Q4 is triggered into conduction Q6 will sustain it's base current whether or not D5 is passing current.

It is a very sensitive configuration but it is possible to use it as a super-sensitive voltage sensor in the right conditions. Here, the regulating circuitry will have to react fast enough to cut off that voltage. It's a very precise feedback loop.

Adding a 100k resistor to Q6s emitter seems to work here without degrading performance, although if the pair still turns hard on I think those transistors might be blown. Hmm. I need a protection measure for that...

My initial idea was that Q2 and Q3 would act as a CCS, except that it also drives Q1s base, giving stable voltage across. Now it looks like Q2 and R3 just act a a current divider. I haven't deviated from this, though, since I've found nothing else that works as well. Believe me, I've tried.

Edit, I would be nervous with this -- there is the potential for Q2/3/5/7 to put a virtually dead short over the unreg supply through all those B-E junctions. It would take only a brief "spike" to spell disaster. Q4 and Q6 "appear" to form a thyristor. Once Q4 is triggered into conduction Q6 will sustain it's base current whether or not D5 is passing current.

Hmm. Looks like we need a fast blow fuse and a sturdy case.

I don't know if thermal switches or fuses would be fast enough, to keep the transistors themselves from acting like fuses.

Any ideas? I think discrete protection circuitry would be best... Hopefully my circuit isn't only good for a toaster. ):

At any rate, here's a AC analysis, using the load as the input, showing the current through the .5 ohm resistor, with the 100k resistor on Q6s base. I found that by using a pass transistor with MUCH less junction capacitance than the 2N3055 (415pF, WHAT!?), this was greatly improved. Unfortunately, that transistor is only rated for 1A, though it will still work in the simulator. Any recommendations for a good pass transistor with low junction capacitance?

- keantoken

[sawreyrw]

keantoken,

I would suggest you run a transient (.tran) analysis of this circuit and look at the currents in the transistors. I could also give you additional feedback, if you like.

Rick

[keantoken]

I remember you.

I'll try not to be such a pain this time. Yes, feedback is fine. If I can't take the heat, that's my problem.

I have been using transient analysis and AC analysis side by side whenever it can help, as much as I know how. I look at the transient at about 10KHz, which is where this PSU operates the worst, and it seems that just plain junction capacitance is what does this. I still can't quite figure out why that bump is always around that area, though. Changing the value of C2 affects ripple greatly. Currently a value of 100p is best, from what it seems. Having C2 at too high a value isn't good, however, because it will trade output stability for input ripple rejection (which isn't what I want).

I've changed the output transistor to a ZTX something, with 7A max current, but the Vceo is 30V and the current Vceo is at max 27V, so I still need a better transistor.

Currently, output changes about 2.2mV from 1A to 2A load.

The image below is an AC analysis with a 1.25-2.25A sine load, probed at the output voltage. The other graph is the load current vs. output voltage at the worst operating frequency, 10KHz.

- keantoken

[keantoken]

AHA!

Deleting the .5 ohm resistor cures it completely! The error was phase related.

When the load current rises, the output voltage lowers. But this is not because the regulation is faulty, because when the load current lowers, the output voltage is steady.

It seems that removing the .5 ohm resistor keeps the junction capacitance from becoming much of a problem.

Now the ripple at 10KHz is about 10mV.

It seems more prone to oscillate now, though, but it looks like 100p is enough to keep it under control.

- keantoken

[Mooly]

Hi,
I will keep looking in . I have the greatest admiration for you guys that design on the simulator ( something I have never got into ) . Ultimately your going to have to actually build it and do some real measurements. Looking at your latest circuit, have you added any rail capacitance to the output and local decoupling for R.F. stability. Does R6 need to be so low ? I can't just grasp the function of Q6. ( I can see what you are trying to do but the whole error amp looks to convoluted ).
Quite a good test (to do for real) is to place a power transistor (HEXFET is ideal) but any high gain high speed power Bjt will do across the output with a resistor that will draw say 2 amps from the supply. Drive the base or gate with a squarewave and look at the actual ripple and rail fluctuations. You can drive it from DC up to say 40 or 50 Khz. Very revealing.

[keantoken]

I only expect to add rail capacitance to the output when I actually try and connect it to something. As for now, I want to get it as good as possible in a linear fashion, with as little as possible AC components.

And what exactly does 'local decoupling' mean? I probably know what it is but the jargon escapes me.

R6 is quite low but I found this was the optimal value by playing around in the sim. Curious that it also matches about the resistance of the 1N34A diode in the same situation...

Documentation on the Q4/Q6 pair can be found here although not in the way I use it. I find it weird that I have not seen a link to this site on DIYAudio once... Maybe everyone else knows something I don't?

Quite a good test (to do for real) is to place a power transistor (HEXFET is ideal) but any high gain high speed power Bjt will do across the output with a resistor that will draw say 2 amps from the supply. Drive the base or gate with a squarewave and look at the actual ripple and rail fluctuations. You can drive it from DC up to say 40 or 50 Khz. Very revealing.

I will remember this one. Sounds like it's liable to make something explode though. I can just hear some cheap electrolytic start fizzing shortly before popping and smelling very very bad. I would actually prefer to be more gentle with my circuits! Something like that really scares me... But I guess I need some more experience to really know what to expect...

- keantoken

[Mooly]

Hi,
"Local Decoupling" The output impedance of any series reg like this will increase with frequency. It just means adding something like a 0.1 mfd across the output to maintain a low impedance at HF. I still can not see what Q6 does other than form a thyristor as I mentioned earlier. As you had it in the earlier circuit it formed a "crowbar" across the rail but now with a 100 k in the emmiter --- what is it's function ? Transistors connected like this used to be available as type BRY39 and BR101 used for triggering of thryistor based PSU in old tellys etc.