DIY Power Supply - JFET voltage problem

[ion]

I've recently completed a DIY LM317 based power supply, mostly based on the attached schematic.  It worked, but not for long - and then there was smoke!

As Q1 and Q2 are obsolete, I substituted them with BF256 JFETs.  JFET current sources were new to me and I focused too much on IDSS at the expense of the other parameters while looking for a substitute.
I failed to realise that on the final tap of the transformer, Vin is around 40V - putting about 50V across Q1.  The BF256 is a 30V part and Q1 is now toast.

I was thinking of putting a suitable resistor after the diodes and potentiometer and before Q1 to drop the excess voltage.
If I understand the circuit correctly this should not have any impact on the current limit adjustment as it is determined by the resistance between the Adjust pin and Vout.

I don't have many JFETs left so I'd appreciate feedback on my solution before I try it, and any unintended consequences it might have.

[Ian.M]

That will FUBAR the limiting.   The purpose of the Q1 JFET current source is to maintain a constant voltage drop across D1 and D2 so the 1K pot adds a predictable voltage to that developed across Rsc, to reach the regulator (1) reference voltage at the desired limiting current. At that point the output voltage will start to drop, and your extra resistor's voltage drop will cause Q1 to run out of headroom, the current through the current source will sharply decline and the current limit will increase to the maximum permitted by Rsc.

The best option is to substitute BJT current sources.  I discussed this at https://www.eevblog.com/forum/beginners/lm317-variable-power-supply-question/msg687356/#msg687356/.

The current sources have to be able to pull the regulators' Adj terminals below ground by Vref to get zero current and zero output voltage.  Just make sure the BJT current sources' base bias voltage is at least 3V below ground so they have some headroom.

[ion]

Thanks, I was afraid it wouldn't be that simple!

I read the post you linked to and simulated a few BJT current sources.  The simplest one I ended up with is a NPN transistor with the collector connected after the diodes, the emmiter to -10V and the base to 0V through a resistor that sets the current.

With a 330k resistor I should get around 8mA using a BC547, and the base would be about 9V below ground so I guess it should work.

[Ian.M]

No that wont work well because the H_FE is not a well controlled parameter - it varies by a large factor from device to device and also with temperature.

Fix the base voltage 3.3V above the -10V rail using a Zener, anode to -10V, and a 2K2 pullup to 0V.  The emitters will be held about 2.7V above the -10V rail, so set the collector currents by choosing appropriate emitter resistors.

Target currents:

A JFET's Idss is not a well defined parameter and its probable that the current would vary over something like a 2:1 ratio between different FETS with the same part number.

e.g:
A 2N3822's Idss is expected to be between 2mA and 10mA
A 2N5640's Idss is expected to be between 5ma and 15mA

==> Aim for 6mA for Q1 and 10mA for Q2.

However your above PSU design is deficient as a LM317 needs up to 10mA quiescent current to regulate so worst case Q2 current must be >10mA, and half the 2N5640 JFETs wont meet this requirement.  ==> increase Q2 current to 12mA.  You would also need to change the regulator (2) Out to Adj resistor to get 10mA load current, and use a voltage adjust pot of half the value.   You can probably get away with leaving things as they are if you carefully select a LM317 for regulator (2)  that doesn't drop out of regulation till well under 5mA load current. (The minimum load current for regulator (1) isn't a problem because it also 'sees' all the quiescent current of regulator (2). )

This gives emitter resistors of 470R for Q1 and 220R for Q2 or 270R if you decide not to mod the rest of the circuit so 10mA is enough.

[not1xor1]

I was thinking of putting a suitable resistor after the diodes and potentiometer and before Q1 to drop the excess voltage.
If I understand the circuit correctly this should not have any impact on the current limit adjustment as it is determined by the resistance between the Adjust pin and Vout.

IMHO you should just set aside diodes and constant current sinks and use LM385-1.2 and resistors.
You will get a much more stable voltage drop and so a more precise current control.

LM385 supports currents varying from 10uA up to 20mA so a 100k resistor (100-500uA) would be fine.

Another LM385 with a 100k resistor would suit perfectly the second regulator to provide adjustment down to 0V.

I forgot the 1k  pot... you need more current through the LM385 and/or an higher value potentiometer...
OK... later I'll post a full schematic

[Ian.M]

Its a nice idea to get rid of the diodes, replacing them with something more stable, but unless you want the current through the reference shunting the pots to vary wildly with the output voltage (or regulator (1) output voltage when current limited), you should keep the current sinks.

[not1xor1]

OK this circuit should work provided that you have typical LM317, i.e. that your ICs work correctly with just 5mA of load.

The second LM317 would sink about 5mA from the first one so the current sink connected to the output of the first one can use just a couple of mAs or something less (i.e. might work with 390/470 ohm too)

[ion]

No that wont work well because the H_FE is not a well controlled parameter - it varies by a large factor from device to device and also with temperature.

I thought I'd avoid that problem by measuring the H_FE of the specific transistor I was using, but the temperature variation was huge - it seems I won't get away with a 2 component solution!

Fix the base voltage 3.3V above the -10V rail using a Zener, anode to -10V, and a 2K2 pullup to 0V.  The emitters will be held about 2.7V above the -10V rail, so set the collector currents by choosing appropriate emitter resistors.

Tried it with a 5.6V zener (I don't have anything lower on hand) and it's working fine on a breadboard.  The base is still around -4V so I guess this will be a good replacement for the JFET.

I have no intention to change Q2 at this point - the voltage regulation seemed fine with no load before Q1 failed.  The pot is actually a 4.7K and 1K in series, so around 5mA minimum load at 30V.  The regulators I'm using state typical minimum load is 3.5mA.  Thanks for pointing that out though - I didn't get to fully test the power supply so it's good to know a possible source of instability.

[Zero999]

The mains disadvantage of connecting linear regulators in series, is the drop-out voltages add together. Each LM317 needs a total of 3V headroom, plus 1.25V is required for the current sense resistor, giving a total of 7.25V voltage drop!

This is why the input voltage is 32V, for a maximum output voltage of 25V. The maximum input voltage is 40V, since that's the rating of the LM317. It's difficult to get 32V to 40V from a transformer, rectifier and smoothing capacitor. You might be able to get away with a 24V transformer, but it's marginal.

Another issue is when the input-output voltage differential is >15V, the LM317's safe operating area protection kicks in, further limiting the current to 500mA.

[Ian.M]

If you are short of Zener's to choose from, not1xor1's proposed current sink (Q1,Q2, R3,R4) above is also good,  Its a bit more stable with changes in collector voltage than the Zener one and needs less headroom.   Q2 never sees more than a couple of volts so doesn't need the Vcbo rating that Q1 does.

+1 to Hero999's comments - cascaded LM317 regulators are a *LOUSY* way to make a 'lab' supply.  The only worse one I've seen was a brute force job: a 2N3055 Vbe multiplier as a shunt regulator with a 2N3055 adjustable current source feeding it!

[ion]

IMHO you should just set aside diodes and constant current sinks and use LM385-1.2 and resistors.
You will get a much more stable voltage drop and so a more precise current control.

Thanks, but I'm going to stick with a replacement current sink.  The circuit is already assembled, and it won't be easy to make significant changes.

[ion]

The mains disadvantage of connecting linear regulators in series, is the drop-out voltages add together. Each LM317 needs a total of 3V headroom, plus 1.25V is required for the current sense resistor, giving a total of 7.25V voltage drop!

This is why the input voltage is 32V, for a maximum output voltage of 25V. The maximum input voltage is 40V, since that's the rating of the LM317. It's difficult to get 32V to 40V from a transformer, rectifier and smoothing capacitor. You might be able to get away with a 24V transformer, but it's marginal.

Another issue is when the input-output voltage differential is >15V, the LM317's safe operating area protection kicks in, further limiting the current to 500mA.

I'm actually using a transformer from a scrap UPS and I do get around 40V from the rectified 30V tap.  I've also added some tap changing circiutry, so the input-output voltage differential is <15V for most of the voltage range.

If you are short of Zener's to choose from, not1xor1's proposed current sink (Q1,Q2, R3,R4) above is also good,  Its a bit more stable with changes in collector voltage than the Zener one and needs less headroom.   Q2 never sees more than a couple of volts so doesn't need the Vcbo rating that Q1 does.

I like the LM385 solution, but I'm going with the Zener one because it will be easier to squeeze in to the board I put together.  And I already have the parts.

+1 to Hero999's comments - cascaded LM317 regulators are a *LOUSY* way to make a 'lab' supply.  The only worse one I've seen was a brute force job: a 2N3055 Vbe multiplier as a shunt regulator with a 2N3055 adjustable current source feeding it!

Oh, I know - this is not intended to replace a proper 'lab' supply.  Just something I'd use in environments where I don't want to risk damaging a proper lab supply.

[Zero999]

The mains disadvantage of connecting linear regulators in series, is the drop-out voltages add together. Each LM317 needs a total of 3V headroom, plus 1.25V is required for the current sense resistor, giving a total of 7.25V voltage drop!

This is why the input voltage is 32V, for a maximum output voltage of 25V. The maximum input voltage is 40V, since that's the rating of the LM317. It's difficult to get 32V to 40V from a transformer, rectifier and smoothing capacitor. You might be able to get away with a 24V transformer, but it's marginal.

Another issue is when the input-output voltage differential is >15V, the LM317's safe operating area protection kicks in, further limiting the current to 500mA.

I'm actually using a transformer from a scrap UPS and I do get around 40V from the rectified 30V tap.  I've also added some tap changing circiutry, so the input-output voltage differential is <15V for most of the voltage range.

That's a marginal design. 30VAC will give over 40VDC out, especially at light loads and when the mains voltage is near the upper tolerance limit. The circuit should be able to tolerate a bit above 40V, since neither LM317 drops the full voltage.

How accurate does the current limit need to be?

[ion]

I've implemented the BJT/Zener current sink and the power supply is operational again 

How accurate does the current limit need to be?

It doesn't need to be too accurate as the current display I'm using has a resolution of 10 mA.  Although it's completely off at the moment - I just noticed the Zeners I'm using on the supporting circuitry are a different voltage from what I ordered.
At least that would explain most of the things that aren't working quite the way I designed them to, but I'll have to put this project aside untill I get replacements.