I need to build a linear voltage regulator like this:

enter image description here

The problem with the usual linear regulators (LM317, LM337, ..) is the limited voltages they can withstand.

I have taken inspiration from a Texas Instruments article: http://www.ti.com/lit/an/snva583/snva583.pdf

The load on the ±15V rails will be around 20 ~ 50 mA, certainly no more than 100 mA.


1.  Positive linear regulator

Please have a look at the circuit below. All protection diodes are colored grey to distinguish them from the rest of the circuit.

enter image description here

1.1  How the circuit works

From my understanding, the darlington pair is connected as an emitter follower: voltage V2 (the emitter) follows voltage V1 (the base). Why? Imagine V1 > V2, then a small current flows through the base of T1, activating the darlington. The darlington conducts until V1V2. Of course, V1 will always be slightly higher because of the diode drop over the darlington.

V1 is 6.2V above the output of the LM317, which is +15V. So V1 is around 21.2V. V2 will be around 20V.


1.2  Protection diodes

I've installed plenty of protection diodes:

  • D2, D4:  If The input of the LM317 ever goes below the output voltage (for example when the power source at the input is taken away), then these diodes prevent the LM317 from being exposed to a reversed voltage.

  • D3:  The LM317 input must never exceed the output by more than 40V (or more than 60V for HV parts). Zener diode D3 prevents that from happening.

  • D5:  If the LM317 output gets shorted to Gnd, the stored energy in C1 causes a reversed voltage between the OUT and the ADJ pins. Diode D5 makes sure the capacitor can discharge.

  • D6:  More than 15V on the output can damage parts that rely on a stable 15V power rail. Zener diode D6 should protect them.


1.3  Components selection

The components I chose:

D1:          PDZV6.2A  (Digikey: PDZVTR6.2ACT-ND)
D2, D4, D5:  1N4004    (Digikey: 1N4004DICT-ND)
D3:          PDZV33B   (Digikey: PDZVTR33BCT-ND)
D6:          PDZV16B   (Digikey: PDZVTR16BCT-ND)
LM317:       LM317HVT  (Digikey: LM317HVT/NOPB-ND)
T1, T2:      MJD47     (Digikey: 497-2482-1-ND)

Note 1: I realize that the zeners cannot dissipate enough power. Sometimes I place zeners in parallel, each of them in series with a 50Ω resistor. Those series resistors make sure that the current distributes itself somewhat evenly over the zeners.
Note 2: The MJD47 transistor is electrically equivalent to the TIP47


1.4  My questions about this circuit

I have a few questions concerning this circuit:

  1. The Texas instruments article puts a 100Ω resistor in series with capacitance C1. Why? Do I need to do that also?  

  2. The Texas instruments article puts a 2.7Ω resistor in series with the output capacitance C2. I believe that one is intended to damp inrush currents. I've used LM317 parts before, and they always worked without such damping. So how important can that resistor be? 

  3. The article says:

    The zener impedance is low enough that no bypass capacitor is required directly at the LM317 input. (In fact, no capacitor should be used here if the circuit is to survive an output short!)

    So a bypass capacitor (like C3, drawn with dotted lines) can keep the input voltage on the LM317 fairly high while the output gets shorted, thereby destroying the chip. Would my protection diode D3 be sufficient to protect the chip from such calamity?


2.  Negative linear regulator

I've designed the mirror circuit to create -15V:

enter image description here

All components are the same, except the transistors (they are PNP instead of NPN):

T1, T2:      MJD350T4    (Digikey:  497-3137-1-ND)

So here comes my last question: Is this mirror-circuit okay?


3.  Notes

3.1 DC-DC-converter

Several comments mention the use of a DC-DC-converter. It would take too long to explain in detail why this is not an option. Let's just focus on the linear voltage regulator. Discussing the pros and contras of DC-DC-converters, and their usage (or non-usage) in noise-sensitive circuitry, can be a great topic for another question ;-).


3.2 Battery choice

I know it looks a bit silly: using a 200V battery to create ±15V. But this has to do with the application at hand. Again, it would take too long to explain why there's no other option (like putting another battery, ...). For now, let's just consider this huge voltage drop as a challenge ^_^.


3.3 "Harmful derivation of the pre-regulator reference"

I read in the answer of Mr. Spehro Pefhany:

Aside from the pointless (and possibly harmful) derivation of the pre-regulator reference from the regulated output this is just a Darlington/Zener pre-regulator followed by a three terminal regulator.

I'm trying to understand his point. Reading his answer further, I learn that my circuit is not putting enough load on the LM317. Thank you for pointing that out. If I put enough load, does this make the circuit okay? Is the "derivation of the pre-regulator reference from the regulated output" still wrong? Please enlighten me to understand.

Note: I've taken this circuit from a Texas Instruments article (http://www.ti.com/lit/an/snva583/snva583.pdf). It's unlikely that these folks design a circuit with serious flaws. Nevertheless, I don't want to rule out the possibility completely.


Big thanks to all of you for the efforts!

  • \$\begingroup\$ You haven't mentioned load regulation expectation. If you expect something below what LM317 can do, Zener diode + R setup would be suffice. \$\endgroup\$
    – Long Pham
    Commented Aug 25, 2018 at 14:49
  • \$\begingroup\$ Zener diode + NPN follower is not good enough? \$\endgroup\$
    – winny
    Commented Aug 25, 2018 at 15:10
  • 1
    \$\begingroup\$ A linear regulator would be dropping 170V. At your worst-case 100mA, that's 17W of dissipated heat in the regulator. Even at your best-case 20mA, that's still about 3.5W. A buck converter might be more appropriate. \$\endgroup\$
    – marcelm
    Commented Aug 25, 2018 at 15:33
  • \$\begingroup\$ Buck converter is not an option. The DC output must be almost perfect (no noise). The DC input comes directly from a battery without any electronics in between. \$\endgroup\$
    – K.Mulier
    Commented Aug 25, 2018 at 16:16
  • 1
    \$\begingroup\$ Then choose a more suitable battery. If this battery exists to power something else, you have to consider what noise that might inject. \$\endgroup\$ Commented Aug 25, 2018 at 16:37

1 Answer 1


Aside from the pointless (and possibly harmful) derivation of the pre-regulator reference from the regulated output this is just a Darlington/Zener pre-regulator followed by a three terminal regulator.

By possible harmful- there is insufficient load on the LM317 for low regulator load conditions- there should be at least 5-6mA minimum (typically a 220 or 240 ohm resistor is used where you have 910 ohms) and the Zener current (minus base current) through R1 subtracts from that- you can work it out, somewhere between 0.5 and 1mA reduction. Otherwise the output goes out of regulation- perhaps the reason there's a zener on the output. Without the Zener on the output, the output voltage would rise, causing the input voltage to rise.

It's really wasteful to use a linear regulator like this - a switching regulator would be much more efficient, but if you really need to avoid the switching noise, complementary pre-regulators (zener, resistor, darlington with a protection diode) followed by three-terminal (or other low-voltage) regulators would make as much sense.

Use perhaps a 24V zener and you can get complementary darlingtons in one package with appropriate ratings.

If you are very concerned about noise, there are better choices than the LM317/337. As well as noise, just the 5-6mA Iq results in 0.6W of dissipation before adding any load at all.

  • \$\begingroup\$ I understand your remarks about the harm caused by insufficient load on the LM317 (thank you for pointing that out). But if I provide enough load, why would "the derivation of the pre-regulator reference from the regulated output" be "harmful" or "pointless"? \$\endgroup\$
    – K.Mulier
    Commented Aug 26, 2018 at 14:42
  • \$\begingroup\$ @K.Mulier It’s harmful if it reduces the load too much. It’s pointless otherwise in that there is little advantage for fixed output voltage-the Darkington gain is high enough that little Zener current is required. Where it could be useful is if the output voltage was adjustable, then the preregulator voltage would “track”, but that’s not the case here. \$\endgroup\$ Commented Aug 26, 2018 at 15:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.