LM317 output voltage drop out of spec (maybe)

[francesco_97]

Obviously, it is another post in the venerable LM317. I am reviewing a stabilizing module that I've done in the past for my ULTRA WOW LOW RIPPLE DUPER SUPER STABILIZED BENCH PSU but the lack of experience teaches me that a fixed voltage can't be kept fixed under load how I dream.

As you can see from the attachments the stabilization module is a dual-rail version of the scheme in Fig.9 at pag. 11 of the LM317 datasheet BUT with the pot replaced by a set of fixed resistors, because I wanted the PSU with fixed voltages selected by a rotary switch, with a 120R resistor to keep more stable the IC and with a ripple filter at the beginning that I copied from the Dave's video. Under no load and at ambient temperature everything works fine.

Now let me show the issue.
From the Electrical Characteristics table at pag. 6 we can read that under V_0 > 5V the max load regulation is 1.5%V_0 (supposing at high temperature) and that the Output-voltage
temperature stability is 0.7%V_0 so we SHOULD (and here could be the issue) sum them to obtain a maximum voltage drop of (1.5+0.7)%V_0 so, using a 21V input voltage and an open circuit output voltage of 9V, the drop should be at maximum 9V*2.2/100 = 0.198V leaving 8.8V clean at the output.

This is not the case.
As two 17W 4.7R resistors put in series are connected, the clean 9.00V drops to 8.76V at 38° 0.95A and with a stabilized temp of 80° the voltage reaches 8.43V at 0.90A. It is too low for the specs.
Probably I didn't read the right way datasheet and maybe summing everything that could be the thermal error like the base 1.5% load regulation + 0.7% temperature stability  + 10W*0.07%/W = 0.7% Thermal regulation + I don't know what else to sum we can obtain a 2.9%V_o drop = 0.261V that corresponds to an output of 8.74V. However, I'm too out.

I tried to change the output capacitor from 100u to 1u and to avoid the fixed resistors using an external pot to manually regulate the open voltage but nothing.
What I'm not considering?
You can find the schemes attached below.

Please don't hit me too hard for my mistakes.
Thank you.

[wraper]

Now let me show the issue.
From the Electrical Characteristics table at pag. 6 we can read that under V_0 > 5V the max load regulation is 1.5%V_0 (supposing at high temperature) and that the Output-voltage
temperature stability is 0.7%V_0 so we SHOULD (and here could be the issue) sum them to obtain a maximum voltage drop of (1.5+0.7)%V_0 so, using a 21V input voltage and an open circuit output voltage of 9V, the drop should be at maximum 9V*2.2/100 = 0.198V leaving 8.8V clean at the output.

Didn't bother figuring out from where you pulled that strange math but voltage drop does not depend on output voltage whatsoever. In fact LM317 always "sees" 1.25V between its output pin and ADJ pin, so it doesn't even "know" what actual output voltage is. As of dropout voltage, it's <2.5V (for ST part) and depends on current and die temperature.
EDIT: answered what wasn't asked.

[tooki]

Wraper, I don’t think the question is intended to have anything to do with dropout voltage. The title tripped me up, too! I think the title is meant to read as “the output voltage goes out of spec.” I was also looking at the dropout specs until I realized OP has 21V going in and 9V going out, so more than enough dropout for any situation.

As I understand it, the question is why the output voltage is plummeting to 8.43V under a 0.9A load.

Edit: and I see you also realized this after the fact, just like me!

[ArdWar]

Where do you measure the voltage?

With how the circuit are laid out. Output power trace going a long way to the connectors while sense trace going exactly the other way in what could be described as anti-Kelvin connection it bounds to cause bad load regulation.
From eyeballing alone and assuming 1 oz PCB you may have about 40 milliohm separating your output connector pins from the sense pins, that alone will cause additional 40mV load regulation drop at 1A load. Strangely matching your missing voltage... (no it isn't caused by that alone)

[francesco_97]

Yes, sorry, maybe the world "dropout" is not the best choice but i meant drop "space space pause space" out of voltage 

[tooki]

Francesco, have you monitored what is happening to your input voltage under load?

Where are you measuring the output voltage? If your load is at the end of long, thin wires and you measure at the load, you could be seeing simple voltage drop across the wires.

Have you put an oscilloscope on the output to see if it’s actually stable?

The really bad capacitor component placement is creating big loops, exactly as the datasheet warns not to do. It is conceivable that the large loop areas are causing oscillations.

[AnalogTodd]

Two things I would look at right away:

1. Check the signals with a scope. Make sure that the output is not oscillating and that the input stays at 21V while you are loading the regulator. More than once I have seen people fail to check for oscillations and they find that the output compensation cap just isn't keeping things stable. That, or they find that their input supply can't handle the load they thought.
2. Check voltages around the board, not just at the output pins. Your grounding isn't that great and you may have too narrow of traces on inputs and outputs such that you are getting voltage drops due to trace resistance that are throwing everything off.

[TimFox]

Agreeing about using a 'scope.
For actual dropout voltage on linear regulators, it is important to verify that any ripple on the input does not dip below the dropout point:  you can't tell that with a DC voltmeter reading only (which will read mean voltage).

[wraper]

Wraper, I don’t think the question is intended to have anything to do with dropout voltage. The title tripped me up, too! I think the title is meant to read as “the output voltage goes out of spec.” I was also looking at the dropout specs until I realized OP has 21V going in and 9V going out, so more than enough dropout for any situation.

As I understand it, the question is why the output voltage is plummeting to 8.43V under a 0.9A load.

You're right.
Minimum load requirement is barely met by current through R3/R4, so it shouldn't be the problem with no load.
However PCB layout is an utter shit to put it mildly and you want C3/C4 way more than 0.1uf for good noise rejection and LM317 stability. It would be fine if they were just for local decoupling (that isn't local at all with this layout) but not like this where they're the only capacitance. IMHO this is either oscillating or there are some huge voltage drops along this PCB layout.

[tooki]

It would be fine if they were just for local decoupling (that isn't local at all with this layout) but not like this where they're the only capacitance.

Aren’t Q1/R1/C2 a capacitance multiplier providing the bulk capacitance?

[francesco_97]

Thanks you all for the answers. Now I'm out but tomorrow i'll check with the scope.
Just to fill the questions i used 2.5mm wires at the end to avoid strage resistances due to wiring and i made the "signals" traces with a 1mm width and the power traces with a 1.5mm width.
The 4.7R resistors are connected twisting their legs together and inserting the others directly into the connector.

Unfortunatly i couldn't place the caps in the right spots far away fron the heatsinks due to limited space and the needing to use a single side pcb but tomorrow i'll try with the scope.

[enut11]

Could the LM317 chips themselves be out of spec? I bought some on eBay that looked like the real-deal but were only able to regulate to 20% of spec! Price was good :-). I now use them in less demanding applications. Buying components on the internet can be risky...

[wraper]

Thanks you all for the answers. Now I'm out but tomorrow i'll check with the scope.
Just to fill the questions i used 2.5mm wires at the end to avoid strage resistances due to wiring and i made the "signals" traces with a 1mm width and the power traces with a 1.5mm width.

If it's 1oz copper, trace resistance from vreg/GND to the output connector should be enough to be responsible for at least 1/4-1/3 of this voltage sag. And the rest probably comes from a poorly laid out resistor divider GND or/and oscillation. Not to say output voltage must be sensed by divider at the connector to reduce voltage sag (or even closer to the load if possible). This PCB layout is so bad it gets me vibes that it should be burnt in exorcism ritual to never come back to life.

because I wanted the PSU with fixed voltages selected by a rotary switch

Solder all those resistors on rotary switch terminals as it's done by sane people and run just 2 wires to PCB.

[francesco_97]

Ok, I have news for you.

I forgot the ACTUAL power supply. When the board is in the PSU it gets the power from the line with the 50 HZ ripple filtered by the capacitance multiplier but NOW i'm testing it with my switching PSU. Connecting the scope with the 20MHz filter engaged at the output with a 10x probe and measuring the voltage at the output AND at the input of the LM317 directly (I'm getting the input at the leg of the diode D1) I've got this:

At 30° we have 17.82V at input, 8.73V at output and something like 0.9V pkpk ripple 
At 80° we have 17.60V at input, 8.48V at output and the same ripple

Maybe measuring it with the probe and the ground flying wire is not the best idea as Dave says but I don't have any idea how to get a better result with my setup. Could the issue be the switching power supply? Maybe using the 50Hz power the issue disappears? Or maybe is the LM317 that is cooked?
You can find the screenshots down below.

[ArdWar]

I don't have any idea how to get a better result with my setup.

Fix your layout.

Could the issue be the switching power supply?

Not really, LM317 are capable of better PSRR. Switchers are indeed capable of making bad design worse however.

Maybe using the 50Hz power the issue disappears?

Reduced ripple for sure, but I doubt it'll improve any load regulation especially dynamic one.

Or maybe is the LM317 that is cooked?

Maybe, who knows 🤷‍♂️. Tho 317 are usually robust enough that it rarely goes bad this way.

[wraper]

Or maybe is the LM317 that is cooked?

Or maybe schematic and especially PCB layout are crap. It's clearly oscillating and you cannot blame LM317 for that.

[wraper]

This and make a sensible PCB layout with short traces and not running them all across the board.

[francesco_97]

Ok ok don't get angry, I'm not an engineer so I don't know how to design the PCB but I'll try to fix this and if it is needed I'm going to remake the board. If i have understood well the main issue should be the placement of the input capacitor and the loop area. From the datasheet we can read at pag. 18 that "The optimum placement is closest to the input terminal of the device and the system GND. Take care to minimize the loop area formed by the bypass-capacitor connection, the input terminal, and the system GND." Now they are placed 15mm from each other and the leg of the cap is directly on ground.
Moreover, they are placed in a straight line so the surrounded area should be equal to nearly zero. Maybe the "loop area" cited in the paper includes also the output terminal? Could you draw on the PCB picture this area so I can understand better this concept?