Transformer voltage for dual power supply (+12V / -12v)

[Ian.M]

Yes, as I thought, the best option is to move the bridge or discrete diodes off-board.   Just link the outer pins of the bridge footprint, each to the nearest middle pin and Sec1 becomes bridge +, and Sec2 becomes bridge -

Using individual Schottkys isn't too horrible if they are isolated tab packages - just bolt them down with a dab of heatsink compound, and wire them up to a nearby tagstrip.  Its probably worth putting Kapton tape on the chassis under the diode legs so they cant short to the chassis if they ever get bent down.

[belzrebuth]

Could you recommend a particular schottky package that would fit my needs?

I've never used those so I'm a bit clueless on what to get.
I've only seen them on SMPS power supplies I think..

[Audioguru]

I haven't looked at the waveform of my mains electricity for many years. I betcha it has its peaks squashed by all the full wave rectifiers and filter capacitors in nearly everything I have turned on most of the time. Then the formula about the peaks are 1.414 times the RMS voltage goes out the window and your LM317 powered from a 12V transformer and rectification will have poor voltage regulation.
Does your mains electricity have a voltage drop each time the refrigerator or air conditioner turns on? Mine doesn't because the transformer feeding mine and a few other homes is next door.

[Audioguru]

Most bridge rectifier packages state Vf of about 1V.

For each diode. A bridge rectifier has two diodes in series so the voltage drop is 2V.

[belzrebuth]

I've just discovered that I also have two identical toroids (also 12V).
These output 13.8V unloaded.

They're 50VA each so I could maybe try those two for a 2x12V substitute.
I think I have other transformers laying around too but they're not toroids..

I really need to get the psu board populated in order to perform some hands-on tests (I could monitor the voltage while turning the air conditioner on for example).
Getting a new transformer is not that expensive but I wouldn't mind trying if what I've already have works..

[Ian.M]

Could you recommend a particular schottky package that would fit my needs?

I've never used those so I'm a bit clueless on what to get.
I've only seen them on SMPS power supplies I think..

This looks good: https://www.tme.eu/gb/details/stps1045fp/tht-schottky-diodes/st-microelectronics/
45V 10A Schottky in a two pin Iso-TO-220 package.

Bend their legs up at right-angles just beyond where they neck down, bolt four down on a smear of heatsink compound using a plain washer + a thin belleville or crinkle washer on top for even clamping pressure (a spring washer is a poor but usable substitute) and wire up as follows:



Use solid wire for the links and sleeve the +DC and -DC ones.  It should be reasonably neat.

[David Hess]

Something like a 1N5822 which is 3 amps would be my first choice.  Use a terminal strip to wire the diodes in the air and mount them to the chassis.

[belzrebuth]

Thanks Ian.Much appreciated

[Ian.M]

I've just discovered that I also have two identical toroids (also 12V).
These output 13.8V unloaded.

They're 50VA each so I could maybe try those two for a 2x12V substitute.
I think I have other transformers laying around too but they're not toroids..

I really need to get the psu board populated in order to perform some hands-on tests (I could monitor the voltage while turning the air conditioner on for example).
Getting a new transformer is not that expensive but I wouldn't mind trying if what I've already have works..

Time to do some load testing.  12V 20W G4 halogen bulbs will do nicely for test loads assuming you are shooting for 1A DC out from the finished PSU.   You can hook them up with croc clip leads and put them on a ceramic tile if you are only doing a short test, but its worth buying some ceramic G4 bulb sockets  if you want to do a longer test or for future use.

[belzrebuth]

First thing tomorrow I'm off to buy those bulbs.
It's Sunday I won't be finding any shops open today..
And I've loaned most of my high wattage resistors to a friend that needed to load-test some PSUs so I only have a 100W 20Ω here

[Ian.M]

N.B. bare halogen bulbs blow if you touch them before use.  They have a quartz glass envelope, and if you leave a finger print on the glass, the oils in your sweat carbonise and cause the glass to overheat + the salts in your sweat fuse with the glass and lower its melting point.   Always handle with a tissue and wipe with methylated spirits or IPA after handling.

[mariush]

Just go to a car or motorbike parts store and you should easily find 10w/20w/40w incandescent light bulbs , and cheap... 

Otherwise, any regular store should have cheap bulbs ... for example my desk lamp uses G4 bulbs , which are 12v 20w : https://www.amazon.com/CBconcept-10XG412V20W-Halogen-20-Watt-12-Volt/dp/7542904760/ref=sr_1_3?ie=UTF8&qid=1535300104&sr=8-3&keywords=g4+bulbs

Just use some paper napkin or something if they're the halogen kind... otherwise the grease can cause problems.

Oh... and it's overkill for this project, but when you really must reduce the voltage drop on diodes an alternative would be to get rid of the diodes completely and use a chip like  Linear (now Analog?)  LT4320: http://www.analog.com/media/en/technical-documentation/data-sheets/4320fb.pdf

Get this chip in a DIP package along with 4 through hole mosfets on a prototyping board and you got yourself a bridge rectifier with maybe 0.1-0.2v voltage drop, depending on your mosfet selection... the only downside is the cost... a few dollars for the chip and less than a dollar for the 4 mosfets.
Or get this in a surface mount package along with 4 surface mount mosfets and you can basically have a whole bridge rectifier in a TO220 style package, as the losses would be very small at 1.5-2A current output.

[floobydust]

Measure your main's voltage, all this math can be wrong.

I did a linear (audio) power supply like this and it worked great in the lab.
At the venue, the line voltage was ~5% lower and the musicians had hum coming out of the power supply, worse when stage lighting was busy. Had to change the transformer from 12.6V to 14V. Schottky rectifiers can give you another volt - all assuming your line voltage is not high. It also changes throughout the day.

The PCB you are showing does not have ideal connections to the transformer center-tap, so expect more ripple on the -ve rail.
To get less than 5mVpp output ripple, you have pay attention to the PCB trace layout.

[belzrebuth]

Okay I load-tested the transfomers, here are the results..

100VA transformer with 20W g4 halogen bulb on one of the secondaries-> 12.6Vac
50VA transformer with   14W g4 halogen bulb on one of the secondaries-> 13.2Vac


The parts for the PSU pcb should be here within 3 days so I will post results then..
I've ordered 4 of the low drop diode packages IanM suggested.
I have a KBU4J here I could also test.
I'll need to get find heatsinks though..

[belzrebuth]

Should I have measured both the secondaries of the 100VA transformer?

I thought about loading each with a 14W bulb.
That's 1.2A at 12V.

I could also use the 20W bulbs,then it's 1.67A for each winding but then I'm too close to the max current rating..
I could do that for a short period though since I will still be within the transformer's tolerance.
If I still get more than 12V I think it's going to be ok if I use the schottky diodes..
I'm going to probably try to turn a heater or an airconditioner on while still monitoring the voltage..

[Ian.M]

For transformers with two secondaries load *BOTH* secondaries!  If you don't, the losses in the primary are far lower, and its impossible to calculate the true equivalent resistance referred to the secondary.

You aren't going to damage the transformers running them at 100% load for 10 minutes in free air.  In fact they'd probably be OK with 150% load (50% overload) for that long starting from cold as there is a fair bit of thermal mass so no part of the winding is likely to exceed its maximum insulation temperature.    Its therefore good practice to aim for 80%-100% of full load when testing.

For the 50VA transformer please use 2x 20W bulbs, one on each secondary, as 20VA@12V is 1.67A RMS which, when you apply a factor of 0.62 for bridge rectification with capacitive input load, corresponds to a DC load current of 1A, which is right where we want to be operating as max load for a LM317/LM337 based PSU.

Normally you'd load the 100VA transformer with 2x 20W bulbs PER SECONDARY, but as you specifically are using regulators with a max rating of 1.5A DC, there is no point in going over 2.4A RMS which is 30W @12V.

Due to the non-linear resistance vs voltage curve of the bulbs, it would also be helpful if you measured the actual load current of one of the bulbs for each transformer.

Also, for our convenience and to guard against mains voltage fluctuations with time of day or week, please repeat the unloaded voltage measurements immediately before and after the loaded measurements and post all the results in the same post.

So far it looks like the 100VA transformer is likely to be the best bet. 

Turning on your heater or air-conditioner etc. is unlikely to have much effect unless your house's wiring is in particularly poor condition.  Problems start when a high proportion of the population turn on a high load appliance at the same time.  The classic case was the U.K's Coronation Street soap opera on ITV, which regularily aired late afternoon.   During its commercial breaks, you could bet that a high proportion of its viewers (over 9 million households in the '60's and '70's) would turn on a 3KW kettle.

https://en.wikipedia.org/wiki/TV_pickup
Dinorweg pumped storage powerstation in the UK was built to time-shift peak loads, but as it has the fastest spin-up time for a major power station in the UK, its currently mostly used to supply the rapid demand surges caused caused by TV pickup. https://en.wikipedia.org/wiki/Dinorwig_Power_Station
(Its pretty incredible - 12 seconds to go from near 0 to 1320 MW)

[belzrebuth]

So to summarize:

For the 100VA transformer I need to attach a 20W bulb per winding and actually log the current draw as well as the voltage.
I'll leave it wired and test at 4-5 different times a day.
Correct?

I didn't quite understand what you've said about the 50VA transformer though..
The 50VA transformer is 1x12V so I only have one secondary.
I have two identical units so essentially they could be used as one 2x12 that's why I mentioned it.

I may need to focus on the 100VA transformer; having two transformers on such a small case might be inconvenient.
For reference this power supply will be mounted on a wooden case used for a modular synthesizer so there's no metal chassis or too much space for huge heatsinks etc..

I also wonder what kind of heatsinks I could use for my LM317/LM337:
Mounting them off the pcb will be difficult so I thought about getting somethink like this:
https://www.tme.eu/gb/details/sk104-38sts/radiators/fischer-elektronik/
Or this:
https://www.tme.eu/gb/details/hs-132-38/radiators/stonecold/

[IanB]

For reference this power supply will be mounted on a wooden case used for a modular synthesizer so there's no metal chassis or too much space for huge heatsinks etc..

This will be awkward since wood is an insulator. If you put a heatsink on a part inside a wooden box it won't be a heatsink any more since the heat will be trapped inside the box and can't escape. You might need to consider drilling vent holes in the box and having a small fan to move air through it and carry the heat away. Heat sinks always need some kind of air flow to work, either natural or forced convection.

[belzrebuth]

Fan is not an option unfortunately.
It's awkward enough as it is.
This a studio instrument so no fans is absolutely essential..

Anyway, I've tested the transformer again with a 20W bulb attached to each of the secondaries.
Current draw is 1.75A and the voltage is 12.5V (per secondary).

I can try it with 2x 20W in series but wouldn't I exceed the rating?
Is the 3.5A rating per winding or for the whole transformer?!?

[Ian.M]

For the 50VA transformer, 1x 20W bulb would be an appropriate test load for its single winding.  I expect the result will be somewhere around 12.9V loaded.

100VA/(2*12V)=4.17A so 3.5A per winding would be a pretty conservative rating, (or a 100% rating if its actually an 85VA transformer)
2x 20W in PARALLEL per winding (4 bulbs total) would put it right at or possibly slightly over its max rated current.  Not a problem for 10 minutes per hour, but not something you'd want to do 24/7.  However you only need to test it to about 50VA which you've already done.

For 1A out per rail, assuming your 100VA transformer, neglecting unloaded losses in the primary and core, input power is 2*12.9V*1A/.62=41.6W.   Output power is 2*12V*1A=24W.   The difference of 17.6W must be got rid of.   

That's enough to make a medium sized closed wooden box pretty toasty after a few hours running, and it gets a lot worse if a significant portion of that 24W is also used inside the same box.   I'm with IanB - its either going to need a fan or an externally mounted heatsink.  As you are dead set against a fan, the only real option is a rear mounted external heatsink with its fins vertical for better convection.

That PCB layout is very stupid from a heatsinking point of view.  The regulators and the bridge rectifier should have been mounted along one board edge so they could bolt to a vertical heatsink.   The best option, if its got plated through holes would be to mount the regulators under the board with their legs bent up.  However that doesn't mesh well with your possible heatsinking options, so you may have to bite the bullet and mount the regulators off-board, with 0.1uF to ground from input and output right at the regulator so it doesn't break into RF oscillation due to wiring inductance.

N.B. LM317 tab is Vout and LM337 tab is Vin so they both need isolated mounting kits if you want to use a common heatsink.   A minimum thickness mica washer, thermal grease, and an insulating bush on the screw is the best option.

[belzrebuth]

I've built the power supply with a standard bridge rectifier (although I've bought the diodes mentioned by Ian.M).
I can draw about 0.6A without any voltage drop but at after 1.0A I get a voltage drop of about 0.2 to 0.5V which may or may not matter for the modules I'm going to power.
I better be on the safe side and have 12,000 at all times if possible.

I was thinking instead of replacing the diodes and mess with the outboard wiring and heatsinking couldn't I just find a pin to pin replacement LDO regulator for the LM317/LM337?
I need low-noise low need for regulation voltage (about half of the that of the LM317 would be nice) and about the same or more max current rating..
Is there something that fits these specs?
I'm not particularly concerned about the cost since it's only two pieces I'm interested in.

[Ian.M]

Have you scoped the output rails under load?  Your 0.2V to 0.5V voltage drop may actually be a ripple breakthrough of double that!  If you don't have a scope, put a 1uF non-electrolytic cap (or nearby value) in series with your DMM on AC V and multiply the result by 2.8 to approximate the peak to peak ripple voltage.

I was thinking instead of replacing the diodes and mess with the outboard wiring and heatsinking couldn't I just find a pin to pin replacement LDO regulator for the LM317/LM337?

You could, but you'll have a hard time finding one that will 'drop in'.  Even if the pinout is correct, LDOs typically have far tighter constraints on input and output capacitance and especially the ESR range required, than LM3x7 regulators and if you get it wrong there's a risk of them breaking into oscillation, which I've seen bad enough cases of in the HF band to disturb other electronics over 30m away. 

Depending on the load you are powering, it may be acceptable to tweak the Ajd pin to Gnd resistor values down a bit to drop the regulated voltages to a symmetrical 11.5V to give the regulators more headroom.  That's certainly better than letting them droop on their own as it avoids ripple breakthrough.

The diodes I proposed wont get as hot as the existing bridge rectifier.  Bolt them down on a  sheet of Aluminum that's big enough for a 1" margin round them and they'll do fine.

Another option if its an open center toroidial transformer would be to add two overwinds to put in series with the existing secondary windings to boost the voltage.   Start by putting 10 turns of hookup wire round its core, through the center.  Measure the AC voltage across the hookup wire ends, then calculate how many turns you need to get the voltage headroom you need.  Post a photo and we can suggest how to wind it.

[David Hess]

I was thinking instead of replacing the diodes and mess with the outboard wiring and heatsinking couldn't I just find a pin to pin replacement LDO regulator for the LM317/LM337?
I need low-noise low need for regulation voltage (about half of the that of the LM317 would be nice) and about the same or more max current rating..
Is there something that fits these specs?
I'm not particularly concerned about the cost since it's only two pieces I'm interested in.

The LT1086/LM1086/LD1086 is a low dropout improved 1.5 amp replacement for the LM317.

The LT1085/LM1085/LD1085 is a low dropout improved 3 amp replacement for the LM317.

The LT1033 is a low dropout improved 3 amp replacement for the LM337.

[floobydust]

I found no benefits to increasing the filter capacitors past 10,000uF because the noise of the LM317/337 dominates. Some manufacturers LM317's have a lot of popcorn noise. LT I found the best and Fairchild the worst for noise performance.

There is a performance limit with these convenient 3-pin parts. To do much better, you need more than 3-pins in an IC Vreg package, in order to connect a reference filter capacitor.

Adding huge output capacitors 1,000uF-4,700uF to the LM317's is what I see commonly done, as a compromise.

[David Hess]

There is a performance limit with these convenient 3-pin parts. To do much better, you need more than 3-pins in an IC Vreg package, in order to connect a reference filter capacitor.

Bypassing the lower resistor in the divider with a capacitor goes a long way to reducing the output noise but I agree; better performance requires a different part or circuit.