Filtering when designing a 4A linear power supply

[jtruc34]

Hi,

I'm designing a 4A linear power supply for a modular synthesizer. (This may seems a lot of current but modular synthezisers may have 20 modules at the same time.) It's 12V dual rail, using an LM317 and 337 (I think everyone goes by this route at one point in their life). There are two points of difficulty I'm facing.

The first is I'm not sure how I should specify the allowable ripple. Modular syntheziser use a common +/- 12V rail and almost all signals are large signal. So ripple shouldn't be to much of a concern, at least I first thought. However, I'm worried, as their can be so much unshielded signal wires in a modular synthesizer that the ripple would get coupled somewhere. In particular there are control voltage following the standard of 1V per octave, and I've experimentally found out that my audible threshold would be about 2mVpp of ripple directly in the signal. However, I don't know how that would translate in the allowable ripple in the supply rail. There will also be a single small signal output for headphones (about 500mVrms), on which I estimated 50μVrms of ripple would be inaudible, which is much lower than the previous figure. However, I could specifically filter the supply of the headphone module, but again I'm not sure whether the unshielded 12V unshielded bus running everywhere in the synthesizer wouldn't couple by some way or on another.

My second problem is how to choose the filter topology. I know this can be a complex topic, and I tried to apply my knowledge about linear systems, however, because of the bridge rectifier, this is a highly non linear system and this doesn't help much. The first obvious choice is a single capacitor (or capacitor bank), but I would need between 10 and 100mF, which I guess is fairly common for a power supply. But then, I wondered if I could make things better by using a pi filter, and that maybe I could use lower values of capacitors with the same ripple. However, I found it hard to design, and even simulations with constant current were not very intuitive, where sometimes increasing the inductance would have almost no effect on the ripple. What are the pros and cons of those topologies in my situation? Could I go even higher order? I'm aware I could have resonance problems, and I can't go higher than 100mOhms of series resistance without having too much voltage drop.

I'm not very concerned about the price of an inductor since I don't plan on building ten thousand power supplies.

Thank you for your help

[Benta]

Normal term is "filtering capacitors", but that's misleading.
The correct term is "storage capacitors", as they need to keep the DC voltage up during the 100 Hz voltage dips after the bridge rectifier.
Rule-of-thumb is 2000 uF/A, but this can vary.
So pi filter etc. won't really help unless the inductors are enormous, plus they will slow down the transient response of your PSU.

I assume you're aware that LM317/337 types wil not deliver 4 A.

Concerning ripple: I see very little risk there. The linear integrated regulators are very good at ripple suppression; on top of that modern audio opamps have excellent PSRR.

I'd focus on potential signal ground loops, those can give you a lot of hum and can be hard to diagnose and fix.

[floobydust]

Better recording consoles settled into using local voltage regulators fed by a higher regulated voltage i.e. I recall Neve used +/-19V main rails feeding 7815/7915 on critical strips but I could be wrong. Look at recording console designs.
I tried using just a pair of master regulators but wiring/connector resistances came into play, as well as single-point (star) connections where required. You can do a bad PC board design and the high ripple currents from say 10-100mF filter caps creates ripple voltage due to trace/wiring resistances that the Vregs can amplify. I'm saying what you think is a good power supply design can actually perform terrible.
You also don't want coupling between cards of audio frequency ripple, cross-talk. I think is more a concern than mains ripple.

I found grounding is of more importance because even 5mV of hum/ripple (audible and easy to get at 1-4A) can get into many cards. You need a signal and power ground. Even then, RF shows up on the mixing bus node and is another challenge where single-ended audio runs between cards exist.

Using LM317/LM337 powering a 24-track board near 1A +/-15V, I also found the recording engineer complaining about noise, I ended up adding huge 10,000uF caps on the regulator's outputs to filter that I think it was popcorn noise from the regulators. Then I got about 2mVpk and it worked fine.

[Benta]

All very good points from @floobydust, and the one about using POL (point-of-load) regulators is the best. Have local regulation on each module, and you're pretty much out of the woods.

[dobsonr741]

Seem counterintuitive first, but modular folks like a particular Meanwell switching supply: https://modularsynthlab.com/product/eurorack-power-supply-mean-well-rt65b

The key is per module filters, proper ground and POL as already pointed out.

[srb1954]

Better recording consoles settled into using local voltage regulators fed by a higher regulated voltage i.e. I recall Neve used +/-19V main rails feeding 7815/7915 on critical strips but I could be wrong. Look at recording console designs.

19V sounds about right for the DC input voltage for a 7815.

These are specified for RR with a 3.5V I/O differential and that RR gets significantly worse if the regulator headroom drops below that level. So, allowing for a little bit of ripple on the incoming supply, a minimum DC input voltage of 19V would required to comply with the manufacturer's recommended operating conditions. A little bit extra headroom on the incoming supply could possibly improve the RR even further.

It should be noted that with the 78xx series the RR gets worse with the higher voltages as there is progressively less loop gain within the feedback loop to counter noise introduced on the supply rails.

[pardo-bsso]

Our studio console has every module fitted with a pair of 78 and 79 regulators fed from about ~16 volts and an extra wire and connector with the audio ground only.

[jonpaul]

Euro Rack system is notorious for the poor power supply and grounds, creating noise and interference.

Avoid SMPS, a normal linear regulator with electrolytic caps,and HF bypass should I fine, if,good grounds and wiring practice is used

Jon

[jtruc34]

Rule-of-thumb is 2000 uF/A, but this can vary.

This seems extremely low. This would mean about 5V of ripple at 100Hz. I know linear regulators can reject ripple but that seems way too much.

Better recording consoles settled into using local voltage regulators fed by a higher regulated voltage

All very good points from @floobydust, and the one about using POL (point-of-load) regulators is the best. Have local regulation on each module, and you're pretty much out of the woods.

So, in practice, I would have a global 15 or 16V regulator and many local regulators, each one with bypass caps?

The key is per module filters

By filter, you mean simple bypass capacitors, or do you mean something more involved?

I'd focus on potential signal ground loops, those can give you a lot of hum and can be hard to diagnose and fix.

good grounds and wiring practice is used

The key is [...] proper ground

Do you have any specific advice on how to have a proper ground? What should I do now to avoid ground loops later?

I found grounding is of more importance because even 5mV of hum/ripple (audible and easy to get at 1-4A) can get into many cards. You need a signal and power ground.

So, you would mean I would have two seperate bus tied down together at the PSU, one used for power and the other for signal? I don't exactly see how I would use that at the board level. Taking a simple example of an oscillator: all op-amps would be power by the +/- 12V. Where would I use the power ground? I'm not sure to see where I would use the power ground and where the audio ground.



Thank you all very much for your feedback!

[dobsonr741]

One common fault I found in my euro rack: the alu face plates are in contact with the signal in/out 3.5 mm jack’s GND. No noise if I keep the module “in the air”. Once screwed into the rack aweful noise appears. I’m planning to add input ground lift and differential output to the modules I’m about to build, so I control the GND reference point of the signal.

Unfortunately not completely compatible with the basic euro rack design ideas, however a mixer output stage can reasonably separate the noise from the output of the whole rack.

[Benta]

Rule-of-thumb is 2000 uF/A, but this can vary.

This seems extremely low. This would mean about 5V of ripple at 100Hz. I know linear regulators can reject ripple but that seems way too much.

Actually. it's more like 4 V, but never mind.

To cross the t's and dot the i's here, all suggestions go in the direction of a main pre-regulator capable of handling the 4...5 A for the whole rack, and having POL regulators on each module.
The pre-regulator could even be a super-simple BJT/Zener combo circuit.
Throwing big capacitors at it is a waste of space and money IMO.

[mariush]

If you use a classic transformer, you can't depend on its output voltage to be a specific value.

At very low output, the voltage may be up to 10-20% higher than the advertised value. So if your product is on stand-by or most modules are idle, the transformer may output a much higher voltage.
Then, you have to account for much lower mains (AC) voltage - ex your 110v AC may be 100v AC, or may be 120v AC ... if the transformer ratio is 1:9  (ex 110v : 12v ac) then you could have almost 13.5v AC with 120v AC on the input, before we even get to that 10-20% extra at low current.

You rectify that AC voltage with a bridge rectifier and then smooth out that rectified voltage with capacitors.
You can sort of approximate how much capacitance you want to guarantee a minimum dc voltage with the formula 
Capacitance = Current /  [ 2 x ac frequency x ( Vdc peak - Vdc desired) ] 

So for example, let's say you have a 110v AC - 15v AC transformer, and you know you may have 14v AC if mains voltage is too low. At idle, and/or high mains AC voltage, you may have 16v AC or even more
Rectified DC voltage will have a peak voltage of Vdc peak = [14 .. 16] x 1.414 - 2 x voltage drop on diode in bridge rectifier (around 0.7v..1v per diode)  = 18.2v .. 21v 

So you'd want to use at least 25v rated capacitors, but 35v or 50v rated would be better.

So going back to the formula , you have Vdc peak of 18v , you have 4A current, let's say you want minimum 14v DC all the time so you could then use linear regulators to produce 12v....
C = 4A / ( 2 x 50hz  x (18v - 14v ) )  = 4 / 400 = 1 /100  = 10000uF

I'd probably consider 3 x 3300uF  or 2 x 4700-5600 uF , depending on how much space you can afford for them.

You may want to think how much current you're gonna consume on negative voltages  and how much on positive voltages

It may be easier and cheaper to not have 2 identical secondary windings to have same current on positive and negative, maybe you could have custom transformer with less current on negative side. Or may even be cheaper to get just one secondary winding, and then use  a switch mode regulator to produce some negative voltage, followed by a linear regulator like 337 to smooth that out even further.

As for regulators, like others said ... you'll probably want to have multiple regulators, maybe let's say 5 of them, 1 for every 4 modules / slots.
Regulators like LM317, 7812 , 7815 can typically do 1 - 1.5A of current, but there are other regulators like let's say LM1085  and *1085 (made by multiple companies) which can do 3A

Even so, keep in mind linear regulators dissipate the difference as heat, so for example if you do 18v input, 15v output at 1A , that's (18-15)x1 = 3 watts of dissipated heat  ... better to have a nice wide heatsink and attach multiple regulators on that heatsink to spread the heat instead of having one super hot regulator.

There's also some linear regulators that are very good at filtering, rejection, maybe you'll find it better to go overboard and use a single 500mA... <1A linear regulator for each module, because such ICs have much better filtering and stuff.

[Vovk_Z]

This seems extremely low. This would mean about 5V of ripple at 100Hz. I know linear regulators can reject ripple but that seems way too much.

2000 uF/A is ok for a typical power amplifier. In your case you'll have a linear regulator after that, so you'll have 60-80 dB rejection (if you do it right).
If 1.0-1.5 A is sufficient, then I would use LM317 instead of LM78xx, because LM317 may deliver higher rejection.
LM338 works with 4 A fine and may give you somewhere near 0.1-0.2 mV RMS noise.
It is easier and give much less space for mistakes if you'll do +- power channels as identical, with two isolated secondary windings.

[jtruc34]

Well, actually I've realized that I have much less headroom than anticipated. The transformer I have (and I can't justify buying a second one) is 12V, which means 17Vpp in normal conditions, but knowing mains can get as low as 210V and factoring the diode drops (480mV for the one I use), I only get 14.3V, which is way too low to have two LM338/317 in series, and with 1V of ripple it's only 13.3V.

I'll have to opt for LDOs. Would you still recommend to regulate twice? Or maybe I could do something else at the PSU and only regulate at the point of load.

Thanks

[mariush]

480mV sounds like schottky diodes... those aren't ideal to use to rectify AC voltage, especially in audio applications. Or maybe check the datasheet and see the voltage drop at higher currents, 480mV may be at 1A, not 4A. Also, you'd probably want a 10A or higher bridge rectifier, something much higher rated than 4A or whatever you think your device is gonna use.
May even make more sense to trade low voltage drop for a package with bigger voltage drop, but which can be more easily attached to a heatsink, for example one of the GB* series which can be easily screwed down to a heatsink : https://www.digikey.com/short/j89wbrr3

There are alternatives to using diodes to rectify AC voltage, but most are quite expensive or have some gotchas (for example the LT4320 in the video below is great if your product uses a lot of current, but it doesn't actually work while the ac output is below around 9v AC, so it's not great for low ac voltage output, and there's gonna be a very small period every cycle where there's no output to capacitors, which means you may need bigger capacitors to account for that if you insist on using it... for let's say a 48v rectified amplifer, it would be great)

[jtruc34]

480mV sounds like schottky diodes... those aren't ideal to use to rectify AC voltage, especially in audio applications. Or maybe check the datasheet and see the voltage drop at higher currents, 480mV may be at 1A, not 4A. Also, you'd probably want a 10A or higher bridge rectifier, something much higher rated than 4A or whatever you think your device is gonna use.

Why would I want to double the current rating? Why are Schottky diodes non ideal for this application? I'm working at about 15V, well below their low breakdown voltage. I've checked and the 480mV drop is at 5A.

[Benta]

Well, actually I've realized that I have much less headroom than anticipated. The transformer I have (and I can't justify buying a second one) is 12V, which means 17Vpp in normal conditions, but knowing mains can get as low as 210V and factoring the diode drops (480mV for the one I use), I only get 14.3V, which is way too low to have two LM338/317 in series, and with 1V of ripple it's only 13.3V.

I'll have to opt for LDOs. Would you still recommend to regulate twice? Or maybe I could do something else at the PSU and only regulate at the point of load.

Thanks

Concerning transformer/storage cap selection, you're in even deeper trouble than what you describe.
You can NOT say: 210VAC min. = 10.96 VAC output = 15.45 Vpeak - Rectifier drop = cap charge voltage.
It's much worse.

The issue is, that the caps are only charging at a fraction of the 100 Hz period time, meaning the charge current gets larger and larger when you increase the cap size, as the charging time gets shorter.
This reduces your peak secondary voltage due to resistive losses in the transformer. And this drop can be several volts. At some point you'll reach the "point of diminishing returns" where increasing cap size will bring zilch, because the ultra-short charging time will drop the transformer peak voltage to a point where nothing more can be gained.
Trying to calculate this is very, very complicated, Spice modelling quite easy if you can measure your transformer parameters.

In your spot, I'd do the simulation in any case, as your headroom is severely limited, and you need to have an idea of where you are.
317/337s et al are out, whether you have enough for LDOs is open.

A last ditch attempt if all the above fails is adding windings to your secondary to increase the voltage, but this only really works with toroids.

Let us know how you get along.

Cheers.

[Benta]

those aren't ideal to use to rectify AC voltage, especially in audio applications.

Why? Please explain.
Thanks.

[mariush]

Maybe I shouldn't have said.  It was based on various "audiophile" advice I've heard in the past, but didn't think about it thoroughly before writing it and now that I think about it I can't find a good reason why they shouldn't be used.

As for my statement about why double the current, I was simply thinking of how to keep that rectifiers diodes cool - passing 5A through a single diode, even with forward voltage of 0.5v, will cause a couple watts to be dissipated on it, which means the diode will heat up significantly.
One of those bigger packages will be easier to mount to a heatsink, and there will be more headroom, you may be able to use a smaller heatsink or maybe even no heatsink, because the package helps a tiny bit to spread the heat.

[srb1954]

480mV sounds like schottky diodes... those aren't ideal to use to rectify AC voltage, especially in audio applications. Or maybe check the datasheet and see the voltage drop at higher currents, 480mV may be at 1A, not 4A. Also, you'd probably want a 10A or higher bridge rectifier, something much higher rated than 4A or whatever you think your device is gonna use.

Why would I want to double the current rating? Why are Schottky diodes non ideal for this application? I'm working at about 15V, well below their low breakdown voltage. I've checked and the 480mV drop is at 5A.

Using a capacitor input filter the diode current waveforms are very spiky i.e. they have a high peak current to average ratio. Due to this high peak current the diode power dissipation can be higher than a simple measurement based on the DC rating would suggest.

It is instructive to do s SPICE simulation to see this effect but you will need to include the transformer parameters to get an accurate simulation of the current waveform and the diode power dissipation.

[jtruc34]

So, I've realized there is no way my 230:12 transformer is gonna do it, even with LDOs (without buying LDOs so expensives that I could buy a second transformer), and unfortunately my transformer is toroidal but its hole is filled with some hard polymer which I couldn't remove, so adding turns is out of question.

The two options I have is 230:15 or 230:18, and I decided against the latter because I would need some massive heat management. The only way I could be safe with 15V is still with LDOs (taking a margin of 10% on mains power), and I've opted for mariush's  suggestion of GB's bridge rectifier.

To chose the 13.5V, there will be a maximum of 1V drop on U2/U4 taking a bit of margin gives 13.5. So 15*sqrt(2)*0.9 - 2.2 is 16.9V. There will be at maximum a 1V drop on Q3/Q6 and a 0.5V drop on U3/U5, it thus leaves a margin of 16.9 - 13.5 - 1.5V = 1.9V. If I can squeeze the ripple below 1.9V I'm good. I will have to do some testing to be sure.

Do you think my solution is sensible?

Thanks

[Benta]

You're ignoring the issue I outlined in reply #16 completely. You really need to take it into account.

[mariush]

I think it's over-engineered actually.
You calculated 2.2v for the bridge rectifiers when it will be more like 0.8v per diode, and it would be cheaper to parallel a couple bridge rectifiers to get lower voltage drop, than complicate your life with those transistors.

For example here's this $1.3 30A bridge rectifier : https://www.digikey.com/en/products/detail/micro-commercial-co/GBU3010-BP/12177159
Check datasheet .. the 1v drop typical for 15A is at 25c, while in use the rectifier's gonna be more like 50-70c on a heatsink ... and at 4A and 75c you're looking at around 0.8v drop

40 mF may be a bit too much, it may blow the fuses and cause a big current spike on the bridge rectifier at startup - again a simple solution could be just paralleling two bridge rectifiers.

If you're gonna have regulators for each set of 4 why even bother with pre-regulators... lose those and just use a couple bus bars or thick traces to bring whatever voltage to the local regulators

[Benta]

Check datasheet .. the 1v drop typical for 15A is at 25c, while in use the rectifier's gonna be more like 50-70c on a heatsink ... and at 4A and 75c you're looking at around 0.8v drop

You still don't get it.
Peak current is going to be 10..,.20 times higher than average at the voltage peak. Ever heard of "crest factor"?

[jtruc34]

You're ignoring the issue I outlined in reply #16 completely. You really need to take it into account.

Fair enough, I ignored it because I don't know the characteristics of my transformer yet but maybe I'm too optimistic of actually being able to get 2V of ripple with 4-5A. What values are relevant for the problem you outlined? Is the series resistance of the secondary enough to model it in SPICE or should I get the inductance as well as the coupling factor? I'm not sure. The inductance is not very hard to measure but I guess it will strongly depend on the saturation of the core which is a mess to model.

I'll measure the secondary resistance of the one I've got and hope the new one will have similar characteristics.

than complicate your life with those transistors.

They are not there to reduce drop, they are there because the linear regulator is not capable of delivering 4-5A. (LDOs with such a high current are really expensive)

If you're gonna have regulators for each set of 4 why even bother with pre-regulators... lose those and just use a couple bus bars or thick traces to bring whatever voltage to the local regulators

n
That's a good point, for which I blindly assumed without thinking that it was better to do so, as many people directly suggested to use preregulation. I'm not sure if it is strictly necessary, but in my case it serves the purpose of dissapating most of the heat in the PSU and not on each per-module regulator, which is good because I won't have to mount heatsink on each of them but only on the one in the PSU.