Multi Voltage PSU +/-15V +/-24V +48V +5V [pics/updates]

Hi Guys,
I've searched around on the PSU meta, but noone has really mentioned such "complicated" design =) :

I need to build a quality PSU that would output +/-15V, +/-24V, +48V, +5V... "Why?" would you ask? Well... I've built a stereo version of EZ1290s (hense the +/-24V & +48V) and I'm trying to integrate the PCM4222EVM into the same chassis, essentially making the unit a 1073DPD. The PCM4222EVM requires +/-15V for analog stages and +5V for digital things.

Also, I'm trying to separate the power trafo/rectifier/smoother part from reg/bypass/etc. stuff so that I have the trafo/etc in an external box (for hum and dimensions related issues) and the regulation would be inside of the actual rack.

Also, I THINK, I have to have a separate (floating) ground for the PCM422EVM... At least for the digital part of it - if you look at the schematic - http://focus.ti.com.cn/cn/lit/ug/sbau124/sbau124.pdf  - page 7 and pages 17-18... Would it be OK to merge the PCM4222EVM analog ground with the EZ1290 ground?

So, summarizing everything, I'd like to design 2 boards - one would reside in a box along with IEC inlet+fusing+on/off switch and the transformer. It would contain a couple of rectifiers and smoothing following them. Then I would have smoothed unregulated DC travel through a multiconductor shielded (?) cable (as few conductors as possible). Then inside of the actual rackmount chassis would be the second board with all the regulation & further smoothing.

What would be the most efficient (read: "smart" =) way of accomplishing all of this?

Thanks!!


PS: I will need about 2A for the whole thing (2x EZ1290 + PCM4222EVM + Phantom + extra LEDs etc..). Currently I have a 100VA 22-0-22 transformer that feeds the FiveFish PSU-2448 to power EZ1290s. It be awesome to know if there is a way to keep the transformer and not have some custom wound stuff, etc. =)

JLM 5 rail psu?

Yes, except it would have a single common ground for all the voltages. And I would need to have at least a 6-conductor cable carrying DC from external PSU box to the rack. I remember someone reputable on this board wasn't particularly fond of this method and was recommending regulating everything physically close to the load (who knows what could happen to a perfect regulated DC running through a cable) =)

But yeah, JLM PowerStation seems like a good plan-B =) Thanks!

Other thoughts?

....

Even VCC and VDD on the 4222EVM and 0V on EZ1290 and chassis being a single ground? Why did TI guys separate them in the first place then?


EDIT: Also, out of curiosity, would it make sense to use PSU2448 +/-24V outputs, splitting it 3 ways - one for each EZ1290 (there are two) and putting the 3rd one into another set of LM317/337 to regulate down to +/-15V? Then I would only need to come up with separate 5V...

comp pwr supply

they are cheap, and dirty.

comp pwr supply

they are cheap, and dirty.

Click to expand...

...and noisy, and loud, and have +3.3V, +/-5V & +/-12V voltages only =)

Question: Where exactly in the PSU (or perhaps before?) should the ground be split to be considered a separate potential ground? Like in case of VCC & VDD in the PCM4222EVM - my trafo's secondary center tap becomes ground, then what? Can it be the same throughout the whole PSU? What about digital noise being dumped into the VDD ground on the way back and interfering with the analog stages, and vise versa? How do you (and should you) isolate these two grounds at the PSU?

[quote author="promixe"]Also, I THINK, I have to have a separate (floating) ground for the PCM422EVM... [/quote]
Nope. Look at Figure 4 on page 17 of the PCM4222EVM. Just below the converter you can see that AGND and DGND are linked. Besides, the PCM4222 will not like it very much if there's more than a couple dozen mV between those nodes.

[quote author="promixe"][...] Why did TI guys separate them in the first place then?[/quote]
To keep noise on the digital ground out of the analog circuitry.

[quote author="promixe"]What would be the most efficient (read: "smart" =) way of accomplishing all of this?[/quote]
Simply have separate +/-15V and +5V regulators powered from the unregulated voltage lines coming from your transformer/rectifier/smoothing caps. According to the EVM manual, you need to supply up to ~200mA on each rail, simple LM317/337 regs will do fine. Heat will be a bit of an issue, so have the regs on heatsinks and consider a dropping resistor between your rectifier cap and the regulators, especially for the 5V reg.

JDB.

What do you guys think of this:

External part of the PSU (IEC inlet, fusing, EMI filtering is not included):


Internal part (in the actual rack):



This is all very rough, I haven't finalized values of all of the components yet, but you get the idea.

...and consider a dropping resistor between your rectifier cap and the regulators, especially for the 5V reg

Click to expand...

What would be the recommended wattage/value of R8? Is this the correct placement for it?

Also, I'm not sure about the +48V part... Does it look ok to you?

Thanks for your answers!


PS: I know it might be an overkill to use LT1085/1185 regs, but I'd like to have high quality stuff for this and for future projects, so...

[quote author="promixe"]What do you guys think of this:[/quote]
Looks OK. I haven't checked your voltage setting resistors, though. Didn't you also need a -24V line?

Nits:
- Even though the LT1085 data sheet doesn't explicitly mention it, I would include a bypass diode between ADJ and out.
- Use good high-temp electrolytic capacitors on the regulators' outputs rather than the recommended tants.
- Up C18 to ~1000u and parallel it with a 100n ceramic cap, so it and R8 can make a nice low-pass RC filter to keep the digital supply's crap out of +VUnreg and vice versa.

[quote author="promixe"]What would be the recommended wattage/value of R8? Is this the correct placement for it?[/quote]
This is the right place. As for value/wattage, just ask Mr. Ohm. You need to know the maximum current drawn on each supply line.

Rough example: let's say that the max draw on +5V is 500mA, and you want to keep 5V across the regulator to keep it well outside its dropout region. This makes the max dissipation of your regulator 5 * 0.5 = 2.5W, which a TO-220 reg should be able to handle just fine with a small heatsink. +V1_unreg is approx 30V (give or take line voltage variations, which I'll leave to you to figure out). So R8 needs to drop 30-5-5 = 20V at 500mA. Resistance would need to be V/I = 40 Ohm, call it 39R to get to a standard value. Max dissipated power is V*I = 10W, so use a 20W type to be on the safe side. Better yet, split R8 in two equal-ish value resistors (in this example: 18R+18R or 18R+22R) and connect their midpoint to ground with another 1000u cap. This creates another RC lowpass filter stage, and two 10W resistors may well be cheaper than one 20W type.

I would recommend you do something similar with the +/-15V regs, as they can do with dropping 10V or so. Dissipating heat in a resistor is almost always cheaper than upsizing the regulator's heat sink, the resistors provide a measure of overcurrent protection, and the extra RC filtering doesn't hurt either. Just make sure you keep the capacitors away from the (potentially hot) resistors and regulators!

[quote author="promixe"]PS: I know it might be an overkill to use LT1085/1185 regs, but I'd like to have high quality stuff for this and for future projects, so...[/quote]
What's so great about the LT1x85s in this particular application? Its higher current handling capability isn't needed, neither is the low dropout voltage. Its ripple rejection looks to be similar to that of the LM317, and at first glance the thermal performance is determined by the TO-220 package and heatsink, not by the silicon that's inside. Noise spec is the same too. On the other hand, the AbsMax voltage difference between input and output is awfully close to the expected value of V_unreg, so output shorts (or even large/low-ESR caps on the output rail) are quite likely to kill the part dead, much more so than a plain old LM317 which has 10V more headroom in that department.

JDB.

JDB, Thanks for your tips!

...split R8 in two equal-ish value resistors (in this example: 18R+18R or 18R+22R) and connect their midpoint to ground with another 1000u cap. This creates another RC lowpass filter stage...

Click to expand...

I'm not quite sure how you are getting 1000uF figure for the RC cap. I'm assuming I shouldn't touch anything below 120Hz with the filter? 1000uF starts at around 2.7Hz with 60R total resistance. So, shouldn't I use a 22uF cap instead? That would give me 120Hz corner frequency.

This is the +5V part where I'm trying to incorporate your suggestions:


I'm calculating ~60R for R11+R12 and 7W total dissipation when dropping 20V at 350mA.

Thanks!

[quote author="promixe"]I'm not quite sure how you are getting 1000uF figure for the RC cap.[/quote]
Larger is better (within reason -- a 10000u cap may be overkill, and will likely have worse HF behaviour due to its higher ESL). For regular DIY this calls for 'the largest cap you happen to have lying on your bench'. There is no reason not to have this filter extend down to as close to DC as you can conveniently get (supply sequencing might be one, but that usually calls for more sophisticated solutions than cap value tweaking). I picked 1000u as you already use that value as your post-rectifier smoothing cap.

Note that if you're calculating the response of this R-C-R filter section, you need to parallel the two resistances for a worst case guesstimate, assuming zero supply and load impedances. With 20R/1000u/20R, this yields a LF corner of ~16Hz. 120Hz is 3 octaves higher, and will be suppressed by 6 * 3 = 18dB. Not much, but it helps.

[quote author="promixe"]This is the +5V part where I'm trying to incorporate your suggestions:

[/quote]
Add another electrolytic cap in parallel with C14, or the 317 will not be too happy. I'd suggest another 1000u cap. Those 100p caps will not do very much, I would up them to 100n.

JDB.

The regulators pin numbers seem mixed up for LT1085 and LM317.
From Datasheet Pin-1=Adj., Pin-2=Out, Pin-3=In.
At least for the +48V part you will need a diode in parallel to R12.
Also pull a min.load of ~3.5-12mA (min.load tolerances vary from different manufacturers, but with R12 at 120R you already pull 10mA, so maybe sufficient) for this +48V part, as phantom, other than the rest of your preamps, is not always enabled.
For your voltage doubler I'd up C5+C6 to 220uF or 470uF, 50V or 63V if size matters. YMMV.
just my 2ct

Thanks for your replies guys!

I'm working on a redesign to incorporate most of your suggestions and gonna post it tonight or tomorrow.

I have a couple of design affecting questions:

1) For sake of heat dissipation and part count would it make sense to move the RC filter (R11+R12+C1+C4) to the external PCB (the one with rectifiers\smoothing caps) and make it drop 10V to use for both +5V and +/-15V sections in the internal PCB? At the expense of running one extra wire in between PCBs.

That way the #1 heat source inside the rack will be the +5V reg dropping about 15V and RC filter's heat will be inside of the external PSU box. And I could loose 4 components from the design by using a single RC filter arrangement for both +5 and +15 rails.

Or would it be more beneficial to not combine these RCs, using discrete ones intead for each rail because of digital noise issues?


2) JDB, according to how you explained RCR filter calculations, if I use 30R/1500uF/30R setup, that should filter around -25dB at 120Hz? Just want to verify the math I derived from your posting. =)

I'd suggest another 1000u cap. Those 100p caps will not do very much

Click to expand...

Sorry, that's actually a typo, I meant to use 0.1u caps there, which are 100n like you're suggesting, not 100p...

For your voltage doubler I'd up C5+C6 to 220uF or 470uF

Click to expand...

Harpo, I actually used the C5/C6 values from another design as I'm not entirely aware of the actual purpose of those two caps. If you could briefly elaborate on how you're calculating those figures, so I could understand the semantics behind it, I'd greatly appreciate it =)

Thanks!

Okay... Here is my next revision of both internal and external parts of the PSU:

External:


Notes:
- Changed C5/C6 caps to 470uF as recommended by Harpo

- Added LED/R1 - with unregulated voltage fluctuating in the range of 30-36V I calculated around 1W heat dissipation with a regular 3V/30mA LED.
Question: is this the absolute best place to incorporate an LED? Given there is no regulation in the external part and I need an LED there...


Internal:


Notes/questions:

- Is there a benefit of using polyprop caps instead of ceramic anywhere it calls for 100nF parallel bypass?

- Is this ok to have C3, C13, C23, C33, C43 1000uF caps in the internal part eventhough there are already C1, C3, C7 smoothing caps in the external part that are 1000uF? I'm looking at various PSU designs and it's usually just the smoothing caps that are that large and then smaller bypass caps and then it goes into the regs.

- Given this many large electrolytic caps everywhere in the design I'm starting to question the size of the prospective board layout... I thought I could do something like JLM powerstation board size wise, but my design now looks a lot more space demanding. My first priority is still quality, not size - but a nice ratio would be a benefit.

- has anyone seen this article? - http://tech.juaneda.com/en/articles/lm317.html. This guy states that the less caps you have in your linear PSU the better its load will sound... Of course he acknowledges that there is a lot of noise without the caps but says caps make the sound "boring" and "unnatural". He suggests using a few good polyprop caps.

- Harpo, thanks for pointing out LM317 incorrect pin assignment. FIXED!

- Every RCR filter resistor value is calculated based on 350mA needed current per rail.



If everyone thinks this is a passable solution, I'm going to jump into designing (my first) PCB for this project. =)

[quote author="promixe"]- Added LED/R1 - with unregulated voltage fluctuating in the range of 30-36V I calculated around 1W heat dissipation with a regular 3V/30mA LED.
Question: is this the absolute best place to incorporate an LED? Given there is no regulation in the external part and I need an LED there...[/quote]
That's a good place for a LED, but most small LEDs (3mm/5mm) are much happier running at 20mA, and you may find that one of those ubiquitous HE LEDs are blindingly bright even at 10mA.

[quote author="promixe"]- Is there a benefit of using polyprop caps instead of ceramic anywhere it calls for 100nF parallel bypass?[/quote]
Unlikely. It does help to use X7R/X5R ceramic capacitors over Y5V/Z5U, as the former have less capacitance variation over voltage, temperature and age, and are less microphonic. If you feel like experimenting with the 'sound' of your supply, pick a 5mm/7.5mm pin spacing for your decoupling cap footprint. You can find several PE film caps in that footprint, as well as ceramic decoupling caps, although I would be surprised if you could hear a difference, especially on the preregulator caps.

[quote author="promixe"]- Is this ok to have C3, C13, C23, C33, C43 1000uF caps in the internal part eventhough there are already C1, C3, C7 smoothing caps in the external part that are 1000uF? I'm looking at various PSU designs and it's usually just the smoothing caps that are that large and then smaller bypass caps and then it goes into the regs.[/quote]
Yes, that's fine. To answer an earlier question, you may want to consider moving C1/C21/C31 and associated resistors to the external supply box. While this is not absolutely necessary (and it increases the number of wires in the cable between both boxes), it will reduce interference between supplies in the same way as using a star ground does. The line to the digital supply will benefit most from this treatment.

[quote author="promixe"]- has anyone seen this article? - http://tech.juaneda.com/en/articles/lm317.html. This guy states that the less caps you have in your linear PSU the better its load will sound...[/quote]
...for his preamp design (which looks like it may have less than perfect power supply rejection)
...to his ears. Your Mileage May Vary.

[quote author="promixe"]- Every RCR filter resistor value is calculated based on 350mA needed current per rail.[/quote]
Measure that. The RCR filters act like current limiters, so if your circuitry needs more, the supply will go out of regulation. Having said that, 350mA looks like a safe ballpark estimate for the PCM4222EVM, as long as you don't short anything and don't start hanging relays, lamps or other current hogs off the regulated supplies.

[quote author="promixe"]If everyone thinks this is a passable solution, I'm going to jump into designing (my first) PCB for this project. =)[/quote]
Your call, but a one-off (or even a three-off) of this circuit is pretty easy to build on perfboard, and perfboard is much easier to change/experiment with ("what happens if I add another RC stage, or change these ceramics for these big PP caps I found on eBay?).

Good luck,

JDB.

[quote author="promixe"]

For your voltage doubler I'd up C5+C6 to 220uF or 470uF

Click to expand...

Harpo, I actually used the C5/C6 values from another design as I'm not entirely aware of the actual purpose of those two caps. If you could briefly elaborate on how you're calculating those figures, so I could understand the semantics behind it[/quote]
I guess, you are refering to Joe Malone's 3-rail supply or Keith's KPS-1, probably better explained by themself.
You may find this article or this page helpful on understanding voltage multipliers.
1st. get your power requirements. In case a faulty mic cable shorts your phantom, the worst case current draw would be (2x 48V/6k8)=14mA. Normaly operating condensers draw up to 10mA or from hearsay even 11mA. With matching secondary winding you could get away with ~2uF/mA straight rectified, but in multipliers the caps are in series so the availiable current drops by each voltage step-up, so double this. Keith's Villard- or Cockroft-Walton multipliers (these are 2 in parallel for full-wave, so less ripple voltage, double ripple frequency) with 2x100uF charge pump and 220uF (=2x100uF in parallel) loading cap will sufficiently supply 2 mics. Your C7 at 1000uF (=2x470uF in parallel) would call for at least the same value for charging caps ie 470uF for C5 and C6. This should even be enough for about 10 condensers. Voltage rating for C5/C6 for your 22-0-22V secondary will be 22VAC x 1.1 for 10% overvoltage x squareroot(2) for peak voltage, so at least 35V. Double this voltage rating for C7, so at least 70V.

Hi Guys,

After a looong break from this I'm back to figuring out the PSU for this project again. I have modified my schematics based on everyone's suggestions above and came up with this:





I've designed a double sided PCB that can be used either as a single piece or a split solution - one half goes into external box for rectifying/filtering/heat dissipation, and the other half can be inside a rack where the regulation etc. occurs. Here are the snapshots of top and bottom:




I got the PCBs a few days ago, hooked it all up and it WORKED! It spits out precise +5V, +24V, +15V, -15V, and +48V, which is what I need for the project at hand. Now, here is where all the downsides start:

1) I've used the 1000uF cap diameters based on Mouser's inventory and search filter function... Should've looked at datasheets to find out that the smallest 1000uF diameter is 16mm, not the 12.5mm that I have on the PCB... Oh well, just gonna live with it for now...

2) More serious problem, and this is where I need advice of someone experienced: The original idea of RCR filters doubling as heat dissipation doesn't seem to work. When I touch the power resistors they are slightly warm to touch, maybe a bit above room temp. The regs on the other hand are REALLY hot when I power up the PCM4222EVM and use it for less than 3-5 minutes... Any ideas why this could be judging by the schematics above? I'm hoping this is maybe just a component value miscalculation somewhere.

3) More like a continuation of #2 - The unregulated rails past RCR filters (at the harness junction) don't seem to reduce voltage, it is roughly +32.5V at the +5 rail, +34V at +24 rail (which is OK I think, cuz it doesn't have dissipation resistors on that rail), +33.5 at +\-15V rails and +69V at +48V rail (the last one is normal I assume). So, as you see, there isn't any voltage drop at RCR points... I wonder why...

Any help is, as always, greatly appreciated!

Promixe

UPD: I've removed Co9/Co10 and measured voltage drop across Ro11/12 - it is just 1.9V, which doesn't really make sense... Now, the resistance of Ro11+Ro12 is about 58R when the PSU is OFF. When I turn it on and measure - it's 0R...

Also, when I look at +V24 at UNREG OUT with the scope I can see a noticeable ripple, since it's just a single 1000uF cap before it. When I look at +V5 at UNREG OUT, just past the Ro11/12 without the Co9/10 caps, there is no ripple. It looks like it's been smoothed out by at least one more large cap, but where? I'm kind of puzzled...

Also, it looks that together Co1/Co2 + Ro11/12 are forming a high-pass filter... Would this have anything to do with not seeing the ripple on +V5 UNREG OUT?

promixe said:

I've designed a double sided PCB that can be used either as a single piece or a split solution - one half goes into external box for rectifying/filtering/heat dissipation, and the other half can be inside a rack where the regulation etc. occurs. Here are the snapshots of top and bottom:

Click to expand...

Looks nice, but as was pointed out before you'd want some more space between the resistors and the capacitors. Resistors get warm to hot, and cap life is inversely proportional to temperature.

promixe said:

2) More serious problem, and this is where I need advice of someone experienced: The original idea of RCR filters doubling as heat dissipation doesn't seem to work. When I touch the power resistors they are slightly warm to touch, maybe a bit above room temp. The regs on the other hand are REALLY hot when I power up the PCM4222EVM and use it for less than 3-5 minutes... Any ideas why this could be judging by the schematics above? I'm hoping this is maybe just a component value miscalculation somewhere.

Click to expand...

Have you measured how much current is drawn? At max (or typical) current consumption, how many Volts between input and output of the regulator? Across the resistors?

The numbers we ran earlier would work out to a dissipation of ~1W in the 5V regulator. That's well within limits for a bare TO-220 in a room temperature environment, although a small heatsink would help extend its life, especially when mounted inside a case.

promixe said:

3) More like a continuation of #2 - The unregulated rails past RCR filters (at the harness junction) don't seem to reduce voltage, it is roughly +32.5V at the +5 rail, +34V at +24 rail (which is OK I think, cuz it doesn't have dissipation resistors on that rail), +33.5 at +\-15V rails and +69V at +48V rail (the last one is normal I assume). So, as you see, there isn't any voltage drop at RCR points... I wonder why...

Click to expand...

Did you measure this with or without a load? Resistors follow Ohm's law pretty slavishly, and if there's little Amps through them there will not be too many Volts across them.

Bottom line: please give us the context for the measurements.

promixe said:

When I turn it on and measure - it's 0R...

Click to expand...

Very few Ohmmeters give useful results when the circuit is powered. In the best case a garden variety multimeter will give you bogus values, in the worst case you'll have blown some of its circuitry.

promixe said:

Also, when I look at +V24 at UNREG OUT with the scope I can see a noticeable ripple, since it's just a single 1000uF cap before it. When I look at +V5 at UNREG OUT, just past the Ro11/12 without the Co9/10 caps, there is no ripple. It looks like it's been smoothed out by at least one more large cap, but where? I'm kind of puzzled...

Also, it looks that together Co1/Co2 + Ro11/12 are forming a high-pass filter... Would this have anything to do with not seeing the ripple on +V5 UNREG OUT?

Click to expand...

Exactly. The +/-15V and +5V lines have a C-R-C-R-C filter, whereas the +/-24V supply only has a C buffer cap after the rectifier. Even if you pull Co9/10 you'll still have filtering through Ro11/12 and C3/4, For the values given in the schematics that has a -3dB point of 1/(2*pi*60R*1000uF) = 2.7Hz, attenuating the 100/120Hz ripple from a typical full-wave rectifier by about 30dB (assuming no load!). My napkin doesn't have enough room to do the math for the situation with Co9/10, but I'll guarantee you that it'll be better still.

JDB.