Investigating a full linear PSU for the XXHE PC and NOS1

[acg]

PC POWER SUPPLY SUPERVISION

After several worthless google searches I have come up with the following information for a way to make a linear ATX supply that will start ANY atx power compliant pc...

Google search "PC Power Supply Supervisor" and you will come up with a plethora of chips that do the ATX power stuff for us...for example.

A good result...with a little effort we should now be able to make a linear ATX supply that will start and stop any pc.

Anthony

[acg]

Hi all,

I thought I would post a couple of pictures of different design approaches for the ATX psu.

The first photo below shows a three rail linear psu that puts 7A into the three rails of the P24 ATX plug and contains the appropriate chip for handling the ATX power sequencing in conjunction with the motherboard.  Nice...I think that something along these lines would be an excellent final product and very achieveable.

The other end result is something like this Teradak photo.  Note that the first picture will also require passive cooling (the heatsinks) and multiple transformers.

Which layout would you would prefer, the circuit board or the box of wires?  I think we may be able to get a circuit board of this type into the XXHE PC (to keep the wire very short from the regulators to the motherboard and peripherals) and have an external box containing the transformers.

Cheers,

Anthony

[CoenP]

I think the linear psu will run quite hot especially when challenged, not an ideal match with a hot running processor in one casing. Regulators have to drop voltage to operate so amperes=heat. You will need an awfull lot of cooling surface to keep within thermal operating range of the regulators. To get the necessary thermal resistance with a small cooling surface you may have to resort to active cooling.

Furthermore the 7+ amps let the voltage droop and increase the ripple on the input of the regulator. To fight that you need a lot of smoothing capacitance or a higher input voltage (which means even more heat). This may not be at the important music playing moment, but you really have to overspecify the transformers and capacitance nog to run into out of spec voltages on start/shutdown transients.
(To get the idea: in the priomordial phase all computers used to have -large- linear supplies. You can still find the very large capacitance, low voltage "computergrade" capacitors that were specified in surplus stores)

I think the teradac option and build looks sound to me: plenty real estate, sufficient capacitance, necessary ss for regulation, big recitifiers and a large cooling surface.
I ve been thinking of a setup that provides additional stabilised voltages to the pc wherefrom the voltages of the sata drives/PCIe card will be stabilized in the pc or any "low power" powerline.

I thought of this as a too big a project to pull off in my spare time. I also wondered and still do if a linear supply in combination with multi picopsus is the way to go. As i stated before, it may just be the switching input with PFC on the AC alone that is the culprit of the "lesser" sound of switching psus. You may reap 80% or more of the sonic benefits when changing that input to a linear one providing ie 24 or more Volts and have the wide range picos do the supply duty to the pc components. The 24v+ is just to minimise the current for a given power requirement (with only 4A you allmost have 100Watts available...) and ease the design of the linear part of the supply. Remember that there is allways some serious switching action on the motherboard to provide the cpu with 50-100Amps@1.5V from a 12 V powerline...

Regards, coen

[acg]

Thanks for the reply Coen.  Sometimes I think I am talking to myself in this thread!

I think the linear psu will run quite hot especially when challenged, not an ideal match with a hot running processor in one casing. Regulators have to drop voltage to operate so amperes=heat. You will need an awfull lot of cooling surface to keep within thermal operating range of the regulators. To get the necessary thermal resistance with a small cooling surface you may have to resort to active cooling.

I don't think we will get anywhere near 7A on any rail of a linear psu for the XXHE psu.  Two 5A rails, two 2A rails, two 1A rails and a 0.1A rail are all that could possibly be used (not all will necessarily be worthwhile, but that is the most that could be jammed in there).  On top of this the XXHE PC in its standard case has three case fans (plus at least one extra fan slot possible I think) and two cpu fans, so the active cooling is already in place.  Pull out the SMPS and the expansion drive bays and there is 30cm x 12cm x 12cm to fit in some regulators inside the Silverstone GD06B case so there is some scope at least.  Transformers could go outside.

Furthermore the 7+ amps let the voltage droop and increase the ripple on the input of the regulator. To fight that you need a lot of smoothing capacitance or a higher input voltage (which means even more heat). This may not be at the important music playing moment, but you really have to overspecify the transformers and capacitance nog to run into out of spec voltages on start/shutdown transients.
(To get the idea: in the priomordial phase all computers used to have -large- linear supplies. You can still find the very large capacitance, low voltage "computergrade" capacitors that were specified in surplus stores)

I understand your capacitance/voltage/transient points from  a 'sounds good in theory but have never done it for myself' perspective.  The aim of the project is to have a reasonably priced supply that optimises playback quality with the XXHE PC and NOS1...so while, as you say, it is all well and good to have a computer that works as well as it can whilst playing music, it still does need to be able to startup and shutdown properly.  We now know how much current is needed to achieve these things...the challenge now is to 'right size' everything and minimise the noise in the XXHE PC...which is where I need help. 

I think the teradac option and build looks sound to me: plenty real estate, sufficient capacitance, necessary ss for regulation, big recitifiers and a large cooling surface.
I ve been thinking of a setup that provides additional stabilised voltages to the pc wherefrom the voltages of the sata drives/PCIe card will be stabilized in the pc or any "low power" powerline.

I reckon it looks good to me too, but I have a good idea of how much that case costs and it is substantial - probably more than half the total cost of the supply.  This is another reason that I think it may be worthwhile investigating making use of the active cooling already in the XXHE PC.

I thought of this as a too big a project to pull off in my spare time. I also wondered and still do if a linear supply in combination with multi picopsus is the way to go. As i stated before, it may just be the switching input with PFC on the AC alone that is the culprit of the "lesser" sound of switching psus. You may reap 80% or more of the sonic benefits when changing that input to a linear one providing ie 24 or more Volts and have the wide range picos do the supply duty to the pc components. The 24v+ is just to minimise the current for a given power requirement (with only 4A you allmost have 100Watts available...) and ease the design of the linear part of the supply. Remember that there is allways some serious switching action on the motherboard to provide the cpu with 50-100Amps@1.5V from a 12 V powerline...

I think that you could well be right that a picoPSU fed a single linear rail may provide a solid benefit Coen, but as of yesterday I think I can make my own fully functioning picoPSU (minus the switching regulators), and may be able to provide three high quality rails into it for around the cost of a picoPSU plus a good high wattage off-the-shelf linear psu.  That has got to be worth a shot.

If I believed everything that I read online I would tell you that in order of sonic gain the best things to provide linear power to are:  USB card; o/s drive and P8 cpu socket. The great majority of people give up with the P24 socket and simply provide linear power to a picoPSU, but the few that have bothered to do this well report a sonic gain.  The problem with the picoPSU is that it is unlikely to power up the XXHE PC in its standard un-optimised form.

And yes, there are plenty of switching supplies on the motherboard but there is nothing I can do about that.  I have read about providing linear power directly to RAM and also of changing the motherboard oscillators with higher quality units, but they are thoughts for another day.


So where I am at right now, is that I know how much power to poke into various parts of the XXHE PC, and I am pretty sure I can knock up a little circuit board to manage both the linear power supply and the motherboard, but I don't know how to design and construct high performance voltage regulators, so I am looking for help.

Cheers,

Anthony

[pedal]

Thanks for the reply Coen.  Sometimes I think I am talking to myself in this thread!

Anthony, I do read it carefully but I don't have much to add about the tech talk which is above my capability.
I do, however, intend to purchase both the XXPC and your PSU when it becomes available.

Keep up the spirit and many thanks for the research you provide here!

:-)

[acg]

Thanks for the support Pedal.  I will plug away at this but I am now at the limits of my recollection of high school electrical theory (it was 25 years ago!! - how did that happen??).  I really need help with a lot of the stuff from here on and I can see myself wasting a LOT of time if I am allowed to trudge off down the wrong paths.

Saying that, I have been reading this afternoon and it seems that there are two primary ways of making each linear rail:

• Monolithic Linear Regulator from someone like Linear Technology (eg. LT1086 or LT1084).  Decent noise and ripple rejection (PSSR), good transient response, simple implementation (I think), lower heat and quite cheap
• Shunt + Opamp diy regulator in the style of something like Jung Didden and it's derivatives.  Better ripple rejection, lower noise, not sure about transient response but larger, hotter and more expensive to build

Peter and Coen - is this an accurate summation of the two approaches to linear regulation?  Are there more approaches that I have not learned the existence of?

Anthony

[PeterSt]

The active cooling in the PC ? well, I guess you can put a ship's engine in there and it will still be cool enough. Sadly the fans will make more noise than that engine. Also I don't think that any overly dissipating (because actually too hot) will work at the somewhat higher ambient temperatures, let alone close to 30C/86F. Careful now because people from the hotter parts of the globe may think this will never happen because of air conditioners, but in countries like here we don't have them. We take it for granted that a couple of times per year we almost die of heat and we grab an extra cold beer. In the mean time 26C/79F is very normal during the better weather, and that anticipates cold amps or otherwise ...

So don't underestimate what happens once the heat can't be dissipated because of the somewhat higher ambient temperature. And this goes "exponential".
But yes, the case can have some for sure, but it won't be the way to do it. And, also "know" that no cold wind will cool a small surface efficiently. It really needs a heatsink of sufficient size. Think that metal-metal flows better, and that metal-air actually creates a protective film (little layer).

Lastly, supposed it will be 80C/176F in that PC case. I don't think it will be a problem, and for example the CPU will still be 40C/104F or whatever it originally was. Nothing will die of it and nothing will melt. One problem : nothing will last long either. So, if you look at the life ratings of capacitors, you will see that this is at 60C/140F and then they last 1 or 2 years, depending on the quality (but see spec). Make that 80/176 and in no-time they will be gone. Or, make it 40/104 and they will last forever.

Peter

[PeterSt]

Peter and Coen - is this an accurate summation of the two approaches to linear regulation?  Are there more approaches that I have not learned the existence of?

What I have been thinking about for many times is just making as many PSU's from the ground up as needed. Hey, look in the left leg of your NOS1. Those ...

But ... I never knew the current needed per rail and such but we know that now. I didn't look it up in this topic, but I think max needed is 4A, right ?
Ok, that PSU can do 3A.

I am also not sure anymore whether -12V is needed but that can be done too. It just needs more of these ...

Also see next post.

[PeterSt]

Maybe here is another perspective; this is not new for me, and maybe especially because of that it could be part of the "design" :

Minimize the useage at the hardware side.

For example, I see that we are talking about support for SATA drives and such. But what if no single drive would be in there ? at least that is what I personally am aming for. Not so easy, but I will get it done.
Let's say that this decreases the usage with 6.5W, which is what I measured for a 3.5" spinning disk. So, little too much maybe, but maybe we can shut down a couple of other things because if this (anybody looked what SouthBridge actually does ? well, can probably not be shut off, but could use as much as the cpu).

What I also notice - and without knowing the real merit of this - is that by now I am using 1.5W less than reported earlier. Is my meter burning in ?
So Idle is 51W now and playing is 53 - 55.5W.

Deduct that 6.5W and we have 49W left for playback.
Don't ask me about the needed headroom, but I think I can see 65W in shorter spikes, like when decoding FLAC. This will be 58.5W without the disk in there.
What I'm saying is that for shorter whiles things are allowed to heat up a bit and next we need safety for boot and maybe shut down.


So, never mind the real merits of this, I think it can always be useful. Also see next post.

[PeterSt]

It's a bit of a trick, we might also blow something at attempting it, but I think this can work :

Create a capacitor bank for each rail;
I'm afraid it was a bit before your time here Anthony, so you won't have such a card : PCI supply rail noise (but many of us do). Anyway, almost the same principle I am talking about, and putting this in even might work (but I don't think so). Thus the basic idea :

Have that capacitor bank for each rail and anticipate a power outage of what ? 0.2 seconds ?
Connect your PSU (of whatever due kind) through it.
In parallel connect the normal ATX through it.
Never have switched on both at the same time, although I actually wonder what will really happen when so after all. But let's not do that.

Boot the PC through the normal ATX SPSU;
Runs ? then switch off that one and switch on the lower rated LPSU of which we know it can take all when things run and do it before the caps run empty. Runs empty too soon ? have more of them.

Want to shut down ? then go the other way around.

I probably make a mistake or two but at this moment I think this can work.
??
Peter

[PeterSt]

but at this moment I think this can work.

And two minutes later I already think that we will be subject to the cap's "ripple" and that this won't be so noise-free.

... And when I write *that* I see myself again thinking that I still don't see much how a nice (electrical) noiseless supply in there will help audio SQ ...

... Which tells me that even additional noise from those caps is not a problem, while in the mean time the SPSU is still out of the way which *is* a problem to my belief (spitting into the mains).

Hmm ...

[PeterSt]

Hey, look in the left leg of your NOS1. Those ...

Btw, output ripple of that is 0.5mV. FWIW, way less than that 120mV on the 12V and 50mV on the 5V/3.3V.

[CoenP]

Saying that, I have been reading this afternoon and it seems that there are two primary ways of making each linear rail:

• Monolithic Linear Regulator from someone like Linear Technology (eg. LT1086 or LT1084).  Decent noise and ripple rejection (PSSR), good transient response, simple implementation (I think), lower heat and quite cheap
• Shunt + Opamp diy regulator in the style of something like Jung Didden and it's derivatives.  Better ripple rejection, lower noise, not sure about transient response but larger, hotter and more expensive to build

Peter and Coen - is this an accurate summation of the two approaches to linear regulation?  Are there more approaches that I have not learned the existence of?

Anthony

The lt series is only good for a few amperes of regulated current. That will do a HDD or a fan best. There are other more capable monolithic solutions. The datasheets also tell you how to boost the current with extra power BJTs. They are very low drop and available in the needed voltages.

I feel heat is still an issue. I know that you can get away with other constraints ie a lower ambient temperature, but as Peter said, component life expectancy progressively reduces with increasing remperature.
You also have to realise that most regulating solutions do NOT regulate transients or hf components. Actually the regulator may have an inductive output feeding oscillations.

The shuntreg is a worst case class A circuit, that is that is has to provide the peak current all the time, so if you need say at peak 2.5 Amps, the currentsource and sink have to both dissipate at least this 2.5 Amps. For a 12v currents sink alone this is 30Watts. Adding the source, you easily end up with a 40-50 Watt dissipation. You should save those for small signal audiocircuitry or ultra energy efficient PCs.

As they say on the silentpcreview site: a little air goes a long way. This is very true for the heatsinking of my cpu. When called for duty very little airflow can keep my megahalems cool compared to a hot touch  when absent. If you decide to go the powerdissipation route, you might consider previous generation cpu coolers as heatsink (to save some money). The are designed to get 100+ watts out of the CPU and keep it cool with a little air. As can be seen on the heatsink datasheets forced air can get small heatsinks to archieve 0.8deg/watt or lower.

As you will know most switching PSUs are not very efficient in the low wattage. If you measure the powerdraw on the AC side, the real powerdraw by the motherboard and components will be less so that helps a little.

I fully agree with minimising the powerrequirements for the PC first. This will make life much easier. I hope Peter will share some ideas on the "unconnected" PC he talked about before. Undervolting and undrclocking are a good start.

Regards, coen

[acg]

Awesome idea with the capacitor bank and the ATX supply to simply start the XXHE PC Peter and of course I have an idea or two in addition.

As mentioned earlier, I am sure that I can use a $2 chip and a small circuit board to do the ATX power on and off stuff.  One particular chip in mind can control two 12V lines (as well as a 5V, 3.3V, 5V SB, PS_ON etc.) which is important for this build.  I am sure that a little bit of additional logic can be included on that circuit board to also manage the supply that is to be used when the XXHE PC needs to boot or even on the occasions when the XXHE PC is not running in minimised o/s mode.

Wait for my next post...I will explain more...I have to put the kids to bed right now.

Cheers,

Anthony

[PeterSt]

5V Rail

There are 5 5V pins on the P24 socket and each pin alone will power up the mobo if all the other 5V pins are not powered.  I cut all the 5v wires, twisted the mobo end of the cuts together into one joint, remove all but one cut 5V wire from the SMPS side of the cuts and made the measurements below...

o/s State	Startup	Idle	Music	Shutdown
Maximised	4.0A	2.2A	2.2A to start, 2.0A stable	2.3A
Minimised	4.1A	2.2A	start 2.2A, stable 2.0A	2.2A

Like the 12V rail, the 5V rail does not seem to be impacted by the sate of the o/s.  I am not confident that 4.0A peak during startup has been recorded accurately because it is displayed on the multimeter as a rapidly changing amperage which may mean that the peak has been missed.

Actually this 5V seems the only problem. Ok, see next post.