Future PSU designs

[exxos]

As many know I designed my own PSU..

https://www.exxosforum.co.uk/forum/viewt ... f=27&t=218

Though soldering the regulators is no fun

I have been looking around and found a soic8 regulator which may be slightly more efficient and should be a drop in replacement to fit my output stage.

The specs are similar, though while I could use a more smt regulator of the one I am using now, I would much prefer to bolt the regulator to metal for heat dissipation than use the PCB copper as a heatsink.

I have ordered a 5v demo board to see how hot it gets and to see what its regulation is like...

The new chip according to the data sheet is most efficient at a lower voltage than I am currently using.of course I will test all input voltages vs load and efficiency when it arrives..

How it looks, if I can use a lower voltage transformer,then it will be physically smaller. If I can free up some PCB space then I really hope to fit a 110/230v selector switch. This will save me in having to stock 2 different versions of the PSU.

The new ic has inbuilt over voltage protection.. The problem I see if for some reason someone shorted out the 12 & 5v rails, that the ic would shut down, but 12v into the 5v rail would still likely blow up the motherboard..

What I do currently is the "crowbar" where should either 5v or 12v rise more than half a volt, then the crowbar is tripped and the fuse blows . problem is setting up the trip voltages is a nightmare as even tollerences or brands of sense diodes used can make or break the protection. So I have to spend a lot of time in testing each individual board that the protection does what I want.

Point being, do people want the crowbar protection? Assuming I used the new ic, it still has protection modes, just not the crowbar, if I was to leave out the crowbar I could build the PSUs quicker and keep them in stock easier.

[exxos]

I am currently looking at a texas part also ...

http://uk.farnell.com/texas-instruments ... dp/2536016

While the IC is more expensive, and more parts, it is higher efficiency and being SMT parts is easier for me to solder in batches.

It has one of the best efficiency figures I can find, and can use ceramics on the output. Various protection features etc. So I may buy the EVM to see how it really performs.

It does not mention performance with a ripple filter, but I will test with and without. If there is not much gain with, then it can also be removed.

I am also looking at another part with internal inductor,

http://uk.farnell.com/texas-instruments ... dp/2213255

Efficiency only good for 1amps or there abouts. But as normally nothing uses 12V anymore, then "bad" efficiency under 1amp isn't going to really matter since little load = little heat, and efficiency is a bit irrelevant in more like 100mA areas.

Again more expensive IC, but if I can remove the 2 output inductors then costs even out a little.

It also looks like the bulk electrolytic capacitors may not be needed, but this depends on my analysis .

[exxos]

One demo board just arrived

1.2 V my arse

IMG_1880.JPG (95.05 KiB) Viewed 13870 times

Now doing a bit of initial testing...

IMG_1881.JPG (136.08 KiB) Viewed 13870 times

I am loading the 5V rail with approximately 2 A...


The datasheet said the output ripple would be approximately 10mV but I am actually on X10 so this is actually 100mV

bcc.png (4.15 KiB) Viewed 13870 times

Zooming in a little more...

ccc.png (3.84 KiB) Viewed 13870 times

There does not appear to be any actual output ripple as such, it looks more like transient spikes. It could be temporary requires higher output capacitance values to clamp down on this.

Of course I have only just started with this board, there is much experimenting to do yet


Preliminary efficiency results below.

Code: Select all

OUTPUT
4.95V   2.5R  1.98A  9.801W

INPUT
13V  0.82A  10.66W 91.94%
14V  0.77A  10.78W 90.92% 
15V  0.72A  10.80W 90.75%
16V  0.68A  10.88W 90.08%
17V  0.65A  11.05W 88.70%
18V  0.61A  10.98W 89.26%
19V  0.59A  11.21W 87.43%
20V  0.56A  11.20W 87.51%

So clearly a low input voltage is more efficient. This was actually to be expected, as the output switching stage should be optimal at around double the input voltage as the output voltage. For example 5 V output efficiency would most likely be best with a 10 V input.

However I cannot use such low voltages here, as the power supply also has to drive the 12 V circuit. I preliminary worked out that somewhere around steam volt input with the 12 V regulator circuit I have in mind should work out the most efficient. Obviously I cannot use much less than 16 volts to drive the 12 V output. Efficiency wise on the 5 V rail is still not too bad 16 V input anyway.

However this demo board has been optimised for 1.2 V output. I had to edit it slightly to get 5 V output, so the output inductor value is not optimal any longer. This is actually why I needed the inductor calculator

In any case, I now have a baseline for a comparison to see if the changes I make increase efficiency or not

[exxos]

You results using a 2.2uH inductor.

Code: Select all

13V  0.81A  10.53W  93.08%
14V  0.76A  10.64W  92.11%
15V  0.71A  10.65W  92.03%
16V  0.67A  10.72W  91.43%
17V  0.64A  10.88W  90.08%
18V  0.60A  10.80W  90.75%
19V  0.57A  10.83W  90.50%
20V  0.55A  11.00W  89.10%

There is now almost 1.5% increase in efficiency at 16 V input

The output ripple now is clearly visible at 1 MHz. Overall peak to peak away form is slightly worse, but at least you can see the ripple over just a garbage of noise like before.

cccc.png (4.69 KiB) Viewed 13867 times

So now to increase the inductor value to 4.4uH...

IMG_1883.JPG (62.84 KiB) Viewed 13861 times

Code: Select all

13V  0.83A
14V  0.77A
15V  0.72A
16V  0.68A
17V  0.64A
18V  0.61A
19V  0.58A
20V  0.55A

I couldn't be bothered to calculate the wattages and percentages but clearly this is worse than before

ddddd.png (4.63 KiB) Viewed 13866 times

Oddly the output ripple now seems a bit less..


According to the datasheet calculation, I should be using 5.5uH.. Clearly this value does not test as good as 2.2uH. So likely I will stick to that value.

Of course one thing to bear in mind here, is that while 5.5uH may actually work better, the problem is the resistance of the inductor doubles because of longer length of winding, which will decrease efficiency.

I had to solder 2x 2.2uH in series as I did not have a inductor of that value. Though even if I purchased such a value, the resistance would be higher anyway. I may purchase one of the large flat coil wound inductors just out of interest to see what the efficiency is like then..

The only inductor I have which has practically zero resistance is a 10uH flat wound inductor so I will give this a try...

IMG_1884.JPG (62.03 KiB) Viewed 13861 times

Code: Select all

8V   1.31A  10.48W  93.52%
9V   1.16A  10.44W  93.88%
10V  1.04A  10.40W  94.24%
11V  0.95A  10.45W  93.79%
12V  0.88A  10.56W  92.81%
13V  0.81A  10.53  93.08%	
14V  0.75A  10.50  93.34%
15V  0.70A  10.50  93.34%
16V  0.67A  10.72  91.43%
17V  0.63A  10.71  91.51%
18V  0.60A  10.80  90.75%
19V  0.57A  10.83  90.50%
20V  0.54A  10.80  90.75%

So now the efficiency the same as the 2.2uH inductor. So resistance indeed was a factor.

Just out of interest I also went down to 8 volts. As predicted before 10V gives the best efficiency just over 94%.

Around 13V to 15V is likely thereabouts whether transformer output voltage will end up after rectification.

The output ripple seems to have vanished again but there are still some spikes. More output capacitance might that though..

010.png (4.32 KiB) Viewed 13860 times

Overall, if the 2.2uH inductor had near zero resistance like the 10uH, then I would suspect the efficiency will be slightly better with the 2.2uH and match the 10uH.

Considering the output ripple seems non-existent with 10uH, I am now leaning towards that value may be a better choice. Of course need to reduce the output spikes yet.

The problem here the flat coil wound inductors (10uH) are several times the cost of the smaller 2.2uH inductor. They are also a lot large and take up more PCB space which I do not really have.

Of course this was the same problem I had with my first PSU design. It boils down to paying a lot more just to gain 1% efficiency. It is not really a good trade-off.

Would seem a larger value inductor gives better efficiency with a high input voltage. Of course 10 V input would be ideal, but I cannot use this as I will also need to power the 12 V switching regulator, where this really needs a ideal input voltage of 24 V. Obviously 24 V input for the 5 V regulator is not going to work out well. So likely some compromises will have to be made somewhere..

[exxos]

So I added a 1500uF capacitor on the output and it really did not like this...

222.png (4.6 KiB) Viewed 13855 times

While this design does not need large output capacitance, and indeed could simply not be added on the power supply board, the motherboard itself is typically a 4700uF and a 2200uF and possibly others, dotted around the power supply rail. So while design is pretty awesome, when powering the motherboard, performance could actually end up worse than the original Atari power supply!

This is actually the reverse problem what I had with my original power supply design. Where adding more capacitance simply worked better.

So now I need to evaluate this power supply design is even feasible to use..

Another update..

a1.png (3.58 KiB) Viewed 13854 times

It seems a regulation is better now, but there is a lot slower ripple which is not there without the 1500uF capacitor. Going to say this is an oscillation, and it is at 2KHz according to the scope frequency.. Not really sure why this is there

I am guessing it is the resonant frequency between the 10uH inductor and the 1500uF capacitor.. which kinda sux..

So I went back to the 2.2uH inductor and did not see any odd oscillations anywhere..

Zooming into the 1 MHz switching frequency, I see there is about 70 mV peak to peak. So adding more output capacitance is basically halved the spikes on the output test.

a2.png (3.78 KiB) Viewed 13850 times

Technically without the spikes, we probably have somewhere around 30 mV ripple. This is actually pretty good in one respect as my previous power supply design actually needed a ripple filter and I got the ripple voltage down to 7 mV. So 30 mV without a filter is really good. But also a little annoying as the regulation on this is actually worse than my original power supply. Adding the filter would likely improve matters a lot and equal my previous power supply design, but I am really looking into not using a ripple filter if possible.

Of course the spikes are probably not really there, scope probe actually can pick up RF interference. I had this issue before when testing my power supply design out. For example if the ground wire of the scope is moved closer to the inductor, the spikes get considerably worse. I do know it can be very complicated to take these measurements accurately, even so, these are probably the worst case figures anyway

[exxos]

The second PSU demo board arrived today..

IMG_1889.JPG (88.4 KiB) Viewed 13823 times

Same resistor load as previously. Regulation "out of the box" is about 300 mV. Which is pretty bad really. But then again this demo board was the only one I could find using this particular switch mode IC, and it just happened to be a USB charger, so likely not great regulation is to be expected..

zzzzzzz1.png (5.06 KiB) Viewed 13823 times

With a 16 V input it pulls 0.73A which is 83.91% efficiency. According to my notes regulators I am using currently are 82% efficiency. So this particular new switch mode circuit is about the same. The IC does get fairly warm fairly quickly, but this is down to the internal MOSFET resistance being rather high.

For those interested, older MOSFET technology does have high resistance, but it is likely cheaper as well. One point here is that a lot of switch modes now have internal overcurrent sensing, when a MOSFET has been like 100mR resistance, a sense amplifier to detect a voltage drop does not really require much amplification, and this basically means a cheaper circuit.

When going down to MOSFETs like the previous board I was testing a few days ago, are towards 10mR , this gives really good efficiency as shown, but it also would require a high amplification on the voltage drop for the current detection to work. This means more internal transistors and will increase the cost stop also the cost of low resistance MOSFET is probably higher by itself. So this makes the IC expensive, but also very good.

The problem here is, or the new switch mode ICs as shown previously are so tiny it is not really realistic to use these inboards which are hand made. Even so, when I got my ram boards made, the DRAM chips almost always had some bad connection somewhere. So around percent of boards would need fixing when they came back from manufacturer. These were all supposedly made by machine as well so they should all work can be reliable, but it never seems the case. So I am reluctant to use very tiny parts like this. Not only would sending them off to be manufactured be expensive, but I foresee a high failure rate in which I would end up spending a lot of time fixing them. Of course making these by hand takes a lot of time also, while I could solder this IC, I was actually trying to make life easier soldering not make it more complicated.

The entire goal of this research is to create a quick to assemble power supply design while trying to increase efficiency. Unfortunately these requirements at the moment seem to be conflicting. Also if the regulation is not going to be as good as my original design, then I really do not want to make a product which is worse off than the first one I created.

So I will try and spend a little time and seeing if I can increase the efficiency and regulation of this second power supply demo board, though at this point I think there is little options left than to continue to produce my original design.

I tried changing the output inductor to the one I used in my power supply, and oddly the regulation be slightly worse, but then again, why linking to inductors is not particularly ideal to start with. I could not obtain the part number used in the demo board sadly but I do have a low resistance so I will compare that later.

I tried adding a 1500uF capacitor on the output and this did not affect regulation at all. As I can see 125 kHz, and this is about what I see is running out, and this circuit desperately needs a output ripple filter.

So I added one...

IMG_1890.JPG (86.44 KiB) Viewed 13820 times

.. And increase the capacity value some more, ultimately did not really seem to affect the output

[exxos]

I have had another look at power supply chips and found this one..

2379552-40.jpg (11.13 KiB) Viewed 13814 times

http://www.ti.com/lit/ds/symlink/tps56528.pdf

The datasheet claims 95% efficiency with a ultralow 10 mV ripple, though I think it is a little misleading as they are quoting 1 V output. So probably the actual output ripple would be more like 50 mV. This is around the same regulation is quoted by the regulators I am currently using. Both datasheets quote that output ripple is down to output capacitor value and ESR. So I suspect both ICs have similar output ripple.

Internal MOSFETs seem about as good as things get this size package. It is a SO8 package show should be easy to solder.

I have ordered the demo board, so hopefully it will come next week, and I think this will be the final power supply design that I will be testing as these power supply boards starting to get a little expensive

[exxos]

New efficiency results..

Code: Select all

9V	1.18A  10.62W  92.29%
10V	1.06A  10.60W  92.46%
11V	0.97A  10.67W  91.86%
12V	0.90A  10.80W  90.75%
13V	0.83A  10.79W  90.83%
14V	0.77A  10.78W  90.92%
15V	0.72A  10.80W  90.75%
16V	0.68A  10.88W  90.08%
17V	0.64A  10.88W  90.08%
18V	0.61A  10.98W  89.26%

As suspected, and as found, the general rule of thumb is output voltage times to for input voltage for best efficiency. As shown above, the 5 V output, 10 V input has the best efficiency.

The problem is we also need a 12 V output, which means we are going to need ideally 24 V input based on that role. Realistically 16V input is probably going to work out the best as the efficiency for 5 V is still 90%, and of course steam volt is higher than 12 V so we can still generate a 12 V output with some other IC. The thing I will have to do is hunt for a 12V switch mode circuit which has the best efficiency with 16 V input.

Regulation as shown below..

3rdt.png (3.68 KiB) Viewed 13794 times

I have started using a different scope probe as my previous one was picking up a lot of RF noise from the inductor on the power supply board. Overall I am actually seeing 120 mV peak to peak noise, but this greatly goes up if I place my finger near the inductor. It is actually very problematical to take these readings because of RF noise. I have already documented this before. Likely regulation on my previous results are probably 10 times better and I actually found.

Thankfully this time with my new scope probe, I could actually see the proper ripple voltage and I have dismissed all of the noise as RF noise generated by the circuit itself. So currently I have about 50 millivolts ripple noise, which is pretty good. I'm sure was some experimentation I can improve on this.

I'm also going to order some larger test inductors with near zero resistance, as I want to see the efficiency will jump up a little higher still.

Overall I think this board will be a ideal choice for consideration in future power supply designs. The heat output is practically nothing on the switch mode chip itself, and it is in a standard SOIC8 package.


EDIT:

Ordered 2.2uH, 3.3uH, 4.7uH, 6.8uH..

*drools*

COILCRAFTSER2900-40.jpg (7.05 KiB) Viewed 13791 times

[exxos]

I had a 10uH flat coil to hand as I use previously so I mash that in to see what happened..

IMG_1894.JPG (66.3 KiB) Viewed 13787 times

I checked efficiency and it severely jumped up, this is amazing!

Code: Select all

8V	1.30A	10.40W	94.24%
9V	1.15A	10.35W	94.70%
10V	1.04A	10.40W	94.24%
11V	0.95A	10.45W	93.79%
12V	0.87A	10.44W	93.88%
13V	0.80A	10.40W	94.24%
14V	0.74A	10.36W	94.60%
15V	0.69A	10.35W	94.70%
16V	0.65A	10.40W	94.24%
17V	0.61A	10.37W	94.51%
18V	0.58A	10.44w	93.88%

Even though there is not much between the efficiency figures, 15 V input seems to give the best efficiency at almost 95%!

And for the regulation ripple now, I can hardly detect it...

tt.png (3.67 KiB) Viewed 13787 times

The ripple is about 20 mV peak to peak. Though I suspect the actual regulation figure is 12mV. the datasheet does actually say 10 mV or better regulation. So I am right on the money now Regulation is probably better than I am measuring as my scope probe is picking up switching noise.. I guess around 5mV is the actual output ripple.

I will try all the value inductors as well, these should be here tomorrow.

As normal with farnell of late, I choose items in stock when I search for stuff, add them to my basket, check out, and then they are out of stock!? even worse, the coil is not going to be in stock until end of December

[exxos]

My new coils came. I tried 3.3uH, 4.7uH, 6.8uH. Overall well efficiency figures did change it was hardly noteworthy. As a datasheet says the higher inductance value gives better ripple performance, and that is what I see as 10uH as the best figures. The datasheet also says surge value is also affected by higher inductor values. As my previous power supply used 47uH and the search value seemed fine, and I do not think this is anything to worry about anyway. Obviously 10uH than currently using now for testing is a lot lower value anyway.

The problem I have with this design is start-up current, his PSU is not capable of driving any external capacitance. This means placing even a 1,000uF across the output basically kills the power supply from starting up, which is a little bit of a problem The IC has 5 A output current, but the problem is with low ESR capacitors, when they are discharged, they act like a short circuit, so they could actually need 10 or 20 A for a very short time to initially charge them. Once a capacitors are charged not a problem as the resistance almost becomes like a open circuit. Current IC claims that it will be tripped after about seven cycles. Though as this will be drawing more than 5 A for more than seven cycles that it is a little irrelevant.

After more searching I found there is a almost identical IC, but with a adjustable soft start feature. So I am hoping the soft start will allow the voltage to ramp up slow and reduce the surge current on power up. Rather than ordering another demo board, I just ordered the IC, I only have to change one pin circuit so this demo board should be fine to test this other IC. This chip should ramp up the voltage slowly meaning the turn on current should be more controllable.

This new IC also has even lower MOSFET resistance figures so should slightly be more efficient as well.

EDIT:
The inrush current assuming a 1500uF 20mR ESR capacitor is.. 5 / 0.020 = 250Amps

inrush.jpg (41.42 KiB) Viewed 13767 times

Where the current does not reach below 10 A until after 100uS.


Running a simulation with just one ohms series of the supply, this limits the current to approximately 5 A..

1ohm.jpg (31.92 KiB) Viewed 13766 times

So now we need about 10 milliseconds at 5 A to charge the output capacitor.

Of course this value capacitor is relatively low to what is normally on the Atari motherboard. So we would probably need soft start of 100 ms or thereabouts.