Design aim -
1) Safe low voltage design
2) Best possible output ripple regulation
3) Best possible surge current regulation
4) Fast efficient switchmode regulators
5) Fix onto old PSU framework.
The "front end" is a toroidal transformer. I chose a 50VA type ( about 4amps output). These are actually pretty efficient and possible more efficient than the switchmode transformers and related losses. This also means no nasty high voltage switching transistors and everything is a safe low voltage operation.
I did originally consider a switchmode instead of a toroidal, though likely a custom core would have to be manufactured to fit the design and this would ramp up cost to much. I did look around for "off-the-shelf" transformers, but there seemed to be little choice and some of them were actually very expensive. A cheaper alternative would be to re-use the transformer cores off the old SR98's etc though I think there would be little point in reinventing basically the same thing.
High voltage switching ultimately results in a lot of heat loss in the transformer core and main switching transistor. Such transistors are prone to failure either due to high voltage spikes or bad thermal contact to heatink cause overheat. Overall, the toroidal while being larger, is a lot more efficient and more reliable than a high voltage full switchmode design.
After working on switchmodes for several years and never really seeing a reliable design, I am a lot more confident that a full low voltage operation will be long term reliable and offer greater efficiency than the original ST PSU's.
The regulators are switching regulators, so heat losses are kept a lot lower than using linear regulators. Switchmode regulators give better efficiency (less heat) and offer fast reaction times for regulation over slower linear types. Whereas the original switchmodes offered anywhere from 200mV to 600mV regulation, low voltage designs offer much tighter regulation and can even offer regulation down to just 2mV!
One thing I almost overlooked, is when the original PSU's fail, DC is applied to the switchmode core and the output voltage falls to zero, which means the 5v & 12V rail fall to zero volts quickly. The down side of a regulator is there is no way to know how it may fail. Of course they should never fail, but I did not want to simple ignore this safety feature. So I have added in a "Crowbar" onto both the 5V and 12V lines. So should the voltage rise to high on either levels, the crowbar kicks in and will simply blow the onboard fuse and protect the Atari circuit form damage.
The "black holes" in the PCB are to allow the regulators to be bolted onto the metalwork of the existing PSU base and used as a heatsink and a solid ground plane. The PCB is the same physical size as the original PSU, so will bolt down in all 4 corners as normal. Each screw point is now also used as a ground. This makes the groundplane out of thick metalwork which is advantage and is a free heatsink. The regulators should not get to warm anyway, so the metal area is more than enough for cooling.
The hardest parts were the inductors. This always drives me nuts as there are so many terrible types on the market that I could write a book on it it, in fact I think I did write a paper on it at some point :) With physical size constrictions and cost being a factor, some small compromises had to be made.
Mostly it is the heat dissipation of the inductor. I worked it out at less than a watt which isn't much really, but this could be reduced further with more expensive inductors, but the cost jumps up to like £10 a pop. 4 of them is £40 so you can see my problem there. Really though, the more expensive inductors just are higher current rated, so thicker wire, larger core, so wattage lost isn't as much. But not worth paying £40 just to save about half a watt of heat dissipation. Overall, it shouldn't get as hot as the original PSU's anyway.
The inductors do not start to saturate until about 3-4amps, which is the max rated power of the Atari PSU anyway. Though in general, according to the regualtor datasheets, as the output current rises, the required inductance generally goes down. For example, if you only wanted 500mA then you would use a 680uH inductor. Towards 3 amps is 100uH inductor. Beyond 3amps gets to about 60uH. Which actually works out perfectly as the inductor starts to saturate at that point, so the inductance actually starts to drop, which actually is a win-win situation....
...Of course about that point the internal current limit of the regulators will kick in. Regulators can push about 3Amps RMS like the Atari PSU, and have something like 6amps peak current. On short circuits the regulator will switch into a cycle per cycle shutdown mode which basically shuts down the regulator until the fault is removed. The transformer is 4amps rated so we shouldn't be pulling more than that in total anyway.
The limits are also physical space constrictions on the PCB, a 60VA transformer simply wouldn't fit. Higher amp switchmode regulators start ramping up costs also, so I stuck with 3Amps as its the same as the Atari PSU's. It is doubtful anyone will use more than 3Amps anyway.
Diodes on the board are the lowest voltage drop Schottky diodes I could find. So this keeps efficiency as high as possible in that part of the circuit. Capacitors have about over 3-10amps amps ripple current so all good there. The main reservoir capacitor is about 10,000uF with added 2,200uF on the regulator outputs for best ripple regulation possible. Keeping the reservoir value high also helps with regulator stability and better regulation.
Overall designing these things , like with most things, is one huge compromise with everything. I have tried to keep efficiency as high as possible and keep the regulation as high as possible , without spending stupid money on parts. Each time I build a few up I may tweak the parts used. If I find something slightly better for not much more money then I will fit it. Though with everything optimised already, such "improvements" are not going to have a huge impact on anything.
Some parts not arrived yet (for the protection circuit) but enough to start testing.
I noticed a voltage drop to about 4.8V under heavy load. This turned out to be the output filter inductor :-( Though after more testing it seems the inductor wasn't actually helping regulation anyway. If anything it was making things worse. So currently the inductor has been removed. While the regulations have a 2% output voltage tolerance, I got 4.96V under heavy load (about 2-4 amps) so well within tolerance. The open circuit voltage is 5.05V as shown in the image.
The inductor gets a bit warmer than I had expected but that's the tradeoff with that size of inductor. It is only 0.06ohm so not bad compared to most. I keep looking for alternative, but the next best ones are flat coil wound types which are pretty huge and cost a lot more. If I ever do a new batch of boards I may try to make provisions to later fitting a larger inductor if people want to at a later time. Most likely the board will have 2 less inductors, so it frees up a little PCB space to give the option of a larger type. Though I've got 50 of these PCB's so unlikely a new design will be done any time soon.
The regulation is pretty good so far...
This may look insanely bad, but this is on a 10mV scale (0.01V) From what I can see there is a oscilation around 69MHz. I First though it was the self resonance of the inductor itself, though after looking on the data sheet it lists it as 6.5MHz.
After much pondering, I tried another set of scope probes..
This is exactly the same 5V rail, and both sets of scope probes are connected to exactly the same place (both signal and GND). The probes also don't seem to care the the GND connection is actually on GND or not either. There doesn't seem to be any RF being coupled to the probes. So the 20mV ripple is actually 10mV on another set of probes. So some more thinking to do yet...
So here is the current test setup in case people are wondering how it fits together..
More tests only thing time moving the GND clip on the scope probe, WOW / WTF is all I can say...
More tests above moving the probe around. So I can vary the voltage "read" by moving the ground wire about.
As it stands, I am pretty much unable to get accurate ripple figures from my PSU. As to why this has only just become a issue I have no idea why. I think its clear currently that the regulation must be pretty good considering the voltage is stable and no dimming on the display while the floppy drive turns on. So a job well done at the very least :)
So I had a idea..
The issues were mainly because the probes picking up noise from the motherboard, so , remove the motherboard :) I have a 200watt 20R rheostat resistor so used that instead. I pulled about 4amps of current on the 5V rail and got just 20mV of noise! (0.02V) This is exactly the same with or without a load. So this design has some serious good regulation from 0-4amps!
Now I am working more "as expected" I revisited the ripple filter. Images show with and without a 1uH filter inductor. The data sheet suggested 22uH but with the voltage drop of over 0.2V its just not viable to use a large value without a drop in voltage across it due to its winding resistance. So next up I will try and find a higher current inductor of maybe 1-2uH with less resistance than my test inductor. This way it will have better ripple filtering without the voltage drop (hopefully).
New filter inductors in place. 18Amps rated 0.0013ohm so voltage drop is basically zero. While a higher value inductor would act as a better ripple filer, it starts running the risk of voltage drops which is the main thing I want to avoid. The ripple is almost not visible. We are taking 2mV average ripple so its basically nothing. But I will keep experimenting as should a second design be done, I may update some things in the future.
I have spent some time testing efficiency. 1amp on the 5V rail works out at 84% efficient. 2amps works out at 82% efficient. This is really good for a low cost solution. The regulators probably don't need a heatsink for 2amps load. Though I have loaded it up to 4amps and efficiency starts to drop and the regulators start getting warm. Most ST's operate below 2amps so it works out well anyway.
I also spent some time experimenting with transformer voltages while doing efficiency tests. 12VAC input works best. This works out at about 17VDC under a 2amp load which hits the best efficiency figure of the switchmode circuit. Higher or lower voltages start to effect the efficiency by a couple % so overall I am pleased that it all balanced out for best efficiency :)
Looking more like a complete PSU now :)
Looking at a Phihong switchmode of 240V 0.6A input is 144watts input. The maxium output wattage is stated as 39watts. This would yelid a efficiency of about 56%.
During a typical ST load of 2amps the switchmode section of my PSU can operate at about 86% efficiency. The bulk of the losses are in the switching regulators themselves. Hopefully in the future more efficient regulators will become available. The toroidal I have not checked, but typically there are around 95% efficiency So the best overall efficiency of my PSU it likely to be around 80%. The maximum output power would be around 60watts which is the limit of the transformer. Overall the PSU is optimised for the best efficiency running possible based on a typical STFM machine load of 2amps.
The second batch of PSU's now have a small PCB ontop of the regulator with 2 SMT capacitors on. This is NOT a modification to the design. The capacitors were being soldered on the bottom of the PCB on the regulator pins themselves. Only this was turning into a small nightmare in soldering. So the capacitors are now ontop of the PCB on a tiny PCB.
Also , some boards had 2 resistors on the bottom, now there is only 1. The resistor biases the 5V line zener to trip out at 5.5V. I couldn't get the exactly voltage zener diode so had to bias with a resistor. Some boards may now just have 1 resistor as I changed to a different brand of zener to which needed a slightly different resistor value where it was a value I could actually get.
2 Inductors (lighter gray in above image) have been changed for a slightly larger type. I managed to find a larger value inductor with still low resistance, so I used them. In theory the larger inductor should have given slightly better ripple figures, but didn't seem to change anything. So there is no advantage to either inductors which have been fitted.
At some point in the mix the SMT diodes have been changed. I managed to find some lower voltage drop diodes so later builds will have those. This doesn't effect the performance, but may help with gaining marginally more efficiency.
Later batches will be missing a small capacitor and 2 resistors. It was a small RC delay network to delay the turn on for the regulator but its not required.
Some capacitors may be changed for larger values or physically larger types in due time. While the cost is higher with larger capacitors, I am generally using similar values for my PSU re-cap kits. So in order to reduce the amount of stock I have to keep buying, it will allow me to use the same capacitor values in multiple projects.
There are some possible mods listed below. Both are non-crytical "faults" and unlikely anyone will have these issues. Though in case someone does, then they can do the mods as listed below. If you do not have the issues, then there is no need to do the mods. You can however still do the mods if you so wish.
I don't know of any issues with these symptoms on PSU's which have sold so far. Though there is always that chance that someone may run into some odd issues in the future. Generally later batches have had the modifications done already.
MOD1 - Trip voltage testing & fix (aka fuse blowing for no reason)
As mentioned above there have been some minor changes over various batches. I have had problems with tolerances on the "4V" zener diode. Where I later soldered a resistor or 2 on the bottom of the zener to correct the trip voltage. However its possible some zeners might be even more out of spec than I initially found. No ill effects will happen other than the 5V trip voltage being "over sensitive" which results in random blowing of the fuse.
To check the trip voltage, a meter need to be placed on 0V and the "back end" of the 4V zener diode. One end will measure about 5volts, the other should measure somewhere around 0.20V-0.30V or lower. Due to tolerances on the SCR the trip voltage can be anywhere between 0.4V and 0.6V. So if the zener voltage is higher than about 0.3V then it may run the risk of blowing the fuse for no reason at all. The SCR should not trip until 0.6V though again with tolerances on parts its possible for it to trip at 0.4V. So I suggest anything over 0.3V is start to get a bit close to tripping.
Typically I worked out the values to trip at 5.5volts. Though I did not expect the tolerances to vary so much, so the "typical values" fitted might not suit all PSUs. Of course all PSU's have been tested before I send them out. Though as parts warm up over several hours of use, they may drift enough to blow the fuse.
Option 1-
Should the voltage be higher than 0.3V then I suggest increasing the resistor values on the bottom of the PCB (soldered under the 4V zener) by about 100R-200R and try the zener voltage again. This additional resistance should be in series with the existing resistors. Typically the value there already will be about 2.4K, so the value should end up being 2.5K and 2.6K. Do not increase the value to high otherwise the trip voltage could end up being over 6volts.
Option 2-
The zener diode and resistors are removed and a new zener used in its place. This zener is a BZX79-C5V1 diode and manufactured by NXP. I have not tried other brands of diodes, though the NXP ones , at least so far, seem to be the ideal ones to use.
If you do not have a problem with the fuse blowing, then there is no need to do the above fix.
MOD2 - "soft start" power up fix.
Later revisions have this "feature" was removed. The fault would show up as showing 3volts on the 5V rail, simply because the regulator was turning off and on very fast. After about 30-60seconds the PSU would work normally.
If you have this problem then there is a 47K resistor near the 4V zener (near the left regulator itself) needs to be removed and a wire link placing there instead. This will make the regulator as "always on" so it removes the possible power up issue.
In later builds of the PSU, there is a 47K and small capacitor missing near the SCR on the top right of the board. Those parts are no longer fitted and no longer needed once the first 47K resistor is linked over.
If you do not have this issue then there is no need to do this fix.
The Falcon PSU is a tweaked design of my ST PSU. Mostly moving over to more SMT based parts. This allows me to assemble them a bit quicker. The main change is I changed the rectifier out for four Schottky diodes with a lower voltage drop. I thought the rectifier got a little hot so efficiency would suffer slightly. So the change to SMT Schottky Diodes allows much cooler running at will increase the efficiency of the PSU.
The PCB sees some minor layout changes. The 12V regulator has been moved so it can now be bolted onto the chassis. There isn't likely any need for it to be bolted unless something heavily uses the 12V line, which is unlikely, but I have not checked. In any case, this 12 V regulator I advised be bolted down.
The PCB copper thickness has doubled and I have used more tin plating to keep track resistance down. 5V actually came out at 4.95V on the ST design due to just a few millimeters of track! Amazing! The thicker copper now shows the proper 5V level. Marginally increases efficiency, but this voltage drop was more of a annoyance than anything. Some parts of the PCB have also been tinplated to keep resistance down.
The PCB corner holes have been made a fraction larger as there seemed to be some alignment issues on some of the metalworks on some PSU's. It is difficult to manufacture a "one size fits all". Though while some hole alignments may still be tricky, it is still a improvement over the previous layout.
There have also been some slight capacitor changes in values. Generally higher capacitance's are now used. While this increases my production costs, I just had so many capacitors for various things I just had to limit the stock of stuff I was having to buy. So now the 2200uF capacitors have been replaced with 4700uF which is also used in many of my re-cap kits. This is also considerably lowered the ripple of the power supply.
Whereas the ST PSU would have 20mV ripple or less, the new design now has 7mV (0.007V) or less ripple. Tested on the 5V line were done with a 2.2amp load. (typical ST and falcon loads on stock machines are around 1.8amps)
Image below is on x10 setting, 720uV is 7.2mV. The 12V line under a 1amp load is a fraction better at about 5mV noise.
I looked at available inductors to see if there was any better ones on the market, but without much luck. One manufacture claimed better ratings than the inductor I am currently using, but it was worse tolerance and resistance wise. So I decided to stick the the inductors I am using already.
The only lower resistance ones are flat coil wound types, theres are at least twice the size and there simply isn't room on the PCB for them. Lower resistance inductors would increase efficiency slightly and run cooler. Though it would be marginal. As mentioned before with the ST PSU design, while lower resistance is always better, they are twice the size and 5 times the cost. Such inductors would add a extra £10 to the PSU price, so IMHO its not really worth the "gain" in using them.
This Falcon PSU will include the motherboard connector. Its not viable to re-use the one from the original Falcon PSU as the cables are simply not long enough. Extending the cables or changing the cables and pins would have to be done. Likely this would cause issues for people who don't have crimping tools etc. So the Falcon PSU is supplied with the motherboard connector. The crimps are gold plated! and the cable used is the thickest I can get to fit in the crimps.
Its worth noting (again) that the main system ground is actually the metalwork itself. So technically the 4 black wires to the PSU connector are not needed. Though I will be supplying the connector will all wires connected as a extra layer of saftey in case of a bad ground between the chasis and motherboard ground.
There are now more holes in the 5V,0V,12V outputs. The Falcon has more cables so the PCB needed more holes for the power connections.
This PSU design will replace my original ST PSU design in due time (No use producing 2 designs of basically the same thing). Though the Falcon design for the board will be somewhat higher due to higher PCB costs and other parts changes. The connector is not fitted for the STFM versions (like my ST PSU design was supplied). The Falcon PSU will have the connectors, but is a lot higher priced due to more work and costs involved. Though people can still buy the PSU without the connector if they want to do that themselves.
