Material from Revision 2020 seminar: Designing a better Amiga PSU

Hi everyone,

as promised in the seminar that was aired around 14:00 today, here's the material:

1) the slides - these are in PDF format, so most of you can view it.

Designing_a_better_Amiga_PSU.pdf

2) the measurement data. This is a Libre Office calc file - not sure if Excel can open it, but it's worth a try. I had to pack it, otherwise this forum software doesn't accept it.

PSU_Load_measurements.7z

I am kinda new to presenting numbers in diagrams in Calc, so the diagrams do have the weakness that the points aren't in the exact right position. This may be improved, and I am happy to take some advice on how to do that.

What the diagram does show is that the voltage of my "Prototype #2" is most stable, and ripple is the lowest in the measurement series. Further, what I did not show in the seminar (because I didn't have the diagram finished in time) is the correlation of 12V-load and 5V drop, which is in the last sub-table. THe 5V drop is not all that high, but if you take ripple and cable/filter drop into account, it may add up to taking a 3.3V regulator into oscillation. There is really no technical necessity to compensate against the 12V-load-caused 5V drop, but I figured that while I was at "compensating" anyway, I might as well compensate that.

Another thing you'll be able to see from the measurement data is efficiency: The existing solutions - including the ones that were "made in 2019 or 2020" are in the low-to-mid 60% area, while the DC-DC converter-based approach holds up nice in the 80% area, so it's actually a money-saver (though a small one).

As soon as I have a VOD link, I'll post that here.

Another question came up in the Twitch chat, and that was about measurement errors. I believe I have addressed that by mentioning the accuracy of the multimeters: The two Uni-T meters are 6000-count, so they are really +/- one digit. Both these multimeters were bought this year, so they are in calibration. The Amprobe multimeter on the 12V rail is a 2000-count, so it's only good up to the second digit after the comma if the voltage is below 20V - which it is. So we're also talking "+/- 1 digit" for that measurement. The clamp amp meter is heavily influenced by the earth's magnetic field, so it is depending on staying in the exact position that it was in when you pressed the "zero" button. The error might be up to 10mA or 10 digits. Since the current measurement only went into efficiency calculation and not the voltage observation, it only affects the resulting accuracy of the efficiency figure.

The accuracy of the Rohde&Schwarz power analyzer is +/- 50mW, and it was last calibrated in February of this year. The Agilent mixed-signal scope is actually out of calibration, but it agrees with the Uni-T 6000-count desk mutimeter down to the third digit after the comma, which is impressive, given that the last calibration is over two years ago.

While these measurement errors do add up, the important message here is that it is actually possible to compensate for all losses with a relatively simple circuit. My criticism for "new PSU"-makers is renewed and has a solid foundation with this data.

I'm happy to add other PSUs to the measurements. Feel free to send me one, or come to an Amiga show where guests are allowed to be present (after we've beaten Covid-19), and we'll do the measurements on site (I'll have at least the resistor-load and the volt meter with me). I'm more than happy if anyone replicates these measurements, which is why I am completely open about the methods. I do want everyone with the required skills to replicate these numbers.

Quote from KKR75

Do you have a specific target price you'd like to achieve ?

It would be great if I can stay below the 80,- EUR mark (that's including tax, plus shipping), which appears possible for the EU version. I'm still looking for a suitable power brick that's 110V-compatible and complies with all the new regulations. Now that USA has put efficiency requirements in place that are very similar to what Europe requires, many vendors are just "falling through the raster".

In other words: I haven't bought that Rohde&Schwarz power analyzer just for fun. I'm using it to verify the papers that vendors are including with their power bricks, and so far, I haven't found one that is 110V-compatible and fully DoE Level 6-compliant (DoE Level 6 is the American almost-equivalent to Europe's ErP Level-2). Granted, I haven't tested a huge quantity yet, but right now, it appears like Asian vendors are attempting to get rid of their old stock first.

Quote from Olli

How do you compensate for the 12V rail shift since the negative rail of the 12V supply is no longer directly connected to the negative rail of the Amiga (there is now a shunt resistor in between)?

There is no shunt in the negative rail - there are two shunts, one in the 5V rail and another in the 12V rail. The picture with the op-amp might be misleading - I do two high-side current measurements.

I actually thought about measuring the current on the GND line, but that would not be accurate enough, because you need to compensate for two line filter windings for the 5V rail, but only for a single winding of the 12V rail (the GND winding). You can't tell which part of the GND current comes from 5V and which part comes from 12V - only high-side measurements give you the required information. This of course requires two op-amps, but the standard 8-pin package mostly contains two op-amps, so the added cost is small (a few resistors and filter caps to keep the setup from oscillating).

Quote from Olli

Then the only thing I could think of that could make a difference is that the original PSUs connect the chassis of the Amiga (Shield) to PE.

In the video, your 12V PSU does not seem to have a PE connection.

Correct - these live without an earth connection.

Quote from Olli

Depending on the noise cancelling capacitor in the 12V PSU (between mains and the low voltage side), there might be some residual voltage on your shield wire.

You seem to know the topic The shield connection from the Amiga is connected with a capacitor to GND, so high frequencies are grounded and not radiated. In prototype#1, I had a normal 50V capacitor in that place, which just burned during the discharge test (remember the "failed badly" part I mentioned in the seminar?). So proto#2 uses a 1kV cap in that place (larger footprint of course), which held up fine in CE pre-compliance testing.

Quote from Olli

If users take care to connect any peripherals properly and the earthing system in their home is OK, that should not be a problem. Just something to be aware of in case of weird issues (glowing LEDs, minor tingle etc...).

That won't happen, as shield and GND have a rather direct connection inside the Amiga. It's just that the shield wire MUST NOT conduct any DC current, as otherwise the line filter (aka "common mode choke") would go into saturation, and the Amiga starts radiating EMI like hell because the inductance of the filter is going near-zero.

Quote from Verminaard

Maybe a bit off topic but I am thinking about modding a good PC PSU to use with my Amigas. I wonder if a stable PC PSU would be enough to resolve any potential issues or do we still need a better designed Amiga PSU?

Generally, PC power supplies do not have the 5V rail as their main regulation rail. It is therefore a bad idea to use a PC power supply, as the 5V rail usually does not have separate regulation. Further, PC PSUs have a very short connection from PSU to the main board, so they would never require cable drop compensation. Another question is if that PSU has low-enough ripple on the 5V rail, as most (if not all) PC gear these days does not need 5V.

Quote from FranciscoJr

I am using in my A4000 desktop, a finest and one of the best sfx psus. The Seasonic SFX 300W model ss-300sfe. 20A +5V two rails 12V 22A combined and 0.8A -12V.

So let's do some calculation:

12V@22A combined means that this rail is capable of delivering 264W

5V@20A means that the 5V rail can deliver 100W

This means that the 12V rail is the main regulation rail. The only way that this PSU is really stable for the Amiga is that you apply lots of dummy load on 12V, or that this specific PSU uses separate DC-DC converters for each rail, but given the low price and low efficiency (80plus rating), it is most likely a classic multiple-winding design. Further, with an average load of 40-50W on that 300W PSU, you're probably on the edge of acceptable efficiency. In other words: You still hear the fan spinning, because you have the conversion heat *inside* the computer.

I am also working on a solution for BigBox Amigas (A2000, A3000 and A4000), where I also use an external power brick. These are extremely efficient these days, and any heat they produce will be *outside* the computer already for obvious reasons. So the fan of the computer only needs to take care of heat that is produced inside the computer, which will be a lot less, given that 94% efficiency is easy to achieve with a modern DC-DC converter.

Quote from FranciscoJr

No problem at all. Very stable Amiga 4000.

Good for you, but definitely not transferrable to a different Amiga, as that will most certainly have different load on the two rails, which might cause votlages to go outside spec. Further, the A4000D is a "lucky draw", as it generates it's separated 5V (for video supply) on the main board. The other big boxes A3000 and A4000 require an extremely-low-ripple version of the 5V for video, otherwise you'll have a wavey picture and artifacts whenever there is a load change on 5V (and there are A LOT of such changes when using an Amiga).

Last not least, even with a converted cable, that PSU is not a plug&play solution, as it does not generate the TICK signal. That's fairly easy to get around by moving a jumper and taking the TICK signal from VSync - though some people prefer to have that separated from VSync, so the internal clock does not start deviating for different screen modes or when using a gfx card.

Quote from FranciscoJr

Is there a way to design an internal power supply for the big boxes Amigas?

Sure, but I won't do that, as the advantages of an external power brick plus the internal DC-DC converter are hard to top. It's actually making the product "possible" for my company, as I can't afford to do compliance testing for three different types of internal PSU.

Quote from FranciscoJr

Much investment.

Not only that - the added advantage of "heat outside the computer" is a good one. Of course the fan always helps, and a DC-DC converter solution with a modern high-efficiency PSU will produce less heat than a cheap PC PSU, but you can never beat heat that isn't generated inside the computer in the first place: If it's not there, you don't need to take care of it. In other words: The fan runs slower, which increases it's lifetime and puts less stress on your ears.

Quote from dbugster

but what happens with people who have no grounded power in their house Atleast outside of kitchen and bathroom

That's no problem, as the grounding is only used to reduce EMI noise (see my posting about the input filter not saturating - there's a cap between GND and shield) . The power supply itself does not have an earth pin, so nothing to worry about, as CE testing was done without extra earth wiring.