AT PSU mod - take two!

[Th3_uN1Qu3]

Well, here we are again. I stumbled upon this little gem at my grandparents' place. I managed to blow it up last year IIRC and pulled some parts from it, but fortunately i remembered that i took good pics, and was able to decipher the values of the missing resistors and caps. What blew was one of the switching transistors, shorted B-E. I happened to have a pair of 13009s on hand, and hey presto, it lives again.

This is as anemic as they get.

• 220uF CapXon input caps,
• no input filtering except 2 Y caps,
• no NTC,
• ridiculously small heatsinks,
• no pi filters,
• small output inductor,
• and small output caps.

On the secondary side we get

• a 10A diode on 5v,
• diodes-on-a-bracket (2x 3A) for 12v,
• -5v and -12v get a pair of 1A diodes each.

Output caps:

• 1x 1000u 16v Jun Fu for 12v,
• 2x 1000uF 10v Jun Fu for 5v,
• 1x 330u 16v Jun Fu for -12v,
• 1x 330u 10v Jun Fu for -5v.

Label claims 200W, but there's no power distribution table, let alone amp ratings! In theory the rectifier ratings add up to 156W, but i'd say it's gonna do 100W on a good day. Given this, it's pretty amazing it survived upwards of 10 years, and no caps are bloated. All secondary caps are either Jun Fu or Fulltec. Well, i'm gonna turn this pile of turd into... wait and see. Certainly, it's going to be miles away from where it currently is.

I lifted the schematic of the controller and power sections. As you can see, the only rail in the feedback loop is the 5v rail, the others (including 12v) are left to cross regulation. There's also a discrete powergood circuit and (i suppose) overvoltage protection, but i don't really care about that so i haven't bothered to trace it. I've attached a couple more schematics with comparator-based protection and powergood circuits, so you can see how the blocks fit together.

Certainly, there is no overpower protection (and it seems no short circuit protection either, but i'm not too keen to test it), so things are likely to get crispy if you push it too far.

This being an AT PSU, there is no 3.3v, and no 5vsb. So watch closely how the TL494 derives its power - it comes from the 12v rail, before the inductor. So the 494 sees the peak voltage coming from the transformer, which in this case is 24v. This is important to note, as if you want to do a radical modification involving rewinding the transformer, you absolutely must have a separate winding that delivers 30v peak at most, at maximum line voltage, or you will blow the 494.

As for how the thing actually starts up, since the 494 is on the secondary side and there's no auxiliary supply to power it, look closely in the primary and you'll find it.

More to come, including a clever way of varying the output voltage, which allows more flexibility than modifying the feedback divider of the error amp on pin 1.

[tom66]

Re: AT PSU mod - take two!

My guess is the 220k resistor turns on the first transistor (provides pulse from mains rectified input), providing an initial pulse to the transformer, this charges up the 330u and the converter starts switching which then keeps the converter running.

That supply has less bulk than some 60W plug packs I have! Wow.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

There's a typo in your post - not sure which 330u you're talking about, maybe you wanted to say the 47u for the controller Vcc. That's it mostly, with another little detail - this wouldn't have been possible in a half bridge without the configuration of the driver transformer.

In all half bridge supplies with bipolar power devices, it has a winding that is in series with the main transformer primary. What happens is when the first transistor turns on, there is a current going thru the driver transformer as well. Since only one transistor is on, a few us later the driver transformer saturates, its magnetic field collapses, which cuts base drive from the first transistor and switches on the other one. So the supply enters a self-oscillation mode, at a fairly low frequency (since there is very little base drive coming from the 220k resistor).

This is what causes the characteristic buzz when an AT supply is in protection mode. Even at this low frequency, the voltage produced is enough to keep the TL494 running, so if a protection was triggered, it will not resume normal operation until it is unplugged and the primary caps have drained.

[tom66]

Re: AT PSU mod - take two!

I was looking at the Seventeam schematic, but the operation is very similar. Interestingly, many cheap self-resonant CFLs use a different way of starting by using a diac to provide a pulse to one transistor, which then pulses the opposite one (through the transformer), and the oscillation begins. I imagine they could do this, but it ends up cheaper to use the driver transformer's saturation properties.

Re Seventeam schematic: What's the point of the 220k from the emittter of the top transistor to the collector of the bottom one?

Like the fake safety Y-caps, too.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

I wound the voltage up till i got 15v on 12v (had to stop there because of 16v caps). In the meantime, 5v got to 6.63v. The thing didn't shut down, so we can about sum up the protections this thing has: none. So the 4-transistor circuit is just a power good signal generator, well that's good news to hear, as i won't be wasting any time on it and just removing the whole lot.

On the low side, i got as little as 200mV stable on the 5v rail, if providing a 0.5A load. With nothing more than the minimum load resistors and two fans on the 12v rail, the minimum stable output voltage is 1.83v on 5v, 4.10v on 12v. And that's because the base drive gets so low that the self-oscillating circuit takes over, if this were an ATX PSU i would have gotten even lower output voltages! Usually they start making nasty sounds with any less than 10v on the 12v rail.

Here's the trick that i used. The "regular" and well-known way of altering the output voltage of a SMPS is to replace the bottom resistor of the voltage divider going from the output to ground, feeding the input of the error amp, with a pot. This however, has one major drawback: it affects the gain of the PSU's Bode plot. Increasing the output voltage via this method lowers the gain (which doesn't usually cause trouble), while decreasing it increases the gain, which brings up the crossover frequency, and will cause oscillation sooner or later. Like i said above, on a 12v output, you cannot go much lower than 10v with this method, without a significant load (which lowers the voltage on the primary caps, hence in effect, drops the gain back down).

You know that in TL494-based PC PSUs, they use a 2.5v reference, since otherwise there would be no way to stabilize the 5v rail. But the TL494 only has a 5v reference! So they use a resistor divider to make 2.5v. You're going to have one resistor coming from Vref to pin 2 of the 494, with another identical value resistor going to ground.

Now, do this.

• Cut the trace coming from this divider to pin 2 of the 494.
• Bridge the gap with a 10k resistor, so that now you have a 10k series resistor from the voltage divider, to pin 2 of the 494.
• Then replace the bottom resistor in the divider with a pot. I suggest a 10 times larger value.

This will allow you to go from ~1 volt output, to nearly double the original output voltage, without affecting the stability of the supply! Of course, any existing protection circuitry should be reconfigured prior to attempting this mod.

As for what it really does and why it works in this "magical" way, you are in effect varying the reference voltage input to the TL494. Since the feedback divider is left alone, the gain of the Bode plot will not change and the supply will be stable over a wide range of output voltages.

As for what the 10k resistor in series from the divider does, and why i asked you to cut the trace and put it in, there are two reasons. First, it presents a constant impedance to the input of the error amp, to keep the compensation working correctly. Otherwise there would be issues at very low output voltages, when the pot resistance approaches zero. Secondly, it eliminates the "hand effect" - touching the pot to adjust the voltage will no longer make the supply oscillate.

You can also supply your own reference voltage, say you want to control the output of the SMPS from a microcontroller, just remember to have the series resistor, otherwise the compensation won't work. It can be any value between 10k and 100k.

[b700029]

Re: AT PSU mod - take two!

This is what causes the characteristic buzz when an AT supply is in protection mode. Even at this low frequency, the voltage produced is enough to keep the TL494 running, so if a protection was triggered, it will not resume normal operation until it is unplugged and the primary caps have drained.

Would you say that almost all AT supplies use the same design or are nearly identical? All the ones I've had the misfortune of shorting would buzz like that and twitch the fan until powered off manually.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

Would you say that almost all AT supplies use the same design or are nearly identical?

Yes. There's no other way to do it anyway, at least not with a secondary side controller like the TL494 is, and i've yet to see an AT PSU based on another chip.

[tom66]

Re: AT PSU mod - take two!

They could be primary side controlled for low loads, e.g. by UC3842. I think I have an old ~60W AT PSU somewhere, maybe I should open it (not sure if it went in the bin though.)

[momaka]

Re: AT PSU mod - take two!

They could be primary side controlled for low loads, e.g. by UC3842. I think I have an old ~60W AT PSU somewhere, maybe I should open it (not sure if it went in the bin though.)

Indeed.
I have a 75W Mitac AT PSU like that. At first, I thought it was dead since it wouldn't start. Turns out it just needed a large load on the 5V rail (and when I say large, two 5V, 5W light bulbs weren't enough). When I connected it to an AT motherboard, it worked great.

Cool mod with the AT PSU. I think I'll try the voltage mod with one of my crappy half-bridge ATX PSUs. I have a feeling the results will be interesting

I also have another half-bridge ATX PSU that I think is oscillating. Do you think that if I do the 10k resistor mod it will stop oscillating? It has a SG6105 PWM controller, though.

[PeteS in CA]

Re: AT PSU mod - take two!

tom and Th3 called it. The start-up comes with the 220K turning on the first transistor, the main transformer current going through a winding on the base drive transformer creating positive feedback ("proportional drive") and the magnetic field flipping at core saturation, turning off the first while turning on the second transistor, self-oscillating until the TL494 and driver transistors have the Vcc to be able to take control, switching at a much higher frequency than the natural self-oscillating frequency. Besides the buzz Th3 pointed out during power limit protection, you may also see interesting stuff when turning off the P/S (the P/S "tries" to turn on during the I/P 'lytic voltage bleed-off, resulting in significant voltage pulse or train of pulses on the +5V - a Mitac ATX supply I saw would actually have a >6V pulse on the +5V).

[PeteS in CA]

Re: AT PSU mod - take two!

They could be primary side controlled for low loads, e.g. by UC3842.

The start-up components - a high-value resistor and a small lytic capacitor - are usual suspects in 384x-based P/Ss that won't power on. The resistor opens up or the cap won't hold/source enough charge, and the 384x can't start-up. Besides being able to work on the primary side, a UC384x is less ancient and can have a faster, better controlled loop than a TL494 or SG3524. The SG3524 and TL494 were voltage-mode PWMs designed in the 70s; the UC384x PWM family is current mode and produced in the 80s.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

I also have another half-bridge ATX PSU that I think is oscillating. Do you think that if I do the 10k resistor mod it will stop oscillating? It has a SG6105 PWM controller, though.

SG6105 is TL494 + 2x TL431 + some comparators, so it's pretty much the same thing. What makes you think it's oscillating? Does it whine to various degrees, depending on the load? Can you draw a schematic of the components around the SG6105? You can find a reference design in its datasheet, i doubt it'll vary much from that.

Putting a resistor between the reference divider and the inverting input of the error amp is to stop the interference from your hand getting into the controller when you are using a pot to vary the output voltage, not in any other case.

@ PeteS in CA: I've seen this behavior (output voltage spike when unplugged from mains) in all AT and ATX PSUs. So it does not appear to depend on the self-oscillating startup of AT PSUs.

[b700029]

Re: AT PSU mod - take two!

Besides the buzz Th3 pointed out during power limit protection, you may also see interesting stuff when turning off the P/S (the P/S "tries" to turn on during the I/P 'lytic voltage bleed-off, resulting in significant voltage pulse or train of pulses on the +5V - a Mitac ATX supply I saw would actually have a >6V pulse on the +5V).

I guess that's the reason for the fan twitching when turning off from protection mode, but all the AT supplies I've seen have the output just die off otherwise when turning off normally.

[PeteS in CA]

Re: AT PSU mod - take two!

The P/S I referred to was an engineering prototype built by Mitac for Sun Microsystems. We reported the problems (the +5V had at least one pulse high enough to possibly do damage, and one or two other pulses that could turn logic on in an uncontrolled fashion) to Mitac, and they fixed the problem without totally redesigning the P/S. So self-oscillation can happen at turn-off with a proportional drive base drive design, but it doesn't have to happen. I should add that the restart attempts would happen several tenths of seconds after turning off power, so unless one had a 'scope set to a slow sweep speed, the pulses might happen but not be detected.

[tom66]

Re: AT PSU mod - take two!

Many ATX PSU's I have encountered will have a slight (~0.5V) voltage spike before they shut down.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

Well, i'm back home, and with more details about this PSU (i brought it home with me).

The common output inductor is wound on a T90-26 toroid as far as it seems, and it has the following values: ~8uH on 5v and -5v, ~28uH on 12v and -12v. My meter isn't very accurate in the 100uH range, so take those numbers as +/-5%. I noted the number of turns somewhere, but forgot and i'm too lazy to count them again. It's 10 turns for 5v/-5v and 24 turns for 12v/-12v IIRC. You know the deal - the common inductor also acts as a transformer, so to ensure group regulation, the turns ratio of the windings must be equal to the voltages ratio, and you can see that in the above numbers.

To be noted is that on a common core, putting two windings in parallel will NOT decrease the inductance by the parallel inductor formula (same as for parallel resistors, 1/Ltotal = 1/L2 + 1/L3 + ... + 1/Ln), it will remain the same, only with higher current capability. Practically all computer PSUs use multiple paralleled windings on 5v, and the good ones do it on 12v too.
Putting windings in series WILL increase the inductance as per the formula (Ltotal = L1+L2+...), while maximum current capability remains the same it would be for the thinnest wire of the series windings (in theory). In practice maximum current decreases because of the added resistance of the wire which causes higher copper loss.

What i will do now is move the feedback on the 12v rail and recalculate the compensation components and see what i get, to check if the formulas i am using are correct. Best to start off with some assurance before i get into the crazy mods.

[tom66]

Re: AT PSU mod - take two!

So if I wind two windings on the toroid at say 8µH very close to each other, I should measure 8µH on each.

Then I put them in parallel, but I still get 8µH. Is it due to the core that the inductance doesn't halve?

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

Yes exactly. Because the core is the same, you have in effect a transformer, so when you are paralleling multiple windings with the same number of turns you get higher current handling while the inductance stays the same. Also note that you cannot parallel windings with different number of turns on a common core - else the one with less turns will short the other.

Edit: Now let's get dirty and get into the compensation.

First, you will notice that unlike textbook examples of type II and type III, the compensation most commonly found in AT and ATX supplies is simply a resistor with a cap in series. I bet most of you have wondered about that. The reason for this is that feedback on group regulated AT/ATX PSUs is always 5v dominant, and the Q (peak) of the LC filter is small, because we have a small inductor with a large capacitance. In addition to that, the ESR of the capacitor is chosen so that its zero compensates the pole of the inductor, making the overall phase shift 90 degrees. This is why it is not recommended to use ultra low ESR capacitors in a power supply.

With just 90 degrees additional phase shift in the negative feedback loop, the system is stable by itself! The only reason the RC is there, is to limit the high frequency gain of the TL494, ensuring the right crossover frequency. It is not a "compensation" circuit per se.

Attached is an adapted version of a LTSpice simulation i found sometime last year on diyaudio, with the opamp model tweaked to represent a TL494 error amp. The original simulation used a LM358 which wasn't accurate in some aspects. There you can see the frequency and phase response of the LC filter, Take note, to see the correct phase you must have the output trace inverted, as in, -V(EAout).

As you can see, at light load, this thing doesn't even respect the Nyquist criteria! Yet, it works... At nominal load the input voltage to the LC filter will decrease since the primary voltage decreases as well, and the gain will drop, making the crossover frequency lower.

In this power supply, at nominal line and minimum load, we had 12v peak input voltage to the 5v diodes, and 29v for the 12v diodes. So we'd have modulator gain of 4 for the 5v rail, and 9.7 for the 12v rail. Feedback was taken from 5v only. How this gain is calculated is explained in the simulation. Now, i will attempt to modify this for 12v-only feedback, as an exercise.

[Th3_uN1Qu3]

Re: AT PSU mod - take two!

Have successfully modified the feedback for 12v only, works fine. Don't have the file right now, as my dv9000 kicked the bucket for good last week, and i don't have a SATA to USB converter.

Anyway. It's about time to start picking parts for the final configuration of this power supply. I want an adjustable split supply from +/-20v to +/-70v at a maximum power of 650W, adjustable current limit 0 to 5A, 50mVp-p maximum ripple.

By my SMPS cheat sheet i am going to need a 2x250uH coupled inductor. I will be making this on a ferrite core, most likely an EI-33. The secondary caps don't need to be more than 1000uF, i will go with Rubycon ZL. I'll also cheat a little and use 63v caps, relying on the surge rating - 100v ones aren't cheap.

Next, i have to pick my diodes. Because the peak voltage at the transformer is going to be a bit over 200v at high line and no load, i would need at least 250v diodes. Common anode diodes are less often found than common cathode, my search at Farnell revealed only one suitable candidate: the Fairchild FEP16JTA. Unfortunately they don't stock the 400v version in common anode. So, for the positive rail, i will be using its common cathode counterpart, the FEP16JT.

This will take a little bite from the efficiency of the power supply as the 600v versions have 1.5v forward voltage as opposed to 1.3v for the 400v, but considering the currents i will be working with, this won't be a problem. 1,5v*5A*2 = 15 watts total dissipation on the secondary heatsink at maximum load, which is quite lax. The switching transistors will dissipate 60 to 80W, so clearly, the dissipation of the diodes is no issue.

[Behemot]

Re: AT PSU mod - take two!

Besides the buzz Th3 pointed out during power limit protection, you may also see interesting stuff when turning off the P/S (the P/S "tries" to turn on during the I/P 'lytic voltage bleed-off, resulting in significant voltage pulse or train of pulses on the +5V - a Mitac ATX supply I saw would actually have a >6V pulse on the +5V).

Yep, I see that in one Enlight Corporation AT supply. Didn't measure +5 V, but +12 V gets as high as over 13 V with this spike. You can clearly hear fans getting faster for a fraction of second.