Adjustable XYS3580 SEPIC Converter

I've finally got around to the XYS3580 converter, which has been written about several times. Why not take a look? It has a universal input and up to 80 W of power. As it turns out, though, it also has plenty of "quirks" that I want to tell you about today.
Inspection, tests, oscilloscope traces, and a modest attempt at putting it to use.


I actually ordered this converter solely for a review, since several of my regular readers had already asked me about it. While I was at it, I decided to make an adjustable power supply—not necessarily the smallest one, but at least close. Unfortunately, things turned out somewhat worse than I’d originally planned. As a result, this review will follow a more chronological than logical order, so there may be a few "jumps" from one topic to another. But let’s get down to it.

The advertising promised an attractive design, digital control, and up to 80 W of power, albeit with some limitations.


The expanded description includes images of several screens as well as more complete specifications.
The key specs are:
Input voltage: 6–36 V
Output voltage: 0.6–36 V
Output current: 0–5 A
Output power: 80 W
Maximum input current: 7 A
Voltage setting resolution: 0.01 V
Current setting resolution: 0.001 A
Voltage setting accuracy: +/-0.3% +1 digit
Current setting accuracy: +/-0.4% +3 digits




The packaging is extremely basic: an unremarkable little box, with something wrapped in bubble wrap inside.


There’s nothing else besides the converter itself—not even a manual.


This enclosure form factor has probably become the standard by now. Quite a few different devices come in the same case these days: generators, converters, even electronic loads. I find it very convenient, except for the small screen.


A dimensioned drawing is provided to make installation easier.


The front panel has a 27 × 27 mm (1.44-inch) screen, a push-button encoder, and a power button. Basically, that’s the minimum needed for control; the only way to have less would be to remove the separate button, as I mentioned in my previous review.


At the back are the circuit board and a terminal block for connecting the input and output. The terminal block is cheap; I think they could have used a better one. When connecting it, keep in mind that the input and output negatives are not at the same potential and must not be connected to each other.



Cooling is handled by a small heatsink and a 30 × 30 × 10 mm fan. The fan is controlled according to the load and output current, which is certainly nice, but there’s no speed adjustment.


Since the converter uses a SEPIC topology, the board has two identical inductors (yes, I know a SEPIC can also use a single inductor), with a coupling capacitor between them.
On the other side of the heatsink are a pair of op-amps, responsible for measuring and setting both current and voltage. The output capacitors are here too: 470 µF and 220 µF, 50 V.


But I didn’t like the construction at first glance. The boards are assembled as a "sandwich," which is perfectly fine in itself, but the top board is held in place only by the connectors joining the two boards. That’s not so great, since the board comes out very easily.
I recommend securing the board somehow before using it.


The boards can be roughly divided into the power board and the control board. The power board carries the inductors, capacitors, transistors, and so on; the control board has the microcontroller, display, and encoder.
The power board has components on both sides, with quite a few on the underside as well. In the center you can see the reverse-polarity protection transistor. That could be quite useful, in my opinion, but for permanent use it’s better to bypass it.


The XL1509 powers the "brains" of the unit. As far as I can tell, the fan control transistor is to its right.
There’s also an inductor on the underside to reduce ripple. SEPIC converters have fairly high ripple, so this solution is a small point in the manufacturer’s favor. The inductor could certainly have been a bit larger.


A microcontroller, a separate voltage regulator—nothing interesting here.


Here it is fully disassembled. Unlike the power board, the control board is difficult to remove.


At the front are several standoffs, the display, encoder, and button. The button is decent and doesn't take much force to press.
Remembering the problems with DPS boards, I checked that the encoder knob was insulated. Back then, the lack of insulation caused problems due to static electricity.


The fan runs on 5 V. I don't think finding a replacement, or even a more powerful one, would be difficult if needed.


Now this is unexpected: not only is the heatsink pressed directly against the plastic component housings, there's no thermal paste either. A double whammy!


The familiar FP5139 PWM controller is installed. I came across it in my review of the SEPIC board. In short, this chip operates at frequencies up to 500 kHz. That helps reduce the size of the inductors, but makes it harder to drive the power transistors.
The transistor is a B75NF75, rated at 75 V, 80 A, 9.5 mΩ. Since the frequency is high, the PWM controller's output is boosted with an emitter follower.
To the left is a B10100G diode array, with its ratings clear from the marking: 10 A, 100 V. There's a small thermistor between the transistor and the diode array.


The soldering looks good—and the power components look almost too good. It feels like they skimped on the solder.


Before putting it back together, I couldn't resist adding thermal paste to the transistor and diode array. I also made sure the heatsink didn't get skewed during installation.


When powered on, a simple animation appears, then the converter switches to normal operation. There's no “default” screen: it remembers whichever screen was displayed before power-off. The same goes for the settings.


Board power consumption.
This is a bit odd. With the output inactive, consumption is low, around 30 mA, and doesn't depend on the input voltage—which is strange given that there's a DC-DC converter. But with the output active and set to 30 V, consumption rises to 100 mA (1.2 W) with 12 V at the input, and to a completely unreasonable 55–60 mA (almost 2 W) with 35 V at the input.

There's a 2.2 kΩ resistor across the output, and at 30 V it dissipates 0.4 W. But even so, why is consumption so high? All that 2 W ends up as heat.

And that's not all: during various transients, the board sometimes drew 3 W, or even 11 W!
The two photos below show the power supply readings, with nothing connected to the converter's output.


Controls and information display.

1. As I mentioned, there's a short animation at startup.
2. Main screen: current voltage/current/power; operating mode and board temperature on the right; set voltage and current at the bottom.
3. Auxiliary screen: accumulated capacity in Ah and Wh, plus the operating time since the last reset.
The board's input voltage and the output voltage/current are shown below.
4. Current and voltage graphs. Interesting, but not very useful on such a small screen.
5, 6. Settings.
M-PRE - Automatically start the selected memory slot.
U-SET - Default voltage setting
I-SET - Default current setting
S-LVP - Minimum input voltage threshold
S-OVP - Maximum output voltage threshold
S-OCP - Output current limit, maximum 5.2 A. It only trips if the current is over the limit when the output is enabled.
S-OPP - Maximum output power
S-OTP - Overtemperature protection threshold
S-OHP - Operating time limit
S-OAH - Capacity limit in Ah
S-OWH - Capacity limit in Wh

There's something odd about the settings. Yes, by default the converter starts with the default current and voltage settings, but mine sometimes remembered the last values I'd set and started up with those. What's more, I only noticed that it starts with the values configured in the settings while I was already writing this review.

7. Additional settings
Backlight brightness adjustment
Automatic screen-off time
Temperature sensor calibration
Color scheme settings

8, 9. By default, the brightness is set to level 3. It can be adjusted from 1 to 5; I immediately set it to 5 because the screen isn't very bright.


1, 2, 3. Setting the current and voltage is straightforward. Click the encoder to highlight the voltage or current setting (select between them by turning the encoder). Click again to choose the adjustment step: up to 1 V for voltage and 0.1 A for current. The selection exits automatically; the next click on the encoder selects whichever setting was used previously.
4. The graph is displayed, but as I wrote, its usefulness is questionable. By the way, I don't understand why cameras always render red more dimly than other colors on screens like this...
5. I was genuinely pleased to see that the output power can be set as high as 110 W!
6. I was surprised that the minimum output voltage is only 0.6 V; similar devices usually go down to zero.


Pressing and holding the power button activates the screen rotation function, which may be useful to someone.


Output voltage setting accuracy.
The specs claim 0.3%; in reality, at maximum voltage, I got 0.7%, which is quite a lot.


The allowed current error is slightly higher, at 0.4%. At maximum current, I got close to 0.5%, but you could say it passes if you allow for the error in the three least significant digits—just barely.


No, this isn't an efficiency measurement yet, just an observation.
While testing current-setting accuracy, I noticed that with an output power of 1.6 W, the power supply was drawing almost 15 W at an input voltage of 12 V. I raised the voltage to 24 V, but that didn't change much—in fact, it got even worse.
In other words, the converter was turning a full 14 W into heat!


I also noticed that while adjusting the input voltage, somewhere in the 28–32 V range, with a load on the output, the power supply would occasionally hit current limiting—that is, switch to CC mode—even though, as shown above, the power was low and my power supply's current limit was set to 5.1 A.

Anyway, I kept playing around with it and noticed the temperature had risen to about 75 degrees. During one such adjustment, the converter's screen turned white, the current drawn from the power supply dropped, and it looked like the board had shut down.
Well, I thought, that's the end of the little guy....



And sure enough, a check showed about 4 Ω across the output. The screen came back to life after about 20 minutes, but trying to turn it on put the converter into OEP error mode. As I later figured out, this means an output parameter setting error: the converter can't produce any of the values that have been set.


Knowing the topology of SEPIC converters, it's clear that this could have been caused by a failed output diode. If the transistor had failed, the board wouldn't have been able to start at all; and for a short circuit to occur in that case, the isolation capacitor would also have had to break down.


That's exactly what happened: I lifted both leads of the diode assembly, and one diode turned out to be shorted. I soldered in the good half and tried starting it up.


The board started up for 10–20 seconds, long enough for me to take a photo, then failed again with an error.


Same result: a short circuit at the output. What's curious is that the datasheet I found said this wasn't a dual-diode assembly, but a single diode. But then it shouldn't have worked after I desoldered one of the leads....
In any case, the assembly was dead.


I salvaged an MBR10100 assembly from a board from some monitor and soldered it in place of the original. The package was different, so I had to mount it vertically. Then I fitted it with a heatsink from the same monitor board.
I turned it on and checked it—everything was fine.


I came across information saying that this converter can be used to charge batteries. In fact, that was obvious just from the converter's topology. But keep in mind that there is a 2.2 kΩ resistor connected in parallel with the output, and when charging is off, it will discharge the battery.



Load tests.
The description said the minimum input voltage was 6 V. In reality, the board starts working at 5 V, but when you try to set a high output voltage, it sets it briefly and then resets.
At 6 V it works reasonably well, but I wouldn't expect to get much power out of it.


But since powering a board like this from 6 V doesn't make much sense, I started testing at 12 V, which is closer to real-world conditions.
First, I set the output to 5 V. The board shut down at almost 25 W output power. The output current limit kicked in because the current had nearly reached 5.1 A. The voltage dropped by 50 mV compared with the no-load reading.


I set the output to 30 V, and this time the board's input protection kicked in, limiting the power to 60 W. The output voltage dropped by 40 mV (the graph shows the voltage at the start of the test on the left and the final voltage on the right).


All right, I realize 12 V isn't enough for proper operation, so I raised it to 19 V, similar to the output of a laptop power supply.

There's no point testing at a 5 V output here; I hit the output current limit, so I set the output to 30 V (the maximum input voltage of my load).
This time the input current limit kicked in, but the output power reached 80 W.


When protection kicks in, different notifications are displayed. In this case, it was OPP—overpower protection.
Then I remembered that I could increase the power limit, so I set it to 100 W.


I repeated the test with the same settings. This time the converter shut down because the input current limit was exceeded, and the output power reached 92 W. That makes sense: 19 × 5.1 = 99 W, and that's before accounting for efficiency.


All right, I raised the input to 35 V, repeated the test, and got the expected 100 W.


But the settings allowed a maximum of 110 W, so I'll try setting that.


With 36 V at the input and 22 V at the output, it delivered 110 W.


Since the test was running automatically, I took a quick photo, but you can still see that it really was putting out almost 110 W.


Then the converter threw in the white flag screen again...
Honestly, that white screen drove me absolutely nuts the whole time I was testing it. Here's what I found:
1. The display comes back after about 20 minutes of "rest".
2. Most often, it happened when the built-in thermometer read around 75 degrees.
3. If you turn the board off and back on, the screen is still white, but everything works: you can turn on the output and adjust the voltage/current. You just have to do it from memory.
4. In this situation, after disconnecting and reconnecting the power and enabling the output, the board output the voltage that had been set before it was turned off, not the "default" setting. You could turn it off and on repeatedly, and it would keep remembering the last setting from before shutdown—even if you changed it while the screen was white (I had to do this to speed up the tests).
5. Sometimes the white screen appeared at a lower temperature too—just like that, out of nowhere. I tried cooling it down, but that didn't help. The only option was to wait and then power-cycle it.



While I was testing, the converter got hot even though I took breaks. Sure, you could say I removed the heatsink, but friends, let's be objective:
1. The heatsink was originally installed without thermal paste and wasn't cooling anything.
2. Adding thermal paste made no difference. Even cooling the heatsink with liquid nitrogen wouldn't have helped, since trying to dissipate heat from a plastic case is pointless.
3. The diode was moved off the board, so it wasn't heating the board, and the transistor's thermal conditions were significantly improved.

In the end, the transistor easily reached 75–80 degrees during short tests, and over 120 degrees at 100–110 W. Long-term testing was out of the question.


As it turns out, I'm not the only one with this problem. There's even a "community fix" for it. In the AliExpress comments, someone showed a modification, and I'm very grateful to them for that. But, as always, there are a few "buts":
1. Finding suitable metal strips is difficult. They need to be thick enough to conduct heat, but not so thick that they're impossible to bend.
2. Soldering them in place can also be tricky, though that's the least of the problems.
3. They need to be aligned in the same plane.
4. The trickiest problem is that the heatsink must be electrically isolated from the plates, since the component cases are not electrically connected. Also, connecting the heatsink to the transistor tab would turn it into a powerful source of interference, because the converter operates at a fairly high frequency of 500 kHz.


I got curious and decided to measure efficiency in different operating modes. Since this converter has not only an adjustable output but also a universal input, I ended up with a lot of tests. To save space, I condensed them into several graphs, grouped by output voltage.

Output: 5 V; input: 12, 19, and 35 V. The horizontal axis is in 5 W increments.

As you can see, the converter struggles to step down when there is a large difference between the input and output voltages. That’s understandable: although the SEPIC topology is versatile, it achieves this at the cost of lower efficiency.


Output: 12 V; input: 12, 19, 28, and 35 V. The results are similar: efficiency drops again at high input voltage, with 19 V being the sweet spot.
Here and in the graphs that follow, the horizontal axis is in 10 W increments. You can also see that the 12 V graph ends earlier, since in this mode the power is limited to 60 W.


Output: 24 V. Essentially, the same story as above.


I ran this test mostly as an extra: output at 30 V, input at 12 and 35 V.
Nothing unusual, except that the 12-to-30 V graph cuts off at just 20 W. Yes, I couldn’t figure out why, but the converter shuts down the output when I connect a 30 W load, even though I was able to get up to 60 W by increasing the load gradually...


Ripple.
No specific ripple limits were given, but I was already expecting it to be pretty bad—and for good reason. The thing is, this is another drawback of the SEPIC topology: its ripple is usually higher than that of conventional buck converters.

The ripple amplitude barely depended on the operating mode and was lower only in one of the combinations tested: 35–30 V at 80 W.


While taking measurements, I noticed the characteristic slope on the oscilloscope trace, and my hunch was right: in some operating modes, especially with high input voltage and low output voltage, you get this picture when you set a slower time base....


While putting this review together, I also tried to build a small adjustable power supply. For that, I needed a primary power source, which was a basic circuit based on the classic IR2153. Unfortunately, I couldn't find the D version for sale, so I had to add a diode to the high-side driver.

The circuit is as simple as it gets and has no short-circuit protection. The idea was to rely on the converter board itself, since it has input overcurrent protection, but just to be safe, I added a 5 A fuse on the output.
It would be best to use 0.22 µF capacitors for C12 and C13, but I didn't have any at home, and with the lockdown it wasn't possible to buy some quickly. The transformer is from an ATX power supply; I didn't rewind it. Most of the other components came from used LCD monitor boards.


I've had a power supply built roughly like this running in my old power supply with a 6005 board for over 5 years. That one is more powerful and has two outputs, as well as fan speed control based on a DC-DC 34063.

Though in my opinion, in this case it would be much easier to use a suitable off-the-shelf 24 V, 120–150 W power supply instead of bothering with a DIY build.


Then I sketched out an equally simple PCB from the schematic and made it using the toner transfer method and paper, as described in one of my reviews.


The goal was to make the power supply as compact as possible to fit a specific enclosure.


I simply bought an enclosure by eyeballing the size. The main thing was that the converter would fit properly.


Since there wasn't much choice in enclosures, I had to put the output terminals on the rear panel. After that, it was just a Dremel, a drill, a file, and a bit of hands-on work.


The layout is tight, but still leaves enough room for air to flow from the side vents to the fan, though I might need to add more vents.


The device is almost finished. Unfortunately, there's no way to put the terminals on the front panel, and the next enclosure size up was much too large.
This is the medium version of this enclosure: one half is tall and the other is short. I also have a version with two short halves at home. I could have bought one with two tall halves, but they didn't have that either. The only option would have been to buy two medium ones and combine them.


I still need to add a fan grille and replace the screws with black ones.


The converter needs better cooling, and I already have some ideas for that, but it works well enough as it is. Under a heavy load, it will quickly overheat and shut down. It would also be a good idea to rewind the transformer.


Now let me try to sum up everything I've said.
In my opinion, this converter has more drawbacks than advantages. It overheats, the manufacturer forgot to apply thermal paste, and the cooling design is poorly implemented in general. The ripple amplitude is high, efficiency is low, and every so often the screen goes white. Overall, it feels sluggish to use: voltage changes are somewhat smoothed out, the interface responds a little slowly, and sometimes it starts up with the default settings, while other times it restores the settings from before it was switched off.

But there are some positives, too. I personally liked the convenient output voltage and current ranges. Of course, the adjustment could have started at zero, but in my experience I rarely need less than 0.6 V.
I also liked the software features. The one thing I would have liked is the option to stop charging completely when the current drops below 1/10 of the set value—in other words, to emulate a proper battery charger.

What else could be improved to reduce heat and increase efficiency?
1. Replace the power transistor. In my opinion, there's not much to gain from this, except that it would be easier to mount on the heatsink, which is what I plan to do.
2. Use an output diode assembly rated for higher current to reduce heating and improve efficiency.
3. Remove the reverse-polarity protection transistor. This will reduce heating of the board beneath the converter’s power transistor and also improve efficiency.
4. Move the resistor connected across the output farther away to reduce board heating.
5. Replace the capacitors, especially the decoupling capacitor between the inductors, with name-brand parts with lower ESR. This may improve efficiency and reduce ripple.
6. Replace the output inductor with one rated for higher current; this may also reduce heating a little.
7. Replace the converter’s inductors with higher-current ones as well to reduce heating and improve efficiency.

Of course, an attentive reader will ask whether a converter that needs this many modifications is worth it. In my opinion, yes—the basic idea is good, but I’m not sure it makes economic sense.
The same converter is sold on AliExpress for $13.90

In general, using a SEPIC here was the wrong choice from the start. This is especially clear from the drop in efficiency at high input voltage and low output voltage. If the manufacturer had wanted to do it properly, I would have used an LTC3780. It isn’t a SEPIC, but it is a versatile converter with decent efficiency and none of the SEPIC’s drawbacks, such as its ripple level — review.

That’s all for now. I’ll probably modify the power supply when the lockdown ends, or if I find the parts I need at home—the capacitors and a suitable transistor. I’ll also need a second fan of the same size, and I definitely don’t have one of those at home :(