In late 2008 I needed to buy a high power true sine wave inverter, for an alternative energy scheme. Scared away the very poor quality I had seen too many times when asked to repair Chinese inverters, but not willing to pay the outrageous prices asked by manufacturers who are known for their quality products, I settled for a Taiwanese made inverter. It was the Power Jack PSW3500, which is rated at 230V 50Hz output, 12Vdc input (at several hundred amperes, of course!), 3500W continuous output, 7000W surge capability, with a true sine wave output. In addition, it has a built-in 50A charger, and the necessary detection and switch-over circuitry to make it work as an UPS, if desired. The price seemed right, and Power Jack made a lot of advertising telling that their product was much better than the typical Chinese ones. I bought that, and I bought the inverter. And I got cheated real good. It has often been said that one gets what one pays for. Not so! I paid good, real, decent money for this inverter, and got a piece of crap, if you excuse the expression, which needed a complete re-engineering and reconstruction to become usable.
I suffered the loss, and now at least through this web page I'll get the satisfaction of trying to help others not to fall in the same trap.
When my Power Jack inverter arrived, I read the instruction leaflet, which is very brief, and shows how to hook it up, in any of its three functions: As standalone inverter, as standalone charger, and as UPS. I intended to use it only as standalone inverter. I hooked it up, using very heavy wire for the battery connections. After all, a 7kW surge will take almost 700 amperes from the battery bank! I switched it on, and measured the output voltage, waveform, etc. The voltage was fine, the waveform was a bit strange: One half cycle of the sine wave was clean, while the other had about 20 volts of high frequency noise riding on it! Also, the inverter produced an impressive amount of radio interference, even with nothing connected to the output! So much, that it completely erased all radio reception. Input current consumption was less than 2 amperes, which seemed fine. But the current was pulsed: Short, high current pulses followed by pauses, averaging out to less than 2A. This is called "hiccup mode". I thought it was by design, so I didn't worry. I did a load test, connecting a 2kW heater to the inverter for just a few seconds, because my battery wasn't very good and I didn't want to tax it too much. The output voltage dropped a little, but not too much. All seemed fine. The only really strange thing was that the LED power meter on the inverter stayed fixed at 700 watts, regardless of how much power really was being consumed! I ended the testing there, and wrote to the seller, who told me that that was normal, and that the power meter was there "only for show" (his exact words!). Oh well... I don't need a power meter... And since the inverter arrived well and was working, I gave the seller positive feedback (this was on ebay).
A few days later I wired up the whole system: A large, good battery, the alternative energy system charging the battery, regulated to 13.8V, the inverter, and connected it to feed my house. There was very little consumption at that moment, not more than 10 watts or so. The inverter happily idled along, taking about 2.5A at 12V. I had to do some woodworking, so I went to get my tools, which include a 1400W circular saw, a 600W planer, and a few more. When I had everything set up and switched on the workbench light to start working, nothing happened. Strange. I went back to where the inverter was, and found the room full of white smoke, awful smell, the inverter giving zero output, but still with its LEDs telling that all was fine: The green "INV" LED was lit, the power meter bargraph was showing the "normal" 700 watts, and the battery voltage meter was at 13V. But no output, and lots of smoke... The nice shiny new Power Jack inverter had commited suicide, while running at almost no load, whithin the first two hours of operation!
I first contacted the seller again, telling what had happened. Since he took some time to reply, in the meantime I opened the inverter, to have a look at the damage.
The shell comes apart in two halves. The bottom half, shown in the upper place here, contains just the power components of the push-pull DC-DC converter, which steps up the 12V to about 340V or so. It consists of six converter groups, each having a transformer, two pairs of MOSFETs, and assorted driving, bypassing and snubbering components. The secondaries of the transformers are connected in series in two groups of three. Each group has a bridge rectifier, and an LC filter. The outputs of the two groups are connected in series. No voltage balancing is employed at all. And there is no current sensor whatsoever!
The other big board is what I will call the mainboard. It carries the sine wave chopper, which uses an IGBT bridge, and an LC output filter with two big inductors. There is also the charger circuit, all the switchover system, and a plug-in control board, which controls the DC-DC converter, the switchover, the charger, and drives the chopper.
I immediately noticed two popped electrolytic capacitors, one of which you can see in this photo. It's a 100uF, 400V capacitor, which is THE ONLY capacitor across the high voltage DC bus on the sine wave chopper board! It has to take the full high frequency ripple current, which can easily be several amperes. And the ripple current rating for such a capacitor is not over one ampere. Was that the reason for it popping open?
And here is the other one. This is the filter capacitor used in one of the two series-connected rectifier/filter group, that produce the high voltage DC. Its companion was still fine, it seemed.
At this point, it looked like some big overvoltage had happened, lasting a good while, because electrolytic caps don't pop on short overloads.
This is one of the connectors linking the high voltage DC from the converter board to the main board. It has spills of capacitor juice on it.
As I looked more closely, I started finding more and more horrible things! Here, for example, is a screw, that came loose during shipping, and got caught in this place. Having been caught, at least it didn't cause any short circuit. But screws coming loose inside equipment are a very bad thing!
Here is a spacer, that came loose when its screw fell out. The spacer got stuck in the white thermal grease, and stayed put there.
Hmm... why is there any thermal grease at all? There is a silicone rubber pad between the power components and the heatsink! This makes thermal grease unnecessary, one would think!
After picking off the spacer, and looking more closely, it became clear: The pressure bar holding these parts against the heatsink was installed much too low, pressing down only on the lower edge of the TO-220 parts! As a result, these lower areas get buried in the soft rubber, and the tops of the parts stick out in free air, with no thermal contact whatsoever!
Instead of fixing the problem, Power Jack "engineers" chose to smear thermal grease all over the flanges! If at least they had smeared the grease INTO the voids, that might have worked, at least a little bit, but smearing the grease OVER the parts and rubber pad is totally useless.
And here is yet another photo showing how these people tried to solve the problem of lousy thermal contact! There is not a trace of grease between the parts and the rubber. There is just air there. The grease is outside the voids.
I then did the complete troubleshooting, and found that all 24 MOSFETs were burned, with the sources open and the drains shorted to the gates. The fuses were fine, thank you. Along with the MOSFETs, the 24 gate resistors also failed. Then there are the two popped electrolytic caps, and also two overvoltage protector diodes (transil). These failed in short circuit, on the side where the capacitor still was fine.
Since it was too expensive to ship the inverter back to Taiwan for repair, and then again to Chile, and the whole import/export would be a real hassle, to avoid paying the import taxes a second time, I proposed that Power Jack simply sent me the required spare parts. They agreed, and sent them very quickly. So I proceeded to disassemble much of the inverter, and replace the bad parts.
At this point, at least part of the failure mechanism became clear: For some reason the DC link voltage had soared to a far excessive value, and for a good time. For that reason, the capacitors popped. But popped capacitors typically keep working for a while! So did these. When the voltage soared further, one of the overvoltage protectors must have failed from stress, while trying to absorb those pulses. It created a short. This overstressed and shorted the other protector, so that now there was a dead short across one set of three transformers. Since there is no current sensing whatsoever, this took out the MOSFETS of that side, before the fuses could react. And the 12V leaking back from the bad MOSFETs, through the shared drive circuitry, to the other three converters, put those MOSFETS into continuous conduction, taking them out just like their companions. An uncontrolled chain reaction like the worst. Fortunately it wasn't nuclear! And the control circuit, dumb and uninformed as it is, didn't notice a thing, and displayed on the LEDs that everything was fine! Is that poor design? What do you think?
With the new components installed, the inverter again worked, but I was hesitant to use it like that. It would have burned out again very soon. Instead, I stored it away, and spend the next several months doing more important work, using a stinky and noisy gasoline powered genset instead of the inverter. So much for alternative energy. At least the genset works.
But in July 2009 I took up the inverter again. I decided to completely check it up, understand it in detail, and cure all its problems, before daring to use it again. And here started one of the most frustrating episodes of my career as electronic engineer! I learned that when a design is bad, it is BAD, really, and when you think you have found all the bugs, and can start fixing them, you find even more!
Instead of continuing to tell the story in chronological order, I will list the flaws, one by one. That will take a lot of time and space, I assure you.
Normally any switching converter uses current sensing. The better converters employ the current sensor to limit the pulse width on a cycle-by-cycle basis. This allows fast, precise and safe control of the current flowing through the system. Simpler systems do not use cycle-by-cycle limiting, but at least they sense the average current, and reduce the duty cycle when the current is reaching the acceptable limit, to avoid burning out components if something abnormal happens. The Power Jack inverter has none. If this is due to cheapness or incompetence, I don't know.
The DC-DC converter in this inverter uses an SG3525 control IC. Its error amplifier was configured as a plain, simple integrator: A single capacitor as the feedback path. This results in the error amplifier having a constant -90 degree phase shift. But in the power section of the DC-DC converter, the dominant elements are the filter capacitance, and the input source resistance (batteries, cables, MOSFETs, transformers), giving the power system also a -90 degree phase shift throughout the low frequency range! This adds up to -180 degrees, which together with the inverting nature of any control loop produces in-phase (positive) feedback. In this case the positive feedback happens from less than 10Hz to over 1kHz. The limits vary depending on the load on the output, which is the third element influencing the power circuit's response! The filter inductance comes last in order of importance. It's much too small to have any significant effect.
With positive feedback, and positive loop gain, the stage is set for self-oscillations to occur. And this was the cause of the hiccupping I noticed. It was NOT by design. It was by poor design!
The control loop is unstable under almost all load conditions, resulting in the inverter drawing extremely high current pulses, followed by pauses, with the frequency and duty cycle given mostly by the load on the output. I measured pulses in the kiloampere range and millisecond duration, while the average current was just a few tens of amperes!
Push-pull converters like these are prone to core saturation and consequent failure from flux walking. To make sure no damaging saturation will happen, the only way is to use pulse by pulse current limiting - which wasn't done in this inverter. Another method, not 100% safe but good enough in most cases, is making the error amplifier so slow that the lengths of subsequent pulses cannot change much, and then rely on the MOSFET's and circuit's resistances to balance out any remaining asymmetry. This was also not done in this inverter! The error amplifier has a gain bandwidth product close to 20kHz! As a result, this circuit is prone to core saturation and MOSFET destruction whenever there is significant coupling of the signal on the transformers into the feedback loop!
The high voltage filter uses two filter chokes, each consisting of 30 turns wound on a yellow and white toroid about 27mm in diameter. This seems to be an Amidon/Micrometals T-106-26, or an equivalent. Assuming it is the mentioned type of toroid, the inductance would be 81uH while the current is low. But the highest current these chokes have to take, at the waveform peak when a 3500W load is connected, is 21.5A, and at this current level the cores are already deep into saturation, with their effective permeability being about 1/4 of the original value, and the inductance having dropped to just 20uH! And at the rated 7000W surge power, the situation is even much worse.
This is just the beginning of the problem. At 81uH, the ripple current in these inductors is 16A p-p, which would be roughly 6A rms. This ripple current will theoretically increase to 24A rms, or well over 118A p-p, at full load! It follows that this DC-DC converter never enters continuous current mode operation at all! It's always in discontinuous mode, with the peak current in the diodes, MOSFETs, etc, over twice the average cycle current! In reality, at 21.5A output current, the inductor current will be swinging between zero and about 50A all the time, which means a peak current of over 160A in each single MOSFET, plus any magnetizing current for the transformers!
I leave it to the imagination of the reader to guess how long the filter capacitors will live, when they are subjected to an average of maybe 15A of ripple current, while their total combined ripple current rating is about 4A maximum.
A quick calculation shows that these poor toroids have to work at an AC flux density of 0.35 Tesla. And this doesn't change very much over the power range. Even at a relatively low load power, the AC flux density is already almost that high, and it remains there, as the current continues to increase, while the cores increasingly saturate. The problem is that this high AC flux density, at the high frequency this converter operates on, produce a whopping 53 watts of power loss in each core! That's enough to make these coils unsolder themselves from the circuit, and probably to burn the insulation and/or meet the Curie temperature!
While testing the inverter at very low power (just 100 watts, less than 3% of the rated output), where the inductors are not yet close to the full AC flux density they reach at higher powers, in a matter of one minute they got too hot to touch, and after five minutes they started discoloring, and smelling burned!
To operate correctly, this inverter needs toroidal cores many times larger than these, so that they can be wound to provide a few hundred microhenries, and not saturate until at least 50A, to enable them to handle the rated 7kW surge load too. This would also make them work at much lower AC flux density, producing acceptable power loss and heating. Alas, Power Jack in Taiwan was too cheap or inept to do that.
The feedback signal of the DC-DC converter comes through an optoisolator, to afford the required galvanic insulation between the DC link, which is at output line potential, and the control circuit, which is at battery potential. In principle this is fine. The problem is that the designer of this inverter apparently didn't know that optoisolators have a strong temperature coefficient! When they warm up, their current transfer rate goes down markedly. And in this Power Jack inverter, the optoisolator is used in such a way that its drift directly affects the "regulated" DC link voltage! As the optoisolator warms up, the DC high voltage goes up, and up, and up, and.... BANG! This is one very likely trigger for the chain reaction that burned out my inverter before two hours were over! To compound the problem, this optoisolator sits in the top of the case, in a pretty warm spot.
Good design practices dictate that circuits should be made so that in the event of a malfunction they will shut down without further damage, whenever possible. Unfortunately the Power Jack inverter is made in the opposite way. The control loop uses an auxiliary power supply, which is derived from another auxiliary power supply, both of which use SMT components stressed beyond their absolute maximum ratings. Specially the ripple current rating of the small electrolytic caps is far exceeded. And if any of these two auxiliary power supplies fails, the feedback loop is broken, in such a way that the DC high voltage will soar far above the capacitor and transient protector ratings. BANG!
Also, the high voltage is sensed through three tiny SMT resistors in series. Each of these resistors is rated at 100mW. One is 360k, another 200k, and the third is just 20k. The DC voltage is about 340V. This results in a total power dissipation of 200mW. If the three resistors were of the same value, this would mean 67mW on each, which is a bit tight for comfort, but OK. But with the values used, the 360k resistor works at 124mW, well above its absolute maximum rating! If it fails, the feedback loop is opened, the voltage soars, and BANG!
Selecting such different values for the three resistors in series is not a matter of cheapness. It's plain, simple, gross incompetence.
Optoisolators are less prone to failing, and this one is used in the low area of its rating, so it should be pretty safe. But anyway, if it fails open, which is the most common failure mode, BANG, again!
The unsafe feedback loop, using an optoisolator so that it causes severe voltage drift, would in itself warrant the inclusion of an independent overvoltage protection circuit: Anything that in the event of the voltage rising to a value that is dangerous to the inverter or to the devices powered by it, would activate a safe shutdown mechanism. But in this inverter, there is even more reason to use it: The filter capacitance on the DC link is so small, that even a quite moderately reactive load on the output would drive up the voltage on the DC link to a dangerous level! Sure, a cheap inverter doesn't HAVE to handle highly reactive loads. But if the load is too reactive to handle safely, it should shut down, instead of exploding! Well, this one explodes. It would be simple to include an overvoltage detection circuit that shuts down the inverter in the event of overvoltage. This would provide protection both in the event of a control circuit problem, or an overly reactive load. And on top of all that, don't forget that this inverter's DC-DC converter has two sections connected in series, and each of these sections consists of three converter units, and each of these six converters is individually fused! If the fuse for just TWO converters open from overload, which is a perfectly feasible scenario, the DC-DC converter would end up badly unbalanced, with one group providing 3 times as much voltage as the other. With the total voltage regulated to 340V, and each filter capacitor rated at 200V, this would cause 255V to show up on one of the 200V capacitors: BANG! But no, despite all this, Power Jack people were too cheap to include any overvoltage shutdown. The necessary components to afford full protection cost about two dollars. Not including them is either being outstandingly cheap, or being incompetent.
This inverter uses an H-bridge of four IGBTs, with pulse width modulation at 20kHz, to produce the sine wave output. The output filter consists of two really large toroidal inductors, and two film capacitors. All this seems good and fine to me. But when looking at the board, which you can see in the first photo on this page, I would never have thought that The Power Jack engineers chose to place BOTH inductors on the SAME side of the bridge's output! Of course, on a piece of paper, it doesn't matter whether both are on the same side, or each on one side. Electrically, they end up in series anyway. In practice, however, it makes a whole world of a difference, in terms of interference to other devices! As it is in this inverter, one side of the chopper appears directly connected to the output, and the negative side of the DC bus is capacitively coupled to the negative of the battery. This results in 340V of square wave 20kHz riding on the output during one semicycle, relative to the battery circuit, which is typically grounded! Talk about a big noise bomb!
And yes, there is an EMI filter in the output line. Unfortunately, it is configured the wrong way, and so it's completely useless: The small ground bypass capacitors were installed on the side of the filter that goes to the chopper, rather than on the output side! They surely blew up when at the lab they actually connected them to ground, because that results in a powerful 340V 20kHz signal connected straight to two little ceramic caps: The one from the "hot" bridge side to ground, and the one from the negative DC link side to ground!
Do you want to know which was Power Jack's solution to this problem? Simple: They left the ground connection of the EMI filter disconnected!!! The result: Absolutely impressive, all-overwhelming radio interference! This is particularly bad because many people buy sine wave inverters to feed sensitive studio audio equipment, or laboratory instrumentation, with a clean AC supply! Ha ha!
I wouldn't have written this web page bashing on Power Jack, if their product was simply unuseable. But when I noticed the gross, extremely dangerous blunter they did by joining the AC input to the output, I decided I had to write this page! I hope that nobody has been electrocuted yet by one of these inverters!
As explained in the beginning, this inverter includes a battery charger, and switchover circuitry for UPS (Uninterruptible Power Supply) service. So there is an AC input and an AC output. The AC output is switched between the AC input, and the inverter output, and in addition the AC input goes to the charger. So far, so good. The problem is that the geniuses at Power Jack chose to use a single pole relay to switch just ONE side of the 220V line, while the other side of the 220V remains connected at all times, to the input, output, charger and inverter! As a result, when using this thing as a standalone inverter, one of the 220V output poles is fully exposed at the male AC input plug!!!
The concept of this was probably to use this permanently connected side as the neutral, but that doesn't work out, because the input and output connections don't use polarized connectors. The output universal jacks can be connected in any polarity, while the input connection comes from Taiwan just as a stripped wire, to which the user has to install a plug of the kind used in his country. And the manual has not even one word of warning about phase, neutral and the like!
So it's perfectly possible that someone connects this inverter to his house, leaving the input disconnected, and uses a standard plug to mate with the jack on the inverter. The house wiring will normally have the neutral grounded. Depending on which way this plug is put in, one pin of the inverter's 220V input plug will have either the neutral, or the phase on one of its pins, exposed for anyone to touch!
Folks, this is more than being cheap, or being incompetent: It wouldn't be hard for a lawyer to bring up charges for attempted manslaughter, or, if someone actually gets killed, erase the word "attempted"! I can only hope that each of these Power Jack inverters burns out quickly enough to become "safe", before getting a chance to kill anybody. And I'm glad I didn't touch the exposed male plug on mine, in the short while it worked!
After this major scare, let's go back to what may look like petty issues.
The charger is a simple half-bridge circuit. It follows the trend of crappy Chinese PC power supplies, and has absolutely no EMI filtering whatsoever! The AC input goes via an inrush current limiter directly to the rectifier, filter caps, and from there to the IGBTs. As a result, the charger causes severe radio interference. Not as brutal as the one the sine wave chopper causes, but still severe enough to disrupt HF and MF radio reception, and even cause strong "snow" on a VHF TV.
The control circuit for this charger has both voltage and current regulation, which is fine. The problem is that the control loop is misdesigned, resulting in instability when operating in current-limited mode. The charger pulses heavily, and if left doing that for long enough, it would fail. It requires re-design.
Here you can see an oscillogram of the charger's output voltage. The main square wave happens at about 160Hz, and the triangle wave on the upper side of the square wave is at roughly 3kHz. The "fog" around the waveform is noise fed through from the main 100kHz switching process. As you can see, the charger's control loop oscillates at two frequencies at the same time, with some modulation on each, and on top of that there is about a half volt of switching frequency noise! How's that for crappy design?
It looks like the designer copied the control loop from some application note or other source, without understanding it, and not realizing that it was meant for a power supply, not for a battery charger with its immense de-stabilizing "capacitive" load attached!
Near the beginning of the page I showed you photos of how the TO220 cases were mounted such that their tops separate from the thermal pad, and how these people smeared thermal grease over it, in a vain attempt to fix the problem. Well, here is a top-down view of one of the larger parts. These have their mounting bars at the proper height, but use just one tiny screw in the middle to press down two big components. Before having any chance to apply enough pressure, the bars bend, and apply pressure only to one edge of the part! The result: Very poor thermal contact, because a big portion of the part's seating surface ends up with a layer of air between it and the heatsink!
You can see the the board through the wedge gap between the pressure bar and the part!
And one more problem: Even electronic hobbyists, let alone engineers, know that power devices first must be mounted to the heat sink, then soldered to the circuit board! This inverter is designed the other way around. There is no way to solder the parts once installed on the heat sink. It becomes necessary to solder them first, and then install the whole board with all the parts in the case, and bolt down the parts as the last step! This results in poor seating, and severe mechanical stress on the connection pins.
By the way: The DC-DC converter board, with its heavy transformers, has no mounting screws whatsoever. It's attached ONLY by the pins of the TO220 MOSFETs and diodes!
Oh, and while we are looking at the cooling, let me tell you that the case of this inverter, which acts as the only heatsink for the big power semiconductors, has the fins on the outside, in horizontal orientation, while the fan draws air through the inside, where there are no fins! Not that this would matter much, because anyway the fan is essentially decorative: The openings through which air can enter the inverter have a total combined area of less than 5% of the fan's area!
It is required by safety standards to leave a certain distance between printed circuit tracks that carry high voltage, in order to prevent creepage, flashover, and the ensuing risk for people, or damage. At usual line voltages, 4 millimeters is a decent distance. Well, this inverter in many places has just 0.8 mm clearance. In the photo above, you can see how close the PCB comes to the case. Whether it touches, or if a small amount of air remains between, depends mostly on luck. The tracks carrying high voltage go to about 1mm of the board edge, and in one place, apparently due to lax manufacturing tolerances, the tracks almost reach the edge.
The same meager creepage distances are kept to the screw holes. Insulating washers were used to keep the screws from directly touching the copper areas, but the copper reaches very, very close to the screw holes, and thus to the screws, which thread into the aluminium enclosure.
I grinded away some copper in the most dangerous places, to regain some peace of mind.
When I first tried the charger included in this inverter, the filter capacitor of the charger's auxiliary power supply quickly got hot and bulged out. I noticed it just in time to avoid another capacitor explosion and electrolyte spillage! It's a 470uF 35V capacitor, which works at well above 35V...
The bad choice isn't the capacitor, but the transformer feeding it. There is no reason to use such a high voltage, to feed a 7815 regulator chip. Fed from that high voltage, the regulator gets very warm too.
By the way, the rather largish heatsink on this regulator is supported SOLELY by the regulator! It has no other attachments anywhere! It's a rather wobbly affair.
And also it's interesting to note that the heat coming from this heatsink flows directly to the optoisolator, whose warming causes the "regulated" high voltage to drift up! So, if the unit is used as an UPS, the optoisolator will be hot whenever the inverter comes in, which will make the inverter produce a very high voltage.
What do you do if a large stranded wire doesn't want to thread through the hole in the printed circuit board? Simple solution, Power Jack style: You cut off half of the strands, and use only the remaining ones!
Shall I continue? I could name quite a few additional flaws! But this page is getting too long already.
To reduce the risk of burning out all the power semiconductors again, the first thing I did was adding protection circuitry. Shown in this photo is a current sensor, that uses two current transformers, each with a load resistor and a bridge rectifier. They sense the current on each group of secondary windings of the DC-DC converter. The outputs are in parallel, so that the highest one will dominate.
I would have preferred, by far, to sense the primary current, but the board layout didn't lend itself to it. It would have required essentially making a new printed circuit board. So I decided to try my luck with this secondary side current sensor.
I installed this current sensor on the DC-DC converter board, and also installed there an overvoltage protection circuit.
This overvoltage protector senses the voltage on each of the 1500uF 200V filter capacitors separately, and if any of them exceeds 180V for a significant time, it triggers a small SCR, which is used to shut down the DC-DC converter, and light a red LED. The circuit is reset when the inverter is switched off and on again. This is the protection circuit cited somewhere up in this page, that costs less than two dollars. It works with DIACs, that form a relaxation oscillator around a tiny ferrite transformer, which triggers the SCR.
Note that both the current sensor and the overvoltage protection circuit are galvanically insulated from the circuits they sense!
In the foreground you can see one of the yellow toroidal cores, which are much too small, and get extremely hot. This one already has black burn marks on it!
Then I moved the error amplifier to the other side of the optoisolator, to eliminate almost all effect of the optoisolator's thermal drift. To do this, I had to build a new error amplifier, and install it on a small board, because on the control board I had too little room to install anything. This error amplifier contains a high voltage divider using two 270k, 1/2 watt resistors in series. That's a lot safer than the original ultra-tiny and overloaded SMD divider!
While doing this, I also inverted the function of the optoisolator. As a result, if it fails now, the output voltage will go to zero, and if any of the auxiliary supplies fails, the voltage will too go to zero.
The control board was modified, to eliminate the SMD high voltage divider, buffer, and the original error amplifier (which used the internal amplifier of the SG3525), and add a few parts needed to inject the signal from the optoisolator directly into the pulse width modulator, and to provide an injection point for my current sensor and overvoltage protector.
Then I got my Dremel out, and severely cut up the main board! I rearranged the chopper's output filter so that now there is one inductor on each side. I also rearranged the EMI filter so that now it works. I had to remove several large components to do all the cutting.
And then I did a modification which is crucial for safety, but also makes the unit better suited for my application: I completely separated the functions of the inverter and the charger, so that they can operate at the same time, and eliminated the UPS switchover functionality. This allowed me to cut up that nasty direct connection between the 220V input and output! As it is now, it should be possible to use the charger, to charge the battery from the primary energy source, while at the same time running the inverter, to power the highly variable loads connected to it. This should allow running a home, including occasional loads of up to 3kW (the highest I need) for short times, while using a 500 watt alternative energy system as primary supply. To this end, I reduced the current setting of the charger, which is originally 50A, to a level my system can feed.
The underside of the board of course also needed to be cut up quite severely to do all this! Later I soldered all the wire bridges in place.
With these modifications, the inverter has become hugely more radio-friendly! The output sine wave is now pretty clean, and radio noise, while still present, is only a minor nuisance, instead of the totally disabling blare it was before. External filtering should get rid of the rest. I also installed a line filter to the charger input, to make it RF-quiet enough for now.
You might say I'm almost there... But not so. Stability of the control loops has been a nightmare. In part it's because I still haven't replaced the much-too-small filter inductors in the DC-DC converter. As they saturate to different degrees, the power loop of the converter changes its behavior. And for the charger, I only managed to obtain conditional stability under most operating conditions, but at that point where it moves from controlling the current to controlling the voltage, during the charge cycle of a battery, it's still unstable, and it seems that nothing I do can stabilize it!
Also the thermal coupling of the parts remains to be improved, the filter inductors need to be replaced, and many other such "minor" things, but the really big obstacle is this: I don't see any way in the world to stabilize the control loop of the DC-DC converter, while at the same time assuring that it cannot develop flux walking problems. The issue is that not having primary current sensing, flux walking cannot be prevented through cycle-by-cycle current limiting. That requires using a slow error amplifier, but such a thing has a 90 degree phase lag, which adds up with the basically RC behavior of the power circuit, to create instability conditions. With a fast error amplifier, that has phase lead instead of lag, it would be easy to stabilize this circuit, but then there would be severe flux walking problems! No solution seems possible other than implementing primary side current sensing, and that's difficult, given that it would have to be done AT LEAST in each of the six converter groups separately, and that the currents involved are several hundred amperes, and that the room available is tiny! Any "solution" without primary side current sensing will be unreliable.
At that moment, I had given up, and intended to bury the shiny almost new Power Jack inverter in a deep, damp hole, then plant a tree on it, the most thorny sort of tree I could possibly find.
During october and november 2009, again I spent lots of time on this crappy inverter. By this time, I have already invested much more time than what this thing is worth, but it has become sort of an obsession, a game and a challenge to try and make it work well enough!
I first digged deep into the charger circuit, to analyze exactly why it was unstable. I simulated, calculated, and finally found out that its current loop compensation was good enough, but its voltage loop compensation was correct only for this thing to work as power supply connected to a resistive load, and not as a battery charger! While without a battery connected, the resonant frequency of the charger's LC filter is in the range of a few hundred Hertz, when connecting the battery the resonant frequency drops dramatically, the exact frequency depending on battery capacity, equivalent series resistance and so on, but typically it becomes less than one Hertz!
So I calculated a new set of values for the voltage loop compensation, based on the largest battery bank I intend to charge, and implemented this. Now the charger is inconditionally stable, and works like a charm! It's interesting to note that the corner frequency of the new loop compensation is 1550 times lower than what the Power Jack engineers used! At the same time, the gain in the middle frequency range is much higher than originally.
To complete the charger refurbishment, I installed a power entry block with switch and internal EMI filter, and to this I added a further common mode choke and bypass capacitor. Now the charger doesn't send any significant noise out on the AC line. But I still have to filter the battery line. Since that one is shared with the inverter proper, I left that for later.
I ordered some T-184-26 cores, and wound new inductors for the DC-DC converter's output filter. This photo shows a new inductor side by side with an original one. The size difference is pretty obvious, I would say! And even these much larger inductors are still marginal, limited by the space available inside the inverter. They have four times the inductance of the original ones, and retain their inductance two times better at high DC current. In addition, they loose only about 8 watts, instead of the 53 watts lost by the original. 8 watts dissipation makes the big inductor get warm, but not excessively hot, while the 53 watts dissipated by the small one make it unsolder itself, loose magnetic properties, discolor, or even self-destroy if run long enough.
With the new filter inductor in place, at least I could test-run the DC-DC converting section under stable conditions. So I ran it in open-loop configuration, measured the real loop response, and tailored an error amplifier to suit its phase/gain behavior. I finally came up with a loop compensation that has 38dB gain up to 600Hz, then starts falling off, introducing a total of three low-pass poles, one of them dominant over the other two, to drop the gain at high frequencies to a point that avoids too fast pulse width changes, which could drive the push-pull converter into single-sided saturation. All this while maintaining a very healthy phase margin. With this error amplifier, the 340VDC output from this stage stays regulated to within a few volts, even while the sine wave chopper section draws a huge pulsed current, and remains fully stable under all conditions tested so far. I don't know yet if this arrangement will prove good enough in practical use, but at least there is hope...
While testing the complete inverter, suddenly the output waveform started to degrade! While one half0cycle remained clean, the other one distorted progressively, getting worse and worse... After a while, it had become almost a triangle wave! At the same time, the inverter started making quite horrible noises.
The problem was quickly traced to one of the optocouplers in the sine wave chopper not turning on correctly, producing erratic output. And the reason for this problem, once again, turned out to be the utter incompetence of the person who designed this circuit! A 6N137 optocoupler was being driven through a 1k resistor, from a 5V CMOS driver output. This will produce about 3.6mA through the optocoupler's emitter. And the 6N137 is rated for an absolute minimum of 6.3mA, with a recommended value of 15mA! At 3.6mA, it will typically still function at a low room temperature, but no longer when it gets even slightly warm. As the optocoupler ages, it will progressively work worse at that low current. How stupid is a designer who uses 1k resistors everywhere, without even checking the datasheets to make sure the resulting current is in the safe range?
After swapping out those 1k resistors for a value that produces a current in the safe range, the waveform returned to a good sine, and remained there even when the inverter warmed up further.
A fan sucking air out of a closed box isn't much good, if air can't enter the box! As described above, the air inlet openings on this inverter's case total just 5% as much as the fan's active surface. To gain some more air flow, I removed the two large power outlet connectors. These multistandard connectors, that will mate with almost any plug used anywhere in the world, worked well enough, but by removing them and installing two standard Chilean outlets (the same type as used in Italy), which are much smaller, I was able to free up a lot of space on the front panel, which I gridded to use as air inlet. The total opening area is now three times as large as before, but that's still very small. I might need to drill a lot of holes into the shells.
This inverter had a peculiar configuration: The only part of it that was grounded in any way was the enclosure. The entire circuit was left floating, including the battery. The EMI filter at the output was also left ungrounded, and thus useless. The battery was also floating. Worse than that: There was a capacitor, installed by means of a piece of wire, joining the high voltage DC bus to the low voltage (battery circuit) side, which left the battery connected to 350 Volt pulses at 20 kilohertz repetition rate, respective to the AC output! This not only turned the inverter into a highly effective broadband noise transmitter, complete with dipole antenna (the high voltage wiring and the battery wiring forming the two poles), but also made one get a quite disturbing shock when touching the battery terminals!
I had already reconfigured the output EMI filter to actually work, by grounding its ground connection. Now I proceeded to remove this stupid shock-generating capacitor, and then grounded the negative side of the battery, which in fact means grounding to the cabinet and thus to real earth the whole low voltage and control circuit. This provides essential safety, and stopped the shocks, but I quickly found out why the Power Jack people added this capacitor: In the new, correct configuration, the control loop was picking up a huge amount of that 20khz signal, driving the loop wild!
The reason for this problem was quickly found: In his infinite wisdom and experience, the person who designed the printed circuit board saw nothing wrong in placing the feedback optocoupler over a "ground plane". Nothing wrong with that, of course... except that the "ground" he chose was the negative side of the DC link, which carries several hundred volts of 20khz square wave respective to ground!
Look at the photo. The small white box is that optocoupler. On the other side of the board there is a transformer. The optocoupler base pin is directly over one of the secondary pins of the transformer, which is at the high voltage side! There is a significant capacitance through the board, directly coupling the several hundred volts of noise into the highly sensitive optocoupler's base terminal, which is not bypassed in any way!
Great design, don't you think? ;-)
To fix this, I lifted the optocoupler's base terminal from the board, to reduce its capacitance to the noise "ground plane". Then I put some bypassing on it, just enough to reduce the noise to an acceptable level. And then I had to spent lots of time on doing some magic tricks with the control loop, to make it stable again, because the bypassing at the optocoupler inevitably degraded the phase margin of the control loop to the point of making the circuit conditionally unstable. At the end, it worked out well, but required international help from two people who took an interest in this marvel of Taiwanese engineering, and helped me with circuit simulations and suggestions!
The little SMD transformer shown above is part of a galvanically insulated auxiliary power supply that feeds part of the feedback loop, and the drive of the sine wave chopper IGBTs. Such power supplies should typically deliver about 15V. But this one only did around 11V. With my new error amplifier, which took a little additional current from this supply, the voltage dropped a little further. And due to the unclever choice of filter capacitors and high-side-driver bootstrap capacitors, strong current surges happen at 100Hz repetition rate. These surges ended up pulling the auxiliary supply just below the 10V limit, making the bridge drivers drop out due to undervoltage! The result was a horribly distorted sine wave at the output.
I first tried to fix the problem by simply removing the transformer and adding a few more turns to the secondary winding. When I removed it, I was in for the next scare! This transformer has one winding doen with common enameled wire, and the other winding uses wire with stronger insulation. This is usual in such transformers that have to isolate high voltages. But the person who designed this PCB negated that safety by placing a groundplane under the transformer, that is connected to the better insulated winding, and TOUCHES the enamelled wire of the other side! Tehre are even some vias, that don't have the green varnish on them, in contact with the enamelled wire! So, in the original configuration of this inverter, the only insulation between the 350V on the output side, and the battery terminals, is a layer of wire enamel a few microns thick, which can easily be scratched off!
I added two layers of insulating tape before reinstalling the transformer. Anyway, the insulation is still unsafe, because the tracks on the PCB are too close together. On the right side of this photo, the rectangular diode (in shadow) is on the secondary side, while the groundplane around it is on the primary side. The clearance between the two copper areas is far less than one millimeter.
At least I now have the battery circuit grounded, so that a flashover at this place would cause damage, but no risk of killing anybody.
Adding turns to the transformer only slightly improved the voltage. Enough to stop the problem of the drivers dropping out, but just barely. I later diagnosed the real problem as being a substandard PNP SMD transistor that fails to meet its hfe rating, compounded by the design using rather weak driving. I increased the driving, which produced about 13V, which should be good enough.
At this point the inverter seemed to be working with an aceptable performance. As a last improvement, I took out my wire clipper and cut out two resistors that were causing totally unnecessary power waste: One in parallel with the battery, another one in parallel with the charger's AC input. The one at the battery was probably intended to discharge the filter capacitors after disconnecting the inverter from the battery. It dissipated about a half watt at all times. This is not too bad an idea, but I removed it because I want to be able to leave the inverter connected to the battery at all times, even when off, without draining the battery. The other resistor, at the AC input, dissipated almost two watts and got very hot, and was totally useless. I'm clueless as to why the designers included it. If it was to discharge any tiny capacitances at the AC input after disconnection, a megaohm resistor dissipating just milliwatts would have been enough. There is such a megaohm resistor in the line filter I added, so I could safely remove the 30k resistor that was causing heat in an undesirable location, close to electrolytic capacitors that don't like heat.
After that, I gleefully closed the cabinet! The inverter now works reasonably well, doesn't cause too much interference, and should be very much safer in every regard than it was before. Time will tell whether it's now actually sturdy enough for practical use, or if the poor heatsinking or some other uncorrected problem will kill it again.
But I have the uneasy feeling of having been cheated. The engineering effort I put into bringing this immature Taiwanese product to a point at which it can be used, should have been made by the people producing it, and not by me, the customer! My feelings of sympathy go to the many people who bought one of these inverters, only to see it blow up, and lack the expertise to correct its many flaws.
Back to homo ludens electronicus.