This inverter project goes back to 2006, as an attempt to build a solid state inverter with around 80W output.
In April 1997, Electronics Australia described a 15W 12V to 240V inverter intended to operate a CD player in a car. It was essentially a much simplified version of the EA February 1979 300W inverter, in that the output regulation is achieved by varying the pulse width of the output waveform. By 1997, MOSFET's were the preferred output devices, and the new design used these instead of bipolar transistors. My intention was to use the circuit with a larger transformer, and uprated MOSFET's. Additionally, I had already developed an auto-start circuit which would be included. The inverter was housed in a case which once held an isolating transformer.

The EA April 1997 circuit which I initially based the inverter on.

In view of the higher power, I used paralleled IRF640's for the switching MOSFET's.  Due to the auto start circuit requiring one side of the transformer secondary being close to earth, the voltage regulation circuit had to operate in half wave. This meant deleting D1, D3, and D4. Because of the resulting higher ripple, the 10uF was changed to 47uF. I also changed D9 to 8.2V, and found it necessary to add 47uF across the now 8.2V supply, for stability.
The transformer was one which I had rewound the low voltage winding of, to 9-0-9V. It had previously been used with this inverter, albeit unsuccessfully with too high of a magnetising current. I thought it might be less problematic with a solid state inverter.
To cut a long story short, the inverter was built up with this transformer and appeared to work. However, the idle current was higher than I would like, drawing about 1A, and the inverter radiated so much RFI that powering a radio was difficult. Bypass capacitors in numerous places only partially helped. The inverter was given up on, and placed in my inverter cupboard for the next 18 years.

In 2024 thoughts had turned to resurrecting it, partly because of the increasing power prices, and wanting to get more onto my 12V solar supply - in this instance, some modems which draw a constant 30W. In the intervening time, I had learned more about  transformer saturation and wondered if this was the problem. Perhaps changing to another transformer would fix it. As it happened, I had an 80VA toroid spare, which would no doubt improve efficiency.

Transformers and Inverters.
The use of a readily available mains transformer used in reverse is a convenient choice for an inverter, and many such circuits have been published. There are several things to take into account before such use of a transformer will be successful.
Foremost is of course the turns ratio. At first glance one might simply assume the use of a 12-0-12V to 240V transformer. To do so will invariably result in a low output voltage; around 190V. The first thing to consider is that what is advertised as a 240V to 12V transformer will not have the expected 20:1 turns ratio. The advertised 12V will be for full load, and the turns ratio is actually slightly lower to compensate for the internal resistance of the transformer. This means the off load output will be higher; 15V would be typical for some transformers. The important thing to know is that the turns ratio is instead related to the off load voltage. In the case of a "12V" transformer which outputs 15V on no load, the turns ratio is actually 16:1.

Now, what happens when this "12V" transformer is used in reverse for an inverter? The output is 12 x 16 = 192V. And that's on no load! On load, again because of the transformer losses, it may drop down to 180V. The output voltage is lower again if the duty cycle is less than 100% (e.g. 80% with a vibrator), and is also further reduced due to saturation voltage of the switching devices. Bipolar transistors have higher losses than MOSFET's.
To compensate for all of this, the turns ratio has to be higher, and a typical turns ratio for a 12V inverter transformer might be around 24:1 to 26:1. In other words, a suitable mains transformer to use in reverse would have  9 or 10V secondaries.

Next thing to consider is the turns per volt ratio. Because we're putting 12V into a 9 or 10V winding, there's a possibility the winding will saturate. That is, there is not enough turns per volt. This means in simple terms, current is still flowing beyond the time when no further increase in magnetic flux is possible. The output voltage stops rising, the input current increases, and the the transformer gets hot.
The problem here is what to choose in terms of readily available mains transformers. Turns would have to be added to the 240V winding so that a 12V low voltage winding could be used. This is seldom practical, but there is a trick which can work fairly well if isolation between primary and secondary is not required. Remember that there will be something like 24V across the low voltage winding, so if this is connected in series with the 240V winding so the voltages add, a useful increase can be obtained; perhaps getting towards 220V.
An example of this method was used with a 12W inverter built to power a 240V LED bulb.

Another consideration is the waveform. Mains transformers are intended for a sine wave. With a square wave of 100% duty cycle, the peak voltage is reached immediately, and remains there until the polarity reverses. It may well be that the flux has built up to maximum amount before the primary voltage drops and reverses polarity, and the transformer saturates.
Also is the question of magnetising current, which can be quite high with some large transformers. This current is multiplied by the turns ratio, so when used in reverse, it can be quite significant. A few watts is trivial when the transformer is powered from 240V, but this can easily translate to well over half an amp when current is fed into the low voltage winding. Toroidal transformers are far more efficient than the usual E-I laminated kind, and as such have a much lower magnetising current. This makes them quite attractive for inverter use, and several designs have been described elsewhere on this site, both for AC inverters, and DC-DC converters.
Generally, the best solution is to simply try a transformer to see what the magnetising current is, and to see if it saturates. This will be obvious by a high current draw under no load. When viewed on a CRO, some distortion will be visible if the transformer saturates.

SG3524.
One of the things that concerned me with the EA circuit was that the output pulse was not centred in the middle of the waveform. I tried experimenting with the circuit with minimal improvement. I suspect the problem was largely that the 'triangle' waveform at pin 2 of IC 4 was not exactly triangle, since the capacitor at pin 2 of IC 1 is not charged and discharged by a constant current. At this point I gave up with the EA circuit and thought about the SG3524 IC. I built up a circuit on a breadboard, along the lines of this inverter. It worked considerably better than the EA circuit, with a much better waveform. Unfortunately, as soon as the inverter was operated with no load it became rather unhappy. The pulse width had to be reduced completely to restore operation. The toroidal transformer was becoming very difficult!

As a quick experiment I tried driving the MOSFET's with a sine wave, but the problem was they are not designed for linear operation, and it seemed the ones I had were not matched, since the resulting (and rather distorted) waveform was not symmetrical over the positive and negative half cycles.

Testing Transformers.
Clearly, the 80VA toroid had a saturation problem, and I briefly considered trying other transformers. I first tried a Ferguson PL18/60VA which is 240 to 9-0-9V at 60VA. This had the same problems as the 80VA toroid. So what about a 12-0-12V transformer? How low would the output actually be? A Ferguson PL24/60VA was then tested with a 12.6V supply. This worked well as far as lack of saturation was concerned, but as expected, output voltage was low.
 

Load	Output voltage
No load	237
15W	224
40W	209
60W	200

Output from the PL24/60VA transformer.

No load current from the 12.6V supply was 400mA. This was tolerable, but these transformers are not the most efficient anyway.
Next, a U.S. made Stancor transformer was tested. This was rated at 12-0-12V at 2A. This was inferior to the Ferguson, with only 209V across a 40W load - and that was with the primary and secondary in series to add the voltages.
So, the Ferguson looked like a contender if the 80VA toroid couldn't be used.

But first, let's see exactly what the limitation with the 80VA toroid is. A circuit using a 4047 was built up to provide a square wave with 100% duty cycle. This drove the MOSFET's in the usual way. The circuit was based on this inverter. The supply voltage was varied to find the optimum operating point, and where the transformer saturated. This was determined by waveform distortion coincident with a sharp rise in supply current.
 

Supply Voltage	Load	Output Voltage
9.6	No load	236
9.8	15W	235
10	40W	228
10.4	60W	230
11.5	75W	225

Supply voltage limit before saturation with the 80VA toroid.

So, we could use this transformer successfully, if the supply voltage was reduced. However, the method of reducing the input voltage was important. At this point, it needs to be mentioned there was a 10,000uF capacitor across the supply to the transformer. Fed from a variable power supply, all was well, and this is how the figures were obtained for the above table.
To make a practical inverter would require the 12V supply be dropped by means of a power transistor, which could also perform voltage regulation. To get an idea of the resistance required (and heat produced) to operate the transformer from 12V, I then tried a laboratory rheostat. Alas, we were back to the violent and unstable transformer behaviour. What was happening is that with no load, the supply voltage of course rises, and the transformer then sees 12V again. The bench power supply of course maintains the same voltage on and off load, so the transformer operation was always stable. The simple fix was to have the 10,000uF across the supply before the resistor, so the transformer did not see the low impedance of the capacitor. Now at last, everything was docile, and a way out of the problem was now possible.

Resistor in series with primary supply prevents saturation and allows control of output.

Series MOSFET.
For the series voltage dropping device, another MOSFET was chosen. The disadvantage with bipolar transistors is the high base current required, and the higher saturation voltage. First thoughts were to put the MOSFET in series with the 12V supply to the transformer. But then the gate voltage has to be higher than the supply. With a gate turn on voltage of say 5V, the gate will need to be about 5V more than the supply voltage, since the source will be around the supply voltage. This would require additional circuitry to obtain the auxiliary higher voltage of about 20V. The other alternative is a P-Ch MOSFET, but these are less common, and not in my stock of parts.

The problem is easily solved by connecting the MOSFET in the negative supply; i.e., in series with the drains of the switching MOSFET's. The idea worked perfectly.
All we had to do was adjust the gate voltage of the modulating MOSFET to control the output of the inverter. Provided the voltage never allowed the transformer to receive enough supply voltage to saturate, all our problems would be solved.

Sine Wave.
But, why stop there? What if we fed a sine wave into the modulating MOSFET? We could modulate the supply to the inverter thus, and create a true sine wave output. Where to get a sine wave? As it turned out, this was quite easy by integrating the oscillator output of the 4047.
With a suitable amplitude of sine wave superimposed on the gate voltage, it worked surprisingly well. However, the voltage range over which the output could produce a sine wave was limited to about 40W. Above this, the transformer had to be driven harder, and while the waveform was a lot less sinusoidal, it was far better than a square wave. In fact, it was quite interesting to see just how docile everything became when the transformer was fed a sine wave. Even a fluorescent lamp operated well without complaint, and the output was free from any RFI.

Circuit of the new inverter.

Oscillator.
The 50 cycle square wave is produced by a free running oscillator based around a 4047 monostable/astable multivibrator. A square wave with anti-phase outputs is available at pins 10 and 11. The output has an even duty cycle which eliminates any DC in the transformer - especially important with a toroid. The MOSFET gates are driven from these outputs via 820R gate resistors which provide stability. The resistors effectively form a low pass filter with the gate capacitance of the MOSFET's. Their value is not critical, but if too high, the switching speed will be slowed down causing the MOSFET's to run hot. Too low, and stability might suffer.
VR1 sets the oscillation frequency. Supply is 7.5V stabilised by a zener diode fed from the switched 12V supply. The 7.5V supply is also used for the regulation and soft start circuits.

Constructed on Veroboard, this circuit contains the oscillator, driver, and regulator circuits. Note the voltage reference LED.

Switching and Step Up.
Four IRF640 MOSFET's switch the transformer primary. These are connected in two parallel pairs for each output phase. More powerful MOSFET's have become available since 2006, and it would be quite in order to use a single MOSFET for each phase. Since MOSFET's have a low 'on' resistance, and they are being operated as switches, they run cool and only require a simple heatsink. In this inverter, the aluminium chassis was quite adequate.
As described previously, the transformer is an Altronics M5109 toroid rated at 80VA. It has a 240V winding and two 9V windings. Note that there is a difference between Volt Amperes and Watts. The 80VA rating means the transformer can power an 80W load, as long as the load is resistive (e.g. incandescent lamp or heating element). However, if the load should have a low power factor, the 80VA rating still stands, but the wattage available is less. For example, the inverter can only provide 40W if the load has a power factor of 0.5.

Toroidal transformer. It's a lot lighter and more compact than the equivalent E-I laminated type. The finned heatsink is for the modulator MOSFET.

The 0.47uF and 0.022uF capacitors across the transformer windings are a legacy of the original circuit. They do clean up the waveform and reduce RFI, so were left in situ. The 0.022uF is rated at 1600V DC, but a 250VAC type could also be used. Bear in mind, most such capacitors are of the self healing type, and their capacitance drops over time as the aluminium foil gradually burns away. A neon indicator is also connected directly across the secondary to show when the inverter is actually running. This is important due to the auto-start circuit, and that the inverter runs virtually silently.

Output Control and Regulation.
The drain circuit of the four IRF640's is completed to earth by another MOSFET; a type P60N, which is rated at 60V, 60A, and can dissipate 150W. As the gate voltage is increased, the P60N presents a lower resistance, and the inverter outputs a higher voltage. The gate voltage comes from the 7.5V supply via the wiper of VR2 and the 1M resistor. And here we see the advantage of using a MOSFET. The gate can be fed by a 1M resistor in view of the gate being effectively an open circuit in terms of DC. It draws no current. If a bipolar transistor had been used, the base current would be a couple of hundred milliamps, and this resistor would be probably less than 100 ohms and dissipate a couple of watts.
VR2 is adjusted so the inverter outputs the highest voltage before the transformer saturates.

The regulation circuit operates in conjunction with VR2. The 240V AC from the transformer secondary is sampled via a half wave rectifier and a 200k series resistor. It is filtered by 22uF and applied to VR3. The 200k resistor consists of two series 100k 1/2W resistors to obtain sufficient voltage rating. It can be seen that as the 240V output rises, so does the voltage across VR3. The wiper of VR3 feeds an error amplifier of conventional design, using a BC183L transistor. The transistor type is non critical, and a BC549 was actually used in the prototype test circuit. The emitter is held at a constant 2V by the LED, which functions as the voltage reference. If the base voltage starts to exceed 2V plus the 600mV base to emitter voltage, the transistor conducts, bringing down the collector voltage. The collector voltage supplies VR2. It can be seen therefore, that if the AC output rises, the BC183L conducts which reduces the P60N gate voltage, and the transformer is fed less voltage.
At this point, the inverter is practical and fully functional. However, since the P60N is operating as a variable resistor, it runs hot and needs good heatsinking. For example, if the inverter is supplying 60W, the transformer should be supplied with 10.4V. Input current under these conditions is 6A. If the supply is 12V, then the voltage drop will be 1.6V. This translates into 9.6W of heat and wasted battery power. If the battery is being charged and is up around 13-14V, the situation is worsened, with 21.6W having to be dissipated with a 14V supply. When the supplied load is 60W, having another 21.6W going up in heat doesn't look very good at all from efficiency point of view (It's about 73%).

One may wonder about deleting the P60N and simply controlling the gate voltage of the IRF640's to perform the same function. Indeed, this was tried, but it was immediately obvious that linear operation was not going to be successful. There was a difference between the positive and negative half cycles, as if the MOSFET's were not matched. Since they are designed as switches, this should not be surprising. Furthermore, it's quite possible that the mismatching extends to the individual MOSFET's in the parallel pairs, with the possibility one MOSFET might be conducting more than the other, resulting in excessive dissipation. When the MOSFET's are used as switches, they are fully saturated and matching is not important.
The single modulator MOSFET approach means that all the IRF640's work equally as switches, but the modulator has exactly the same effect on all the IRF640's.

Waveshaping.
Since a MOSFET is being used as our variable resistor to adjust transformer output, a logical extension of this is to modulate it with a sine wave, which in turn should provide a sine wave output instead of a square wave, thus more accurately replicating a mains supply.
Experiments were done with some success. First thing is what to modulate the MOSFET with. The waveform needs to be a positive going half sine wave at 100 cycles. This is because each phase (and pair of MOSFET's) conducts for 10ms. Remember, the 4047 is actually oscillating at 100c/s since the creation of the two anti-phase outputs effectively divides this by two. So, a 100c/s source is already conveniently available. A simple RC circuit integrates the square wave into something approaching a triangular shape. The circuit below was initially used for waveshaping:

At the junction of the 100k and 0.1uF exists a form of triangle wave. When fed into the modulator MOSFET gate, it just so happens that because of the linearity of the MOSFET, that a reasonable sine wave output is produced. The amplitude of the triangle wave is important, so the 100k preset allows adjustment. Additionally, the gate DC voltage is important, and is adjusted by VR2 as previously described. VR2 and the 100k preset are fairly interactive so need to be adjusted together, one after the other.
I found that with a 15W load the sine wave was quite good, and still acceptable at 40W, though the tops were noticeably flattened. Even at 75W, the sides of the waveform still had some slope. The heatsink could run somewhat cooler than if the inverter was run with a 100% duty cycle square wave. Since it's a sine wave, the transformer voltage can be run closer to the supply for a portion of the waveform. Bear in mind that for a 12V supply, the sinusoidal peak is 9V, and the transformer will not saturate.
The circuit shown above can be used for anyone wishing to make this inverter in its sine wave form. Adjust for best sine wave with a 40W load, assuming the same transformer. With the particular components I used, it was actually possible to dispense with the 100k preset and simply connect the 100k resistor straight to pin 1 of the 4047. This makes setting up easier, although the sine wave was not quite as good.

Variable Duty Cycle Square Wave.
The final version of the inverter, as constructed, came about as a mistake while experimenting with the prototype. Somewhere along the way, I had substituted a 0.01uF for the 0.1uF in the waveshaping circuit. I didn't immediately notice at first, since the light bulb load lit up normally as usual. However, I couldn't quite get the proper sine wave - but then realised the inverter was operating even more efficiently. What was happening is that modulator gate voltage was now of a much higher amplitude triangle wave, and we now had a pulse width modulating circuit. It seems the linearity of the P60N is such that with a higher amplitude triangle wave was driving it into saturation much earlier. Depending how far up or down the triangle the P60N switched determines the pulse width.

Gate voltage of the P60N. Depending how far up or down the waveform the MOSFET switches determines the pulse width of the square wave output.

The heatsinking requirements are reduced. Even so, a proper finned heatsink is still necessary.

Source voltage of the IRF640's (Drain voltage of the P60N). The solid line at the bottom is earth.

It should be noted that creating a sine wave output by the method described is not completely efficient, since the switching device(s) operate in the linear region. Inverter efficiency improves as the sides of the waveform straighten up.

Auto Start.
Quite some years ago I designed an auto-start circuit for inverters. This saves running the inverter when the appliance it is powering is switched off, and is particularly useful when the inverter is remote from the appliance. This situation arises when the battery has to be remotely located. Obviously, it is better to have the inverter at the battery to avoid voltage drop, and to power the appliance through a longer cable, where 240V overcomes any losses. It is often more convenient to use the appliance power switch anyway.

Auto start circuit on PCB. 10,000uF capacitor is adjacent.

Electronics Australia had used an auto-start circuit for some of their inverters. While it works, it loses about 2.4V of the inverter's output voltage. In the overall output it is only 1%, but I decided to improve on that, and also simplify the circuit. The EA circuit is shown for this 300W inverter.

Turning now to my auto-start circuit, one side of the transformer secondary is connected to the live pin of the output socket via a switch. The other side is connected to earth via to back to back diodes. The neutral pin of the socket is connected to earth via a set of normally open relay contacts. The inverter receives its 12V supply through another set of normally open relay contacts. Thus, when the relay is actuated, the inverter switches on, and the neutral circuit is completed.

Imagine the inverter at rest, but with 12V connected, and the switch in the auto position. An appliance is plugged in and not switched on. Nothing will happen since the relay is not actuated. When the appliance is turned on, things begin to happen. 12V flows via the 1k 1W resistor, through the appliance, through the switch, through the transformer secondary, and to earth via the right hand side diode of the back to back pair. 600mV now appears across the diode, and biasses the AC187 transistor into conduction via the 47R base resistor. The collector current actuates the relay and the inverter starts.
Now there is AC across the back to back diode pair. The opposing diode conducts on the negative cycles, so only 600mV is lost with this circuit. During the negative part of the cycle, the transistor is not biassed, and the relay would normally drop out. That is, except for the 2200uF which holds the relay in during the negative half cycles. The 12R resistor limits the collector current when the transistor first switches on. Since the 2200uF is fully charged when the inverter is off, as soon as a load was connected, it would suddenly discharge through the AC187. This current could be high enough to damage the transistor, but with 12R in series, it can never exceed 1A.

When the appliance is switched off, the 600mV is no longer present across the back to back diodes, the transistor ceases conducting, and the relay deactivates, switching off the inverter. The 1N914 across the collector and emitter of the AC187 prevents any back emf from the relay coil damaging the transistor. In reality it probably isn't needed since the 12R and 2200uf would provide sufficient damping. In series with the relay coil is a 1N4007 diode. This provides reverse polarity protection for the inverter, since the relay cannot power up if the input polarity is reversed. No blown fuses if the polarity is reversed with this circuit!

The AC187 is a germanium type, chosen because the base turn on voltage is about 250mV. If a silicon transistor was used, the 600mV available would be right at the threshold of turning on, and the operation might not be reliable. The remaining component to discuss is the 1k 1W resistor. This is required because when the relay contacts close for the neutral connection, the 12V supply would otherwise be shorted to earth. The 1k limits the current to 12mA. The resistor value could be higher, but then appliances with a high resistance might not provide enough bias current to switch on the AC187. In the normal way, only 144mW is dissipated in the 1k resistor. So why the 1W rating? Should the 12V contacts close just before the neutral contacts, the 1k resistor will be in the path of the 240V current flow. A 1W rating should allow the resistor to cope with a few cycles of excess current, before the neutral contacts close. In reality it's very unlikely to happen, and in the 24 years I've used this circuit, it has been completely reliable.

Switching is provided so the inverter can be set to auto or manual start. Some appliances appear as an open circuit or very high resistance at 12V, and will not activate the auto-start. Typical is a fluorescent lamp which presents an open circuit until the starter ionises. In the manual position, the collector of the AC187 is earthed, switching on the relay regardless of load. The reverse polarity protection still works in this mode.
The switch used is a DPDT with centre off. Note that this switch also switches the mains socket. This is to allow the inverter to be switched off with a load connected. Unlike the EA circuit, there is no standby current draw in the auto mode.

Soft Start.
Because of the time constant formed by the 200k resistance and the 22uF capacitor in the voltage regulator circuit, it can be seen that when the inverter is first switched on, the 22uF will have no voltage across it. The result is the inverter operates at full power until the 22uF charges, and the error amplifier begins to function. To prevent excess voltage being fed into loads when the inverter is first switched on, a soft-start circuit is included. Also connected to the gate of the P60N is a 1N914 diode and 22uF capacitor. The 22uF is charged from the 7.5V supply via a 150k resistor. When the inverter is first switched on, the 22uF is discharged, and this effectively earths the MOSFET gate through the 1N914. The inverter produces minimum output. The 22uF begins charging through the 150k, and once the voltage is higher than the peak gate voltage, the 1N914 no longer conducts and has no further effect on the gate voltage. By this time, the voltage regulator circuit has stabilised. It is important that the 22uF discharges immediately the inverter is switched off, so that the soft-start circuit is always ready for next switch on. This achieved by the 1N914 across the 150k. When the inverter is switched off, the 22uF discharges via this diode into the 7.5V rail and everything which loads it. When the inverter is on, the diode has no effect since it is reverse biassed.

Setting Up.

• VR3 wiper to earth.
• VR2 wiper to earth.
• CRO between IRF640 drains. Power supply must be isolated from earth for this connection.
• Increase VR2 to just where a waveform is visible. Adjust VR1 for 50c/s.
• Load inverter with 75W and increase VR2 to the point just before saturation becomes visible.
• Load inverter with 40W and adjust VR3 for 235-240V.
• Load inverter with 15W and see that output does not exceed 250V; readjust VR3 if necessary

Output waveform at 15W and 75W loads respectively. Note the duty cycle increases under load to maintain the rms voltage.

Regulation.
Input voltage at the 10,000uF was adjusted in 500mV increments from 10 to 14V. A 40W load was used. Output voltage was measured to observe regulation vs. supply voltage.
 

Input voltage	Output voltage (40W load)
10	207
10.5	214
11	220
11.5	225
12	231
12.5	236
13	241
13.5	246
14	251

Internal view of assembled inverter.

It is quite remarkable to compare the inverter at the start of this project and what it ended up as. The 80VA transformer which was at first so difficult to use and inappropriate,  turned out to work very well in the end by using some rather unconventional circuitry. The biggest surprise was the overall efficiency achieved.