We recently bought some of LEDsales' wireless LEDs to play around with. We examined them, then set about hacking the system to find out if we can make it bigger.
We stumbled upon these wireless LEDs on LEDsales' website some time ago but didn't see a use for them in DIYODE projects at the time. They have many uses, such as in model making, dollhouses, even wearables, Lego, and anything else you can think of where a light might be needed and wires are impractical. We didn't buy any at first because we don't often build a DIYODE project where wires aren't already everywhere.
However, they still have plenty of merit, and potential. Accordingly, we ordered some to see how well they work, and exactly how they function. They were quite obviously an induction product, but there were still a lot of unknown details. We made a very simple wireless power transfer circuit in Kids' Basics issue 48, which was based on a self-oscillating circuit using a small coil and the parasitic capacitance of a transistor. That is one end of the scale, with sophisticated wireless power transfer ICs at the other end of the scale. Then, we wanted to know if the LEDs simply connect across a coil like the Kids' Basics version, or if they have small circuitry.
We bought a 5V kit, which is the power coil and its driver circuit, plus two each of five colours of LED. We also bought several of each colour LED, which are available separately. The product data states that the LEDs will show light between 0mm and 80mm from the 76mm diameter coil. They will decrease in brightness across that distance, according to the laws of physics which govern all induction and radiation. While there are plenty of free energy devices and related flasities on the internet, none stack up to reality. We knew these LEDs, and any like them, would have a limited working range purely because of the laws of physics. No amount of engineering and clever design can change the laws of physics!
ANALYSIS AND TESTING
On opening the kit packet, we found the ten LEDs were SMD devices on a tiny PCB with an unknown component on it, mounted over a ferrite-bobbin inductor. We checked the unknown component with a 10x power magnifier but could see no markings. We have no idea if it is a resistor or a capacitor. If it is a resistor, then chances are that the LED lights purely from the induced current in the coil. However, if it is a capacitor, then the LED is probably fed by the current from the LC circuit at a resonant frequency, with current-limiting performed by the tiny fine inductor wire only being capable of passing a limited amount current. With no markings and not enough exposed connections to connect test probes (the inductor wires are under the board), we were left guessing.
The driver coil on the other hand, was both readable and testable. In the middle of the operation is an 8-pin SOIC Ic, an XKT-333. An internet search suggests this is a wireless power transfer controller IC, but it is not from a mainstream known manufacturer. Instead, it is one of many Chinese manufacturers producing either their own material, or modified or straight copies of major manufacturer's products. The XKT-333 appears to be an original product, but there is little to be found about the device, manufacturer, or any datasheet for the IC. We do know that LEDsales test their products, so we're not too worried.
We powered up the driver circuit with the requisite 5V, and held the whole bag of LEDs near it. Sure enough, they lit up, but unevenly. That at least told us that the coil was working, but for its exact performance, better testing was required. We chose one LED, and using plastic tweezers to make sure there was no influence on the magnetic field in any way, we moved the LED around. The faintest light is visible 80mm directly above the coil, while directly above the centre of the coil, the LED lights at 90mm.
The brightness increases the closer the LED gets, which is expected, with full brightness right on top of the coil itself. Sideways, we detected the faintest light at 47mm between the outer edges of the coil and the LED sat flat on the bench. By playing around, we came up with a rough and non-scientific approximation of the pattern of influence of the coil, based on visual observation. It is worth noting that the observations took place under full workbench lighting.
Also of interest is that at one point, when an LED was being slid closer to the coil and light was only just visible from the LED chip, it tipped over and suddenly glowed more brightly. The interactions of the electromagnetic field are complex, because the coil has a non-spherical radiation pattern, and the inductor coil does not have an even receptive area, either. Whether the bulk of the inductor coil, or the ferrite bobbin base, or a combination of both at different angles, are facing the transmitter coil, all influence the amount of induced current.
By this point, we had a pretty reasonable impression of how the LEDs perform, and wanted to examine the transmitter section. You can probably already tell that we were thinking of hacking and modifying it. The LEDs are cool as they are, if your use intention happens to follow the pattern of influence. However, we had plans to supersize things a bit.
The key was to understand the circuit. Without any data sheets for the ICs making up the transmitter, we had to analyse the results. Accordingly, we hooked up an oscilloscope to the output of the driver, with the probe on the ground side of the coil. It's a low-side switch design with the other coil connection to the supply rail.
Immediately we could see that the coil is not operating on a straight AC waveform, nor a square wave. The semi-sinusoidal peaked waves are spaced by periods marginally longer than the wave. The overall frequency is 220kHz. This was interesting, because the waveform ruled out a straight up LC oscillator using the coil. That would not produce a waveform with this structure.
The IC and the circuit in general proved hard to probe. We played with an oscilloscope for a while but some parts seemed to be active high while other sections seemed to be active low. Really it didn't matter because we already had the feeling we were not going to be modifying this circuit, but rather building one from scratch. We tried measuring the length of the wire and making our own coil: We succeed here. Using a 10mm mandrel, we wound a cylindrical coil. It had an inductance of 3.4µH, while the coil on the kit driver had an inductance of 3.6µH. We desoldered the kit coil and added ours, then measured with the oscilloscope. Sure enough, it functioned at 221kHz. The LEDs lit well but the pattern of influence was a very different shape. LEDs also gloed much more brightly in the middle of this coil, brighter even than when right next to the coil supplied. That makes sense, as there was still a much stronger magnetic field inside our new coil.
We added a 10mm ferrite rod to our coil, with the view to shaping the magnetic field and maybe making a wand which could influence an array of the LEDs. The trouble is, we were pretty absent-minded and forgot that this increases the inductance. The circuit stopped oscillating, went flat-line, and got very hot! The inductance with the ferrite rod inserted into our coil was 42µH.
The final experiment, then, was to connect a u-shape of wire the same length as the coil wire, to see if inductance was really part of the circuit operation, or if the higher inductance of the ferrite coil simply caused too much impedance. The circuit again got very hot, but this time there was at least a waveform, albeit one with sharper curves, and a much rougher form. The frequency was similar.
Of note in both cases is that although the supply voltage is 5V, the peak waveform voltage was much higher. It comes close to 17V in normal kit operation, and the u-shaped, uncoiled wire showed nearly 30V. Also of note is that the frequency did not change much (less than 1kHz) between having no LEDs and having thirty-five LEDs in and around the coil.
AMPLIFYING
Our goal with this project was to make a more powerful transmitter coil to distribute power to a wider area or from a greater range. The two ideas we had were to make a larger coil that could light over a greater area and from greater separation distance, and to produce a wand which could light an array of these LEDs by being pointed at them. We realised the former is likely easier than the latter, but we decided to pursue both and focus on whichever one gave the most promise.
We set out to take a deliberately 'hacker' or 'experimenter' approach. We could have researched for hours or days and come up with an exact solution. However, many makers just 'go for it' with a bit of research, a bit of existing knowledge, and a goal. We wanted to do that too, to show newer makers how we go about that process, how we handle roadblocks, and how we analyse results and pivot or replan.
We were still unsure if the LEDs had any current limiting, if they worked on purely induced current from the coil or if the mystery component was a capacitor so that the current comes from a tuned LC circuit. As such, we decided to run an experiment. Because the waveform seen from the kit driver circuit is not a half-wave rectified waveform but is close, we decided to build a driver circuit to take the output from a function generator, half-wave rectify it, and then feed it to a power drive stage to feed current into the coil.
Before we did that, however, we needed to know how much current is passing through the coil from the kit circuit. A multimeter is not suitable for this because at best we would get an RMS value, but more likely an average. The waveform here does not fit the AC or DC conditions that most multimeters are designed for. We therefore desoldered one side of the coil and added a 1Ω resistor in series. This is really too high a value but was the smallest we had on hand at the time. A more suitable item would be a 0.01Ω current-sense resistor, but we didn't have one and thought the experiment was worth running anyway. If the oscilloscope showed that the circuit was oscillating differently, we would abandon this plan and get the right resistor.
Analysis of the oscilloscope connected across the 1Ω resistor showed a voltage drop of 4V across the resistor. That equates to 4A of current! This was not constant but the peak of the waveform. Regardless, the circuit and resistor both got hot, fast. Also, the current value shown on the power supply changed: In a steady state with the kit on its own, current drawn from the power supply displays at 75mA. With the resistor inline, it spiked to over 300mA. Given that 4A was only the peak value and the power supply display is the average over a short time that is still far longer than the 220kHz cycle, 4A peak does not seem so far-fetched. The 1Ω resistor was definitely having an influence. The frequency changed as well, but fluctuated so we never got a great number to write down. The 0.01Ω resistor is a hundredth of the value and should have a lot less influence.
However, a sudden realisation occured: The photo shows the oscilloscope connected directly across the resistor, in order to measure the voltage drop across it. However, the circuit acts as a low-side switch. The oscilloscope and the power supply are both isolated and at completely different ground potentials. This means current was flowing through the ground clip, and causing the observed spike in current. It also altered the performance of the circuit. The next plan was to use maths mode on the oscilloscope. This is where two probes are used, one connected either side of the resistor, and both to the correct circuit ground. The oscilloscope can measure the difference between the two. That's great in theory, but that is one of the corners that has been cut to keep these budget scopes affordable: Math mode has a crazy sample rate and the green math trace looked more like static all over the screen. When we say budget, by the way, it's relative. $800 (the RRP of this scope at the time it was new) is a lot of dollars, but engineering-level scopes can cost several thousand dollars to upwards of forty thousand dollars.
While we waited for a chance to go out and buy the right resistor, we decided to try some oscillator circuits. The first was the very simple circuit from Kids' Basics Issue 48, which was a wireless power circuit. This one relied purely on induced power and was not a tuned receiver, so the transmitter relied on a centre-tapped inductor coil, and the parasitic capacitance of a transistor. The frequency can be adjusted over a range by adding a potentiometer between the coil and the base of the transistor, which the original Kids' Basics circuit did not have. Peak voltage on this circuit reaches around 120V as the coil's magnetic field collapses.
We wound a very different coil to the Kids' Basics version, but the current is very low in either case so that 120V is not an issue. The circuit is crude and the transistor overheats quickly. However, we were able to vary the frequency between around 125kHz and 217kHz. The LEDs lit up only above 200kHz and were only of reasonable brightness at the full 217kHz. The circuit is quite unstable as the components heat, so we couldn't get more data than this. However, we drew the conclusion that the frequency is important and is not just an arbitrary number in a means to generate AC for the sake of induction.
While we couldn't measure as thoroughly as we wanted to, this experiment did tell us that we do indeed need a 220kHz oscillator to drive our power section. Accordingly, we set about making one based on our favourite oscillator, despite its faults: The NE555. The circuit is the basic astable, configured with 2.2kΩ resistor R1 between the supply and pin 7, and an 820Ω resistor R2 between pins 6 and 7. There is a 1nF MKT capacitor for the timing capacitor, a 100nF decoupling capacitor to stabilise pin 5, and a 100µF power supply filter capacitor. The output was sent via a 1kΩ resistor to a BC327 PNP transistor, so that the short low pulse would be inverted into a short high. There is a pullup resistor as well, because the NE555's output never gets quite close enough to the supply voltage to turn off the transistor. We used the power supply's inbuilt current limiter to set the current at 200mA, although because of the transistor output stage, the NE555's current limit was less relevant.
The LEDs did glow, but weakly. We swapped out our coil for the kit coil, desoldered from its driver circuit. This did light the LEDs, but not quite as well as the kit driver and the transistor became rather warm. Current draw stayed below 200mA on the display on the power supply, but we suspect it spiked much higher during the peaks, with the display being an average. However, it became apparent that we needed more current in the coil. In addition, the oscilloscope screenshots from both the output of the NE555 IC and across the coil, did not inspire confidence. Whether it's because of the inductance shaping the current pulse or not, the high time is longer than the duty cycle of the NE555 at these component values would suggest. The output also exhibits some significant misshaping. The second image shows the oscilloscope connected across the coil, between the transistor's collector, and ground.
Regardless of what's going on in the circuit, the LEDs do not light well, and the waveform is nowhere near what we want, either.
However, before we did so, it was worth one more experiment. Sine wave oscillators are harder to make accurately and simply when compared to square wave oscillators. We needed to know if the waveform was indeed important, or coincidental. We came up with a rough apparatus for the job: A MOSFET module meant for Arduino, with a 12V supply connected to it and the kit coil on the output. On the input, was 5V from the other side of our dual-channel power supply, and an input from a function generator. The function generator was grounded to the 5V channel of the power supply (ours is independently grounded on each channel, isolated from each other and earth).
Pretty soon, we realised that in between the power being turned off to allow the unheatsinked MOSFET to cool, the duty cycle could be varied between around 20% and 60% and still light the LEDs. There was not a huge visible difference along this range, either, particularly between 30% to 55%. So, that meant a 50% duty cycle oscillator would be viable. All we needed was an accurate oscillator circuit, with its output driving a high-current stage with its own current-limited power supply so that we could connect all manner of coil designs and sizes.
For a more accurate, cleaner oscillator, we decided to try a pretty basic op-amp multivibrator which has a particular advantage. Normally, the maths to find the frequency looks a little uncomfortable for those who don't like maths much:
However, when the gain (ß) is made to be 0.462, the frequency equation simplifies to:
Setting the gain to be 0.462 isn't too hard, either. The gain is set by the relationship of R2 and R3. The non-inverting input of the op amp is connected to the junction of these two resistors, forming a voltage divider from the output. The gain is whatever R3's value is as a decimal ratio of the combined value of the two resistors. So, a gain of 0.5 would have each resistor the same. A gain of 0.333 would see R2 with 66.6% of the total and R3 with 33.3%. A gain of 0.25 would have R2 three times greater in value than R3.
With standard resistor values and 1% tolerances, we are unlikely to get a gain of 0.462, so the easiest answer is to use a 100kΩ trimpot in place of R2 and R3, and adjust its value until the value between the wiper and the grounded terminal is 46200Ω, or exactly 0.462 of whatever measured value the trimpot has. That's the road we went down for our design. In addition, we also decided to add another op amp on the output, connected as a unity gain buffer. This isolates any impact that the load current would have on the voltage across either R1 or what is now R2, the trimpot.
Drawing a load current from an op amp which has such high-impedance outputs, by virtue of ohm's law, will lower the voltage because most of the current will flow through the load, reducing the sensed voltage at the op amp inputs. Having the oscillator output feed straight into another high-impedance output guarantees no appreciable effect on the voltage through the resistors providing gain and timing. On the other side of the buffer, the output is available to drive current into a transistor or voltage into a MOSFET.
The other section we thought about in our build was a current limiter for the supply voltage to the coil. Particularly when fewer turns of wire are used, there is not enough impedance in the coil to provide current limiting below that which the wire itself can handle. This is also really useful if using a bipolar transistor as a switching element rather than a MOSFET, which has its own gate drive current requirements if it is not going to overheat. Even when using MOSFETs, if there is not enough gate drive current to fill the gate capacitance quickly, the MOSFET spends too long in the linear region, and overheats at well below its rated current.
However, there are caveats. Most current-limiting circuits, and certainly all the simpler ones that don't require a week of engineering and a circuit board five times the size of the op amp board, create a lot of heat. This is wasted energy, and is also a disposal problem. So, while we had a couple of circuits lined up to try if we did indeed need a current limiter, the first option was to just make sure our coils had enough inductance to create a reactance that would provide current limiting.
We have rearranged the standard inductive reactance equation because in our case, we know what we want the inductive reactance to be, because we have a desired coil drive voltage and a maximum current that the wire and the power supply can handle. If we want the coil to operate at 12V and have a maximum current of 2.8A to give a bit of a buffer for our 3.1A rated power supply so that short-circuit or overcurrent protections do not kick in, then we use the Ohm's Law format equation but use inductive reactance in ohms in place of DC resistance in ohms. Note that there are better formula for this, that are more accurate. However, all are based on sinusoidal AC and need a known phase angle, but we are using a square wave and do not know the phase angle. The result, in our case for our purposes, is close enough.
With 12V powering the coil and a desired current of 2.8A, we need an inductive reactance of 4.286Ω. Dividing that by the result of 2πƒ where ƒ is 220,000Hz, we end up with a target inductance for our coil of 3.1µH. So, that is the value we aimed for in our build. While there are formulae out there to calculate the diameter, number of windings, wire size, and other factors to produce an inductor of the desired inductance, our object here was not just an inductor: We needed a functional transmitter coil. That meant using objects around us to wind the coil onto, of diameters that would suit the end-use. The wire size would be consistent, too. We wanted to try different coil sizes and shapes as transmitters, so we felt it was better to use the method favoured by makers everywhere: Guess and check!
The original kit coil's inductance was known to be 3.6µH. That's not bad considering we wanted 3.1µH, so we took it as a rough guide. It has four turns of wire, so while we may not need exactly four, we definitely would not need twenty. We made a few different coils that ended up around the target inductance, measured with an LCR meter, and then set about building the oscillator circuit. If we were wrong about this and were going to need a current limiter, it would be added to the power supply side of the coil anyway and not affect the oscillator build, aside from the need for space on the circuit board.
THE OSCILLATOR
The oscillator is built according to the discussion above, with a MOSFET output. 220kHz is not fast enough to render most bipolar transistors problematic in terms of rise/fall time. However, the TIP41C originally chosen for this task has a DC current gain (the hFe figure) of 65 (measured by semiconductor analyser), so to pass 3A, it needs a base current of 46mA. This is above the current the op amp can provide from its output, so a second transistor is needed. Even at 46mA, at 12V the resistor limiting the current through the second transistor in the Darlington arrangement is handling too much power and gets hot. Hot enough that we smelled it before realising. So, we opted for a MOSFET which will drive very differently. We could have used a dedicated Darlington device but had none on hand.
While the transistor will handle the switching speed, some of the most common op amps will not. The key figure is slew rate. This is the rate at which the op amp can swing its output between the supply rails. Our supply rails are 0V and 12V (we are not using dual supply) and therefore an op amp has to swing across 12V total at both the rising and falling edge of the square wave. The LM358, one of the most often reached-for op amps in the common maker inventory, has a slew rate of 0.5V/µs. So, swinging across 12V will take six microseconds (6µS). However, the entire period for a 220kHz wave is 1 ÷ 220 000 = 0.000004545, or 4.545µs. Given that half this time needs to be high and half low, the slew rate is longer than the half cycle! We tried it just for fun to see on the oscilloscope exactly what happens, and the answer is: 'Nothing'. After changing component values to create a 220Hz wave to make sure the circuit was actually a viable design, we suddenly saw the expected square wave, at one one thousandth of the desired speed.
Accordingly, we checked through the slew rates on what op amps we had in stock. The popular TL072, which is quite an old device but one that remains popular, has a slew rate of 20V/µs. So, slewing to 12V will take 0.6µs. That's much better and may just do the job. We had a couple of CA3130s too, with a value of 30V/µs. The others we had were all slower. We decided to pursue the TL072 first, on the basis that a lot more makers have one in their inventory, then swap to the CA3130 if a performance increase was needed.
In the end, the TL072 performed happily. As can be seen in the schematic for the build, however, there was a small change needed between this and the theoretical schematic presented earlier. The original design was for a dual rail (+/-) power supply. To make the circuit oscillate on a single supply, a bias is needed so that the non-inverting input is never stuck at the same potential as the non-inverting input. The easiest way to do so is to just add a resistor to the input from the positive rail. This should be roughly the same value as the setting for the lower half of the trimpot, so in our case, this was 43kΩ. The screenshot above is from the output of the oscillator, before the buffer section. There is a 1kΩ resistor to ground as a load.
We decided to use the IRF540N MOSFET, because they are again a commonly-held Maker part by virtue of being available at retail, and we have a bunch here. Because the circuit is 12V, we get away from the issues with driving MOSFETs from logic levels and do not need a dedicated logic level part. The first step is to figure out the required gate drive current. The less time a MOSFET spends in the linear region, transitioning between on and off, the more efficient it is and the less heat it generates. The gate current is calculated by dividing the gate charge figure from the datasheet, in Coulombs (usually nanoCoulombs) and usually denoted by 'Q', by the time required for the switching operation to be complete. Some discussions online just divided one second by the frequency but of course this would only give the time for the entire wave cycle. Instead, we took a closer look at our waveform on the oscilloscope.
This measurement was taken from the buffer output and shows that, at a best-case scenario, our rise and fall times are 500ns. It's probably closer to 600ns. The output from the oscillator is not terribly square. But that's ok, neither is the original kit driver circuit! This is good because the quicker a MOSFET gate needs to drive, the more current is needed at once. We want to go as fast as we can but within the limits of our drive circuit. That drive circuit is a simple NPN transistor connected to Vcc, and with a limiting resistor between its emitter and the gate of the MOSFET. When the output of the oscillator is high, the NPN transistor is on, and feeds current to the MOSFET gate.
Also present is a PNP transistor, which will turn on when the oscillator output is low. This will provide a path for the gate charge to dissipate through, otherwise the MOSFET will only turn off when the gate charge has leaked away. This transistor uses the same limiting resistor, to stop the gate current exceeding the maximum capacity of the transistor. Both resistors are also connected to the oscillator via the same 1kΩ resistor to limit base current.
We chose BC337 and BC327 devices for this task. To turn on in 500ns only requires 142mA of current. However, as noted, the faster the better. Even though rise time was limited by the oscillator, we set the current to be 450mA, close to the 500mA maximum continuous rating of the transistors (although some brands of the BC337/BC327 handle 800mA continuous). To do that, at 12V, we need a resistor of 27Ω, and that ignores internal losses in the transistors which will add further to the losses. Therefore, we are in safe territory as far as current limits are concerned.
If the transistors are connected with the PNP to the supply rail and the NPN to ground, in the more usual manner, an unfortunate situation can occur. As the square wave transitions from high to low or low to high, there is a point where both the transistors are conducting, as one is turning off and the other is turning on. This is only a very small amount of time but during that time, there is a short circuit between the supply rail and ground. This can kill the transistors very quickly, even at the limited on time involved.
We powered up the circuit, using a 21W/5W 12V brake light globe as a known-value load. We had begun to suspect that we were wrong about our current calculations for the coil, based on some heat we were getting on brief connections of the MOSFET under coil load. So, we felt the light globe was a safer bet until all circuit functions were proven. After all, the calculations we made were for inductive reactance based on an AC sine wave, not a DC square wave.
While the waveform was the right shape, because of the dynamics of using an op-amp circuit modified for single rail use, the output was never swinging to ground. Note the position on the left of the '0V' yellow marker arrow. There was a significant DC bias and that means the MOSFET was spending far too much time in its linear region, which we do not want. The photo above is at the output of the oscillator where it feeds into the drive circuit. The photo below is at the gate of the MOSFET.
The waveform is less than square, too. Just to see how
much the circuit was being affected by the inductive load of the light globe, we swapped it for a 100mA 12V 'grain of wheat' globe. The output at the gate of the MOSFET is shown below.
NE555 OPTION REVISITED
Unfortunately, after quite a bit of development time, this DC offset issue rendered the chosen circuit unsuitable. There may be a fix, another clever way of creating false grounds and such to force the output to go to ground, but we couldn't justify any more development time. Some makers out there will see the solution instantly, but at the time, we didn't see one and needed to move on. We would just have to go back to the spikey, but reliable and familiar NE555 circuit which we know goes to ground, or at least close enough to it.
Interestingly, the component values do not match those from any calculations. This is partly because of the interactions of component tolerance, and partly due to some factor we have not identified. The ICs used are basic models, and certainly not high-precision varieties. We built according to the schematic, and tried the driver section without the driver transistors. We used a resistor on the pin 3 output to limit the current to 180mA at 12V, fed it straight to the gate of the MOSFET, and the result produced the waveform shown here.
At this point, we weren't employing a driver like the op amp oscillator. We also measured on the load side of the MOSFET. However, the MOSFET is still getting quite hot, even when we temporarily added the gate driver circuit. MOSFETs do get hot, but how hot should we be expecting?
MOSFET HEAT TEST
To find out, we hooked up the same MOSFET with no switching circuit, just 12V constant to the gate, and two light globes as a load. We have a total of 47W of filament, which at 12 gives a current draw of 3.9A. That means using our power supply in parallel mode, so the gate voltage comes from the same supply. That is ok in this case, although in a real drive/switching situation there would be some capacitors involved to avoid instability at the gate due to changes in the load current draw.
After almost five minutes running, the MOSFET was still cool enough to touch with fingers. It was warm, and almost too hot to touch, but nowhere near the burn-inducing heat we got from the MOSFET in the NE555 switching circuit after only thirty seconds.
This confirms that the MOSFET is spending too much time in the linear region and not being driven fast enough. As such, the NE555 circuit will need a gate driver circuit as well, but that may not be enough to keep the MOSFET cool.
The Final Attempt
| Parts Required: | ID | Jaycar | ||
|---|---|---|---|---|
| 1 x Solder Breadboard | - | HP9570 | ||
| Packet of Wire Links | - | PB8850 | ||
| 1 x 27Ω Resistor * | R4 | RR0534 | ||
| 2 x 1kΩ Resistor * | R1, R3 | RR0572 | ||
| 1 x 100kΩ 25-Turn Trimpot | R2 | RT4656 | ||
| 1 x 33pF Ceramic Capacitor * | C2 | RC5318 | ||
| 2 x 100nF MKT Capacitor * | C4, C5 | RM7125 | ||
| 1 x 100µF 16V Electrolytic Capacitor * | C3 | RE6130 | ||
| 1 x 1000µF 25V Electrolytic Capacitor * | C1 | RE6230 | ||
| 1 x 1N4004 Rectifier Diode * | D1 | ZR1004 | ||
| 1 x BC337 NPN Transistor * | Q1 | ZT2115 | ||
| 1 x BC327 PNP Transistor * | Q2 | ZT2110 | ||
| 1 x IRF540N MOSFET | Q3 | ZT2466 | ||
| 1 x NE555 Timer IC | IC1 | ZL3555 | ||
| 1 x 21W 12V BA15S Globe * | - | Automotive Parts Suppliers | ||
| 1 x 21W/5W 12V BAY15D Globe * | - | Automotive Parts Suppliers | ||
| 1 x BA15S Globe Socket | - | Automotive Parts Suppliers | ||
| 1 x BAY15D Globe Socket | - | Automotive Parts Suppliers | ||
| 1 x Heatsink for MOSFET | - | HH8580 | ||
| 2 x Terminal Blocks | - | HM3172 |
With a fair amount of time and energy invested so far, we decided that this would be the last attempt. We would build the NE555 oscillator with a push-pull (or totem pole) transistor driver, and use light globes as current-limiters so that any length of wire could be attached and not alter the characteristics of the current draw.
This whole assembly would be mounted on a solder breadboard for stability, as even moving the timing components during breadboarding earlier had changed contact resistance enough to alter the frequency.
The rationale here is that, with light globes providing a constant current draw, we could fit any wire we like, even if it is a straight circle and not a multi-turn coil. This would enable much experimenting with patterns of influence. So far, we had tried to use a coil that matched the inductance of the kit coil, but we were not getting influence on the LEDs even as well as the original.
We included two pairs of screw terminals in series between the MOSFET and the supply voltage, in order to connect both the current-limiting light globes, and the test coil. When testing just the light globes, we used a wire link in the other terminal block.
TESTING
At first, the circuit was loaded with a single LED with resistor, to test the waveforms at various parts of the circuit with no possibility of overheating or the load changing the output waveforms. We tested at the output of the NE555, the input to the bases of the transistors in the driver (the resistor does shape the waveform slightly), the gate of the MOSFET, and the output of the MOSFET. The waveforms are interesting in that the output of the MOSFET when loaded with the LED is a short, sharp spike and not reflective of the waveform at the gate (second image). The third image is the MOSFET output when both the globes and a coil are in series.
With the globes limiting the current, the circuit functions as expected. However, the MOSFET is still getting very hot, despite its heatsink. In fact, it gets too hot. Hoping to eliminate the heat from the globes, and increase the coil current, we tried a coil made from 1mm diameter enamelled copper wire, with an inductance of 5.6µH. This, however, did not provide enough reactance and the current spiked to close to the maximum the lab power supply can provide. While this should have been well within the capability of the MOSFET, the waveform on the screenshot above gives a hint of what happened next: The MOSFET is still spending too much time in the linear region, and it overheated so much that it failed. Put on a semiconductor analyser, the result was a dead short across all three terminals.
The results were also underwhelming. While the LEDs did appear to glow brighter than the kit version when close to the coil, they were not lighting up as far away as the original. After so much development effort, this was disappointing but not entirely unexpected. We had calculated that the original coil was passing around 5A, so the 3.9A going through this coil was unsurprisingly not as effective. The fact that trying to work without the globes as current limiters had resulted in a baked MOSFET had discouraged us from trying that any further.
WHERE TO NEXT?
The wireless LEDs are a great product on their own. However, we still want to pursue the idea. We would love to make a system where an 's' shape of wire or even a folded design like a heating element could light LEDs over a broad area. We also still want to pursue the wand idea mentioned earlier.
As such, we need a more effective current limiting circuit, one which can limit current but not generate significant heat. It also cannot be a clamping circuit which shuts down completely when the current is exceeded, but rather one which will just limit current to a maximum value and keep allowing that current to pass.
We also need a better MOSFET driver, and this will probably have to be a commercial driver IC, capable of delivering much more current. We did end up finding a Darlington transistor in the workshop to explore with, but it was even worse than the MOSFET and that option can be ruled out. The particular example we had was limited to 100kHz, too. In addition to that, a better oscillator is needed, one that is more stable as other things in the surrounding circuit environment change.
Rather than an oscillator and current driver, we are thinking of something like a charge pump, where current pulses are generated rather than passed through a control element. This might allow greater current through the coil while eliminating the heating problem from the MOSFET in its linear region. Additionally, it might let us get closer to the waveform seen in the kit driver, which the square wave oscillator has moved away from.
Failing all of that, we can try a commercial wireless power transfer controller IC. However, these are not readily available (neither are MOSFET drivers, for the record) from retailers, instead being the preserve of wholesalers and trade suppliers. Also, we need to learn a lot more about them first! We have never used one, and never really read or heard a lot about them either.