2010-12-26
Wireless Power Experiments
Along the lines of the season's traditional blinkenlight projects I had the desire to build a rotating persistence of vision display of festive geometry. The usual problem of delivering power to the rotating bus from the static part was the first engineering challenge. Most builders avoid this problem entirely by putting batteries on the spun bus, or by using slip-rings (often in the form of a 3.5 mm audio jack and socket). One inventive idea I've seen is to steal power from the motor armature but this requires removing one of the motor bearings. I decided instead to try sending power to the spun bus wirelessly - an engineering solution which incidentally also solves two other problems intrinsic to the design of spun displays - more on that in a later article...
Some Talk
First lets talk about wireless electrical power transfer. Here is a lengthy video I made on the subject, covering the basic physics and the technical difficulties of achieving high efficiencies. It covers much of the material of this article and can be watched instead of reading the article, or in addition to it. The video, much as this article, makes many simplifying assumptions as doesn't talk much/all about source impedance, matching, the electric and magnetic properties of matter (dielectric and hysteretic losses). Neither do I provide a design procedure, rather I give a rough sketch of the critical parts of the system you need to look at closely if designing your own.
Induction, Transformers, Mutual Inductance and Coupling
For the original display project it would be a simple matter to build a rotary transformer and send power across a small gap, but doing so requires great mechanical precision. Conventional power transformers achieve high efficiency because their closed magnetic circuits give good coupling between the primary and secondary inductances, minimising their leakage inductance. Good magnetic coupling between the windings in a rotary transformer means minimising the air-gap in the magnetic circuit.
A much more easily hackable solution for my toy display is to use so-called "evanescent wave coupling" to send power between tuned resonators. This allows efficient operation at much weaker coupling (much larger air-gap and/or no magnetic core material at all), as long as the resonators are fairly high-Q (low loss). Essentially we allow the much larger leakage inductances associated with low inductor coupling coefficients of "air core" transformers and then tune them out with capacitance, utilising the remaining mutual inductance to transfer power between the resonant circuits. In effect we utilise the large "buffering" capacity of high-Q tuned circuits to render an impedance match from the TX power source into the RX load over a poorly coupled path. As long as the inductors are physically small with respect to the free-space wavelength of the frequency of operation very little radiation (leakage of energy into space as radio emissions) will occur.
Yes, Tesla had this idea working in 1893, but there are a number of practical issues that limit the technique to niche applications, such as toothbrush and cellular telephone chargers where the coupling coefficient can be kept high, placing less demands on the circuit components, especially inductor Q. Sending power significant distances of free-space without radiation demands very high-Q as inductor coupling (mutual inductance) drops rapidly with inductor separation. As the coupling/mutual inductance drops the resonator losses become large with respect to the reflected load impedances seen "across the gap" from the secondary to the primary, efficiency drops rapidly unless resonator Q is increased to keep loss small with respect to the actual useful load dissipation. There is an almost linear relation between the product of inductor Q and coupling coefficient (denoted k) and the total system efficiency (n). As k drops Q must be elevated to maintain the same system efficiency, basically n ~ kQ. With practical inductors, Qs over 100 require special and expensive techniques to achieve, so most implementations of wireless power transfer try to keep coupling as high as possible. Inductor coupling basically comes down to proximity and alignment.
Inductor Coupling
To get a practical feel for inductor coupling coefficients, I performed a simple experiment. I wound a pair of planar ~4 uH coils on ~50 mm OD former using 700 um magnetic wire. Coupling was assessed in face-to-face (axial) and edge-to-edge (planar) placements at spacings up to 2 diameters. You can perform this experiment yourself; measure mutual inductance, and hence coupling coefficient by measuring the individual self-inductances of each coil and the total inductance of the coil pairs in-phase (or anti-phase), the difference is twice their mutual inductance. Their mutual inductance is the coupling coefficient times the geometric mean of their individual self-inductances.
You'll note that axial alignment offers the best coupling, and that coupling actually goes through zero and becomes negative with planar alignment. (Negative coupling means phase reversal; picking up more flux from the other side of the coil, for power transfer purposes phase is irrelevant and we care only about the magnitude of the coupling coefficient).
If you try this with different sized coils you'll find the coupling always drops rapidly below 0.1 as you separate the coils by more than a diameter or so. This means to keep a high coupling coefficient over distance you need big antenna coils. It is instructive to repeat the experiment with one large coil and a much smaller diameter one - as would be the typical case of a practical wireless power system - the coupling is weaker again. Coupling also varies with coil alignment, placed orthogonally coupling plummets almost to zero, indeed you can completely null the mutual inductance between coils by suitable placement. In a power transfer system such an alignment would deliver no power to the load.
Inductor Losses and Q
High-Q inductors are best made by using thick wire (large surface area), space wound with a length to diameter ratio of the overall coil of somewhere between 1 and 2. This optimal length/diameter ratio comes from ohmic/skin effect loss minimisation and inductance maximisation (wire length vrs coil geometry), combined with the need to space-wind to minimise proximity effect losses. Unfortunately this optimal shape is not a very convenient shape for practical wireless power systems, especially as diameter must be large to allow large coil separations and maintain coupling efficiency. Wireless power system resonator inductors are therefore a compromise between coupling, physical form, and Q/efficiency (and material cost).
Before continuing to just hack away on the bench, I did some modelling on paper and simulations in LTSpice to get a feel for the key design problems. Resonator Q is critically important for high efficiency at low coupling. The TX-side antenna Q is in some ways more important, ideally there should be relatively large circulating energies in the resonator with RX-side loss (power transfer) being seen as additional loss only on demand. Of course the RX antenna is important too, but the TX antenna is always powered, better Q means it wastes less power when not loaded. You can think of the system as a double-tuned band-pass LC filter with very weak coupling between the resonators. Insertion loss is related to resonator unloaded Q and system bandwidth (loaded Q). The narrower you make the pass-band the more loss you experience due to finite resonator Q, but the weaker the coupling you need - and also the more accurate your tuning needs to be... Maximum transfer efficiency occurs only for one resonator loading (selection of loaded-Q for optimum power delivery given all system properties).
This basic physics of the required resonator tuning and coupling is one of the main reasons "Witricity" hasn't seen widespread adoption. Some claim to have systems which dynamically optimise either or both ends of the system, but it is quite obviously a nasty engineering challenge to do so, especially in highly-asymmetric systems (the typical usage) of a single TX and multiple RXs with different alignments to the TX antenna, differing couplings, practical Q for a low-cost implementation, etc. Fortunately it is a fairly tractable problem if you just want to send a watt or two across a few inches and aren't too fussed about perfect efficiency. Charging "mats" are very practical and can operate at k in the region of 0.1-0.4 for practical geometries and implementation costs.
A Simple TX for Experimentation
OK, so back to practical stuff on the bench. I breadboarded a simple push-pull power-oscillator using a pair of 2N7000 MOSFETs, operating in the HF region. (I've attached an LTSpice model if you want to tinker with it, I used a 2N7002 model and asymmetric bias resistors to keep spice happy, the real circuit uses 2N7000s and starts just fine with 4k7 bias resistors on both sides.)
The "180p" capacitor tunes the tapped coil to the frequency of operation, select or make it variable as desired. In one implementation I put 6v8 zener diodes on the MOSFET gates to protect them against over-voltage destruction, at low powers this is not strictly needed, but at higher powers you may need them. Similarly the pair of 33 pF feedback capacitors need to be selected with the frequency of operation in mind. The MOSFET drain breakdown voltage is also important if you are trying to scale up this circuit. While simple, other approaches are probably better for high powers, the MOSFETs are spending a lot of time in their transition regions, dissipating a lot of power. A purely switching class-E approach is obviously better, but suffers from sensitivity of tuning to load impedance in my brief experiments with it (using an IRF510 device). (I've attached another spice model attempting to show the class-E TX approach, I started with values derived using my class-E power amplifier design calculator, as shown it is not perfectly tuned. The practical circuit tunes up nicely and is quite efficient > 70%.) The breadboard TX in the video above is a class-C version with weak capacitive coupling to the tank to optimise its Q. Yet another approach is half or H-bridges, these show great promise, perhaps driving a magnetically coupled "link" winding rather than the tank directly, allowing the tank to float, and facilitating easy variation of coupling to it to optimise its Q... A subject for more detailed investigation at a later date perhaps.
Anyway, combined with one of my LC tank jigs I was lighting LEDs and Neon bulbs over a few inches within minutes of breadboarding the simple push-pull prototype.
The NE-2 bulb demonstration is instructive. Without much of a load, the voltage across a tank in the near-field of the TX rises to enormous levels at resonance. A load dump with a sensitive circuit in the RX might be disastrous unless you anticipate this effect.
This combined with the LED example also demonstrates the simplicity of a "shunt" regulation approach, the LED clamps the resonator voltage to only a few volts where as with the NE-2 bulb the tank can swing about 60 volts before the Neon ionises. The resonator Q is degraded much more by the LED, reducing the power transfer and offering some regulation. The Neon bulb was dissipating about a watt in comparison because its higher impedance load allows more efficient power transfer. One can tap-down on the inductor (either physically or electrically using tapped capacitance or other matching techniques) to improve the "match" and power transfer efficiency to lower impedance loads.
In a practical charging system rectification of the RF waveform costs you a lot of voltage, so it is important to use low drop, high speed diodes (Schottky for example) and ideally a bridge topology. (Yet another spice model is attached containing a rather non-optimal but very simple end-to-end system including bridge rectification.) Voltage regulation would be efficiently implemented using a modern switching regulator, but clearly the system needs to be designed to deliver sufficient voltage at the expected couplings to operate the switching regulator (which may then boost the voltage if required, perhaps bootstrapping itself also). Total available power is optimised by tuning the entire system to make the load match the source, and using very high Q components to minimise the system losses. In practice your resonators will have hundreds of volts across them with large circulating currents even just delivering milliwatts into a well-matched low-voltage load. This is why their losses are so very significant.
A Cute Tesla Coil
Naturally enough it is hard to not think of Nikola Tesla and Tesla Coils when doing this kind of work. It wasn't long before I wound a long-thin secondary with a self-resonance of about 21 MHz. As Tesla coils go that is an rather high frequency, but it works quite well. This cute little Tesla coil is limited by the 2N7000s of the power oscillator to about 1.8 watts, but produces a very strong electric field, the amplitude of which is sufficient to light fluorescent lighting tubes and neon bulbs.
The circuit is dead-simple, but the simplicity has some limitations. The power oscillator needs to be tuned to the secondary resonance and loading of the secondary can pull its resonance quite a lot - meaning you need to retune it some times. Coupling between the primary and secondary is slightly tight and the oscillator shows a bit of hysteresis-instability tuning across resonance. Still I think I may have to scale this baby coil up to something capable of producing break-out in air! Don't let this teeny coil fool you though, even at < 2 watts it is still capable of giving you an RF burn. The high-Z output can deliver most of that watt or so into an sub-cubic-mm of flesh, energy densities that instantly cook a pin-point of skin - it hurts trust me! Even initial tests using my C-jig and a planar coil of thin wire over the lashed-up primary on the breadboard gave me an uncomfortable burning sensation as I tuned through resonance with no load except my body - the voltages across the resonator are plenty high enough to push significant power through your skin. At higher power levels you will do damage before you can react.
An Aside - Fluorescent Tube Striations
It was while playing with this toy I first observed Fluorescent Tube Striations.
I may have vaguely observed this before, but it never really registered with me how fascinating they are. I had a lot of trouble finding any explanation of their cause online, but it seems to be a plasma physics effect, an interaction between the emitted photons and the electrons in the plasma. I found it interesting that their number was quantised and ranges of applied field would sustain only a specific number.
Another effect I noticed with the Tesla coil was NE-2 bulbs emitting purplish light in addition to their usual orange Neon lines. I need a spectrometer to tell exactly what is going on, yet another area of investigation for a rainy weekend...
Conclusions
My video is all very doom and gloom about the possibilities of generalised wireless power becoming widespread, but it is obvious that non-contact charging is quite practical and even easy to homebrew. There has been some commercial work on wireless charging mats, and a few attempts to standardise frequencies and near-field communication systems for allowing a range of appliances to communicate their needs to generic charging transmitters. Unfortunately a lot of commercial effort currently seems to be being expended in accumulating patents for ambush purposes, waiting for the day when someone finally gets a system to market and accumulates enough share to make it the defacto standard. Perhaps part of the problem is also lack of real need when balanced against the costs of implementation - plain-old copper wire works fine and plugging in a device to charge isn't really that difficult. Many after-market systems require a bulky module to be plugged into an existing device to "enable" it for wireless charging. While this is a natural intermediate, for consumers where is the benefit? Real progress can only be made by adoption of a standard which can be implemented in devices straight from the factory.
15 comments.
Attachments
| title | type | size |
|---|---|---|
| Class-D Witricity Design | application/octet-stream | 3.356 kbytes |
| Simple Witricity Design | application/octet-stream | 3.050 kbytes |
| Experimental Push-Pull TX Oscillator | application/octet-stream | 2.151 kbytes |