Imagine getting a light bulb as a present. LED replacements for screw-in bulbs are so rare and expensive, I actually did get one last Christmas. Yet we’re approaching a crossover point where LEDs will become common enough and cheap enough that they will take over a great deal of the burden of lighting the world.
Not that we simply will be screwing new bulb-shaped objects with LEDs in them into old fixtures. It is no longer necessary to treat light sources as disposable. Pretty soon, lamps (including the part that lights up) will outlive their owners. Also, since LEDs turn on and off instantly and power-cycling doesn’t involve a method that shortens their life, we can turn LED lighting on only when we need it.
That could be a potential advantage for astronomers who use optical telescopes and a potential source of huge energy savings for cities and building owners. It even turns out that police and private security operators like nighttime lighting that only turns on in response to motion sensors because it helps tell them where the bad guys are. Still, this technology is truly in its infancy, and it takes some getting used to.
When any diode is in its conduction state, whenever an electron and a hole recombine, some energy is released in the form of photons. The color of the light (the energy of the photon) depends on the energy band gap of the semiconductor material. Aluminum gallium arsenide (InGaAs) and other materials yield red. Indium gallium nitride (InGaN) produces green. Zinc selenide (ZnSe) leads to blue.
It’s possible to combine red, green, and blue diodes to produce white light. However, the high-brightness (HB) white diodes we’ve become so familiar with combine a blue InGaN diode with a yellow phosphor, usually cerium-doped yttrium aluminum garnet (Ce3+:YAG).
An interesting and useful thing happens with white LEDs that use phosphors. During a quantum effect called Stokes shift, a photon emitted by the phosphor photon has less energy than the photon it absorbed from the blue LED (Fig. 1). In an HB white LED, a fraction of the blue light is Stokes shifted to a longer wavelength. This is a good thing, as it allows LED manufacturers to use a number of phosphor layers of different colors, broadening the emitted spectrum, which effectively raises the color rendering index (CRI) of the LED.
That is, the light shouldn’t merely look white when it comes out of the LED. It should, as far as possible, accurately reflect all the colors of the objects it shines on. CRI, which is a rather complicated characteristic to measure, is a specification that indicates how closely the colors of an object being illuminated artificially look like those colors when viewed under real sunlight, which is what we earthlings consider the standard for “white” light.
A good CRI comes at a price with phosphor-based white LEDs, because the Stokes shift means they exhibit a lower efficiency than single-color LEDs. But for most general illumination applications, good CRI trumps efficiency.
A curious parallel to Moore’s law called Haitz’s law, named after Roland Haitz, says commercial LED maximum light output doubles every 36 months or so. This has an important implication for product development. Essentially, whatever light output one can expect from today’s most expensive LEDs will be available from mainstream LEDs a year and a half from now. State-of-the-art mechanical designs, then, should scale to fewer and fewer LEDs through future product generations.
One of the factors driving Haitz’s law relates to the difficulty of getting photons out of the diode semiconductor material, which tend to have high refractive indices. If a photon can’t cross the interface between the semiconductor material and the air (or vacuum) surrounding it, it’s reflected back into the material and absorbed.
If the semiconductor material were cubic in shape, it would only emit light more or less perpendicular to one face or another of the cube. (Think back to Physics 102 and the part of the chapter on optics that dealt with “critical angle.”) Thus, simply dicing a wafer of LEDs and treating the chips as if they were just some more semiconductor devices would be unsatisfactory.
If one weren’t dealing with epitaxial diode materials deposited on a flat substrate, one might think of emulating a diamond cutter and faceting the material. But it’s more practical to pot the LED in a transparent plastic material with a refractive index between the indices of the semi material and air while shaping the glob of plastic into something more spherical, or hemispherical, that increases the critical angle at both interfaces.
BASIC LED DRIVING
Driving LEDs ought to be simple. They’re diodes, they have a certain forward voltage drop, and their light output depends on current, for which there is a do-not-exceed value for any given diode. That seems like a manageable set of parameters, doesn’t it? But then it starts getting complicated. To start with, as with conventional diodes, LED current varies exponentially with the voltage. (That’s the Shockley diode equation, which has a nice explanation at http://en.wikipedia.org/wiki/Shockley_diode_equation#Shockley_diode_equation.)
Suffice it to say that one is not dealing with Ohm’s law here. A small change in voltage can cause a large change in current. That’s why, in most cases, LEDs are driven with constant-current sources. Those cheap pocket flashlights that run on stacks of watch batteries are a notable but not appealing exception. Those batteries are expensive, and they don’t last long. Quality flashlights use conventional batteries and tiny boost-converter drivers. They’re expensive, but they can make a conventional AAA alkaline battery last for months.
Before one can even consider driver circuits, an early design decision involves configurations for interconnecting LEDs in arrays. Pocket flashlights aside, few applications use only a single LED. Whether it’s for a screen backlighting array or for LEDs in a streetlight or a replacement for an incandescent or fluorescent lamp, most designs need more than one LED.
One of the first decisions a designer must make is whether to drive the LEDs in series, in parallel, or as a parallel array of strings. Generally, it isn’t a good idea to drive a number of single LEDs in parallel because it can lead to non-uniform current sharing, even when the LEDs are all rated for the same forward-voltage drop.
So that leads to driving LEDs in series, which introduces the question of what happens when a single LED fails-open. This can be solved somewhat expensively by providing parallel Zener diodes or silicon-controlled rectifiers (SCRs) across each LED. SCRs are more attractive because they dissipate less power if they have to conduct around the failed LED.
Or, perhaps the optimum strategy is to acknowledge that a single dark LED is every bit as bad as a whole string of dark LEDs and create a robust thermal design that minimizes the chances of a thermal-stress-related failure of any LED during the anticipated product lifetime.
If a designer settles on multiple parallel strings, a separate driver for each string will be more expensive than fewer drivers (ideally one) with enough output capacity to drive multiple parallel strings of LEDs.
The use of multiple strings tends to even out the current-sharing problem that led to the decision to use strings in the first case, but it is still necessary to use a ballast resistor for each string. Designers can calculate a resistor value, assuming as a target a ±10% variation in forward voltage drop across the string and a need to match the currents in each parallel string so they’re within ±20%.
Assume that, for each string, the sum of the LED forward drops plus the voltage across the ballast resistor should equal 80% of the nominal output voltage of the driver. From that, it’s possible to calculate the ballast resistance and the current rating of the constant-current driver. An easier alternative is to buy the series/parallel string arrays matched and cross-connected, as in Philips’ Luxeon Flood products.
Apart from the light output and the spacing of the array and even evenness of the lighting effect, there is little difference between a general lighting array and a backlighting array. That changes, however, when dynamically matching the backlight intensity to moving video images.
Dynamic backlighting, creating local variations in backlight intensity that correspond to the luminance value of the image on the screen being backlit, is growing in popularity for two reasons.
First, it increases the apparent dynamic range (or contrast ratio) of the images being viewed. Second, by throttling back some of the LEDs in the backlight, it reduces overall energy consumption. The first effect is believed to have a strong effect on customers choosing high-end TV receivers by comparing displays side by side in a store (Fig. 2). The second is more of a way to extend battery life in mobile devices.
I first saw dynamic backlighting demonstrated at an NXP press briefing, and it took me back to my days in college when I worked during the summer at a TV station. I knew lots of theory on my first day on the job, but nothing practical. So, they sat me down behind a camera-control unit and showed me the knobs that controlled the absolute black and white levels.
A human operator was necessary because the station’s format was old movies. That complicated TV camera operation because Hollywood directors like to take advantage of the enormous contrast ratio of film to light scenes “artistically,” which was far beyond what old RS-170 video could manage. My job that first day was to nullify their artistic efforts.
It was a case of déjà vu, four decades later, to be in an NXP conference room watching some ICs drive the LED backlighting on a Hitachi HDTV to do something at the display end of the signal chain akin to what I had once done at the camera end, albeit with less insult to the film director’s original efforts.
One significant consideration associated with dynamic backlighting is granularity. Each backlight LED illuminates a large number of pixels, so the variations in light intensity must be spread gradually over several rows and columns of LEDs. Even so, that’s better than anything I ever accomplished with gain and setup controls.
Pulse-width modulation (PWM) can control LED light output. To implement dynamic backlighting, drivers require a high-bandwidth control channel. Lots of innovation arises from the need to make LED replacements for incandescent and fluorescent light sources respond to dimming input from legacy triac and SCR dimmers. This is one of those situations where a legacy becomes a burden to the heirs.
The dimming part is easy with a constant-current dc power source. The challenge lies in making the dimming and brightening happen in response to a controller that was never made for LEDs. Basic dimming, then, is simply a matter of varying the duty cycle of a pulse-width modulated constant-current square wave that is switched quickly enough to avoid the perception of flicker.
The rule of thumb is that anything over 100 Hz (twice the European mains frequency) is sufficient. In fact, European regulators are starting to be concerned about the combination of such relatively slow rates and short duty cycles when power factor correction (PFC) in the ac-dc front end of the LED driver adds harmonics to the waveform being applied to the LEDs.
Before addressing anything that subtle, consider the more fundamental problem associated with triac dimmers, which is the way they control the brightness of a simple incandescent bulb by interrupting part of each half-cycle of the ac waveform. That’s trivial if the load is the filament of an light bulb, but not simple at all if the load is a constant-current driver IC (see “High-Brightness White LEDs Light The Way To Greener Illumination” at www.electronicdesign.com).
For example, National Semiconductor’s LM3445 buck controller is a triac-friendly dimmable driver. In fact, National’s engineers developed a proprietary constant off-time method to maintain constant current through the string of LEDs for it (Fig. 3). This is really an instance of pulse-frequency modulation because, with constant off-time, the on-time becomes the only variable. It’s then easy to control on-time by varying the switching frequency.
The LM3445 features a Zener bridge at the input. A “valley-fill” circuit after the bridge smooths the chopping action and allows the buck regulator that follows to draw power even while the triac is in cutoff. It also provides passive PFC.
External to the IC, the circuit requires a “bleeder” resistor that emulates the incandescent bulb filament resistance that the triac would see in a conventional lighting circuit. “Bleeder” usually refers to a resistor across the output capacitor of an ac supply. Here, it dissipates the small current that flows through the triac in its OFF state.
At the same node as the bleeder resistor, there is an external circuit with a 15-V Zener and a pass transistor. The pass transistor “stands off ” most of the rectified line voltage so the voltage on the IC’s sensing pin varies with the rectified line voltage whenever it drops below the Zener’s threshold.
The LM3445 delivers a 10% to 100% dimming range, relative to triac-dimmer firing angles between 45° and 135° of the ac line waveform. That part of the circuit is based on a ramp generator and comparator. The ramp comparator’s output drives a common-source N-channel MOSFET through a Schmitt trigger, and the MOSFET’s drain voltage is proportional to the duty cycle of the triac dimmer. There is another circuit for setting PWM off time. A resistor sets the actual current that drives the LED current.
Linear Technology’s LT3799 targets similar triac-controlled applications (Fig. 4a). Its designers put a great deal of effort into the power-conversion stages, starting with the ac-dc stage, which was implemented with a flyback topology in contrast to National’s boost-converter approach. Their objective was to keep power factor in the rectifier section as close to unity as possible across the dimming range, without passing the offending harmonics down the conversion chain.
The point of the strong emphasis on PFC was to meet the requirements of the European IEC61000-3-2 standard, which specifies power factor in terms of harmonics of the line frequency, rather than as a phase angle between voltage and current on the power line. The two definitions of power factor are equivalent, and both relate to load reactance, but in different ways.
Every rectifier circuit includes a large output capacitance that stores and smooths the pulsating dc from the rectifier diodes. As a result of that smoothing, the load draws energy from the line for a portion of each ac-input cycle. For the rest of each cycle, it draws energy from the capacitor. While the ac line voltage is sinusoidal, the ac line current is spiky, with a Fourier series at multiples of the ac line frequency.
The IEC61000-3-2 standard specifies permissible maxima for each harmonic, up to the 32nd. Linear’s chip is designed to meet that requirement, maintaining a minimum power factor of 0.97, using active PFC.
One of the advantages of approaching PFC Linear’s way is that it actually simplifies triac-based dimming. The LT3799 employs an isolated flyback topology, with the transformer secondary output fed back to the primary side for control, as in any flyback. Unlike most flyback stages, however, this one uses a third transformer winding, rather than an optocoupler, for isolation.
In operation, the chip uses the external MOSFETs peak current information, derived from a sense resistor, to calculate the converter’s output current. (See pages 9 and 10 of the LT3799’s datasheet at http://cds.linear.com/docs/Datasheet/3799f.pdf.) When the main power is off, the LT3799 can also use the voltage on the third winding to detect and report any open LED string. In normal operation, the third winding not only senses the output voltage, it also supplies power to the IC.
To achieve that 0.97 or better power factor, the PFC circuitry operates in critical conduction mode, i.e., just on the edge between continuous and discontinuous conduction mode.
After all that complexity in the front end, the actual process of dimming the LED string based on the duration of each half cycle of the applied ac waveform that the triac passes is simple but elegant. According to the datasheet, triac dimmers aren’t ideal switches when they’re turned off because they allow milliamps of current to flow through them.
In fact, that’s a signature of triac. Dimmer controllers from other companies provide a resistive load that behaves like the filament of an incandescent light bulb to cause the triac to trigger. In the Linear driver, instead of turning the main power MOSFET off when the triac is off, it is kept on to properly load the triac. When the triac does turns on, the driver detects this and enables the control loop.
When the dimmer triac is OFF, leakage current still ?ows through its internal ?lter to the LT3799. That current charges up the IC’s input capacitor and would cause random switching and LED ?icker, except the transformer primary winding is used as a bleeder. The MOSFET gate signal goes high so the MOSFET is ON when the triac is off, bleeding off the leakage current. As soon as the triac turns on, the MOSFET seamlessly changes back into a normal power delivery device (Fig. 4b).
HEADACHES AND SEIZURES
A final caution about dimming comes from iWatt, a fabless semiconductor company whose dimmers are found in the bases of many LED replacement bulbs. The challenge in that particular niche, which iWatt has mastered, is to enable the driver to identify which particular triac controller, out of hundreds, is being used and to tailor the behavior of the lamp to it.
The company additionally says that the people in charge of the European safety standards are taking a closer look at what happens to the harmonics that are suppressed in the ac-dc stage of an LED driver for general lighting applications. According to iWatt, these harmonics can show up as a form of noise on the constant current output line, which is not bad in itself, except that when light output is very dim the harmonics may create a subliminal flicker that could trigger headaches or even seizures in susceptible people. Research is in the early stages now, but it could result in more dash-numbers for the IEC61000 standard.