A matter of light, Part 4 --- PWM dimming
By Sameh Sarhan and Chris Richardson, National Semiconductor
In part one of this series, we looked at the basics of LED lighting sources and their driving requirements. In part two, we discussed why a constant-current buck converter should be your first preference when it comes to switch-mode LED drivers. In part 3, we investigated larger LED displays and the applications space for other converter topologies. Here in the concluding part of this series, the authors take a look at how to best implement the dimming function.
Whether you drive LEDs with a buck, boost, buck-boost or linear regulator, the common thread is drive circuitry to control the light output. A few applications are as simple as ON and OFF, but the greater number of applications call for dimming the output between zero and 100 percent, often with fine resolution. The designer has two main choices: adjust the LED current linearly (analog dimming), or use switching circuitry that works at a frequency high enough for the eye to average the light output (digital dimming). Using pulse-width modulation (PWM) to set the period and duty cycle (Fig. 1) is perhaps the easiest way to accomplish digital dimming, and a buck regulator topology will often provide the best performance.


Figure 1: LED driver using PWM dimming, with waveforms.
PWM dimming preferred
Analog dimming is often simpler to implement. We vary the output of the LED driver in proportion to a control voltage. Analog dimming introduces no new frequencies as potential sources of EMC/EMI. However, PWM dimming is used in most designs, owing to a fundamental property of LEDs: the character of the light emitted shifts in proportion to the average drive current. For monochromatic LEDs, the dominant wavelength changes. For white LEDs, the correlated color temperature (CCT) changes. It's difficult for the human eye to detect a change of a few nanometers in a red, green, or blue LED, especially when the light intensity is also changing. A change in color temperature of white light, however, is easily detected.
Most white LEDs consist of a die that emits photons in the blue spectrum, which strike a phosphor coating that in turn emits photons over a broad range of visible light. At low currents the phosphor dominates and the light tends to be more yellow. At high currents the blue emission of the LED dominates, giving the light a blue cast, leading to a higher CCT. In applications with more than one white LED, a difference in CCT between two adjacent LEDs can be both obvious and unpleasant. That concept extends to light sources that blend light from multiple monochromatic LEDs. When we have more than one light source, any difference between them jars the senses.
LED manufacturers specify a certain drive current in the electrical characteristics tables of their products, and they guarantee the dominant wavelength or CCT only at those specified currents. Dimming with PWM ensures that the LEDs emit the color that the lighting designer needs, regardless of the intensity. Such precise control is particularly important in RGB applications where we blend light of different colors to produce white.
From the driver IC perspective, analog dimming presents a serious challenge to the output current accuracy. Almost every LED driver uses a resistor of some type in series with the output to sense current. The current-sense voltage, VSNS, is selected as a compromise to maintain low power dissipation while keeping a high signal-to-noise ratio (SNR). Tolerances, offsets, and delays in the driver introduce an error that remains relatively fixed. To reduce output current in a closed-loop system, VSNS, must be reduced. That in turn reduces the output current accuracy and ultimately the output current cannot be specified, controlled, or guaranteed. In general, dimming with PWM allows more accurate, linear control over the light output down to much lower levels than analog dimming.
Dimming frequency vs. contrast ratio
The LED driver's finite response time to a PWM dimming signal creates design issues. There are three main types of delay (Fig. 2). The longer these delays, the lower the achievable contrast ratio (a measure of control over lighting intensity).


Figure 2: Dimming delays.
As shown, tn represents the propagation delay from the time logic signal VDIM goes high to the time that the LED driver begins to increase the output current. In addition, tsu is the time needed for the output current to slew from zero to the target level, and tsn is the time needed for the output current to slew from the target level back down to zero. In general, the lower the dimming frequency, fDIM, the higher contrast ratio, as these fixed delays consume a smaller portion of the dimming period, XXXXXXXe lower limit for fDIM is approximately 120 Hz, below which the eye no longer blends the pulses into a perceived continuous light. The upper limit is determined by the minimum contrast ratio that is required.
Contrast ratio is typically expressed as the inverse of the minimum on-time, i.e.,
CR = 1 / tON-MIN : 1
where tON-MIN = tD + tSU. Applications in machine vision and industrial inspection often require much higher PWM dimming frequencies because the high-speed cameras and sensors used respond much more quickly than the human eye. In such applications the goal of rapid turn-on and turn-off of the LED light source is not to reduce the average light output, but to synchronize the light output with the sensor or camera capture times.
Dimming with a switching regulator
Switching regulator-based LED drivers require special consideration in order to be shut off and turned on at hundreds or thousands of times per second. Regulators designed for standard power supplies often have an enable pin or shutdown pin to which a logic-level PWM signal can be applied, but the associated delay, tD, is often quite long. This is because the silicon design emphasizes low shutdown current over response time. Dedicated switching regulations for driving LEDs will do the opposite, keeping their internal control circuits active while the enable pin is logic low to minimize tD, while suffering a higher operating current while the LEDs are off.
Optimizing light control with PWM requires minimum slew-up and slew-down delays not only for best contrast ratio, but to minimize the time that the LED spends between zero and the target level (where the dominant wavelength and CCT are not guaranteed). A standard switching regulator will have a soft-start and often a soft-shutdown, but dedicated LED drivers do everything within their control to reduce these slew rates. Reducing tSU and tSN involves both the silicon design and the topology of switching regulator that is used.
Buck regulators are superior to all other switching topologies with respect to fast slew rates for two distinct reasons. First, the buck regulator is the only switching converter that delivers power to the output while the control switch is on. This makes the control loops of buck regulators with voltage-mode or current-mode PWM (not to be confused with the dimming via PWM) faster than the boost regulator or the various buck-boost topologies. Power delivery during the control switch's on-time also adapts easily to hysteretic control, which is even faster than the best voltage-mode or current-mode control loops. Second, the buck regulator's inductor is connected to the output during the entire switching cycle. This ensures a continuous output current and means that the output capacitor can be eliminated. Without an output capacitor the buck regulator becomes a true, high impedance current source, capable of slewing the output voltage very quickly. Cuk and zeta converters can claim continuous output inductors, but fall behind when their slower control loops (and lower efficiency) are factored in.
Faster than the enable pin
Even a pure hysteretic buck regulator without an output capacitor will not be capable of meeting the requirements of some PWM dimming systems. These applications need high PWM dimming frequency and high contrast ratio, which in turn requires fast slew rates and short delay times. Along with machine vision and industrial inspection, examples of systems that need high performance include backlighting of LCD panels and video projection. In some cases the PWM dimming frequency must be pushed to beyond the audio band, to 25 kHz or more. With the total dimming period reduced to a matter of microseconds, total rise and fall times for the LED current, including propagation delays, must be reduced to the nanosecond range.
Consider a fast buck regulator with no output capacitor. The delays in turning the output current on and off come from the IC's propagation delay and the physical properties of the output inductor. For truly high speed PWM dimming, both must be bypassed. The best way to accomplish this is by using a power switch in parallel with the LED chain (Fig. 3). To turn the LEDs off, the drive current is shunted through the switch, which is typically an n-MOSFET. The IC continues to operate and the inductor current continues to flow. The main disadvantage of this method is that power is wasted while the LEDs are off, even through the output voltage drops to equal the current sense voltage during this time.


Figure 3: Shunt FET circuit, with waveforms.
Dimming with a shunt FET causes rapid shifts in the output voltage, to which the IC's control loop must respond in an attempt to keep the output current constant. As with logic-pin dimming, the faster the control loop, the better the response, and buck regulators with hysteretic control provide the best response.
Fast PWM with boost and buck-boost
Neither the boost regulator nor any of the buck-boost topologies are well suited to PWM dimming. That's because in the continuous conduction mode (CCM), each one exhibits a right-half plane zero, which makes it difficult to achieve the high control loop bandwidth needed in clocked regulators. The time-domain effects of the right-half plane zero also make it much more difficult to use hysteretic control for boost or buck-boost circuits. In addition, the boost regulator cannot tolerate an output voltage that falls below the input voltage. Such a condition causes a short circuit at the input, and makes dimming with a parallel FET impossible. Among the buck-boost topologies, parallel FET dimming is still impossible or at best impractical due to the requirement for an output capacitor (the SEPIC, buck-boost and flyback), or the uncontrolled input inductor current during output short circuits (Cuk and zeta). When true fast PWM dimming is required, the best solution is a two-stage system that uses a buck regulator as the second, LED driving stage. When space and cost do not permit this approach, the next best choice is a series switch (Fig. 4).


Figure 4: Boost regulator with series DIM switch.
LED current can be shut off immediately. On the other hand, special consideration must be given to the system response. Such an open circuit is in effect a fast, extreme unloading transient that also disconnects the feedback loop and will cause the regulator's output voltage to rise without bound. Clamping circuits for the output and/or the error amplifier are required to prevent failure due to over-voltage. These clamps are difficult to realize with external circuitry, hence series FET dimming is practical only with dedicated boost/buck-boost LED driver ICs.
In summary, proper control of LED lighting requires careful attention right from the start of the design process. The more sophisticated the light source, the more likely that PWM dimming will be used. This in turn requires the system designer to carefully consider the LED driver topology. Buck regulators offer many advantages for PWM dimming. If the dimming frequency must be high, or the slew rates must be fast, or both, then the buck regulator is the way to go.
About the authors
Sameh Sarhan is a staff applications engineer for the Medium Voltage/High Voltage Power Management group in Santa Clara, CA. He has been involved with power electronics in various forms since 1998, having worked for FRC Corp. and Vicor Corp. His experience includes the design of hard/soft switching power supplies from a few watts to 600 watts. Sameh received a bachelor's degree in electronics engineering in 1996 from Cairo University (Egypt).
Chris Richardson is an applications engineer in the Power Management Products group, Medium and High Voltage Division. His responsibilities are divided between lab work, bench evaluation of new ICs, written work such as datasheets and applications notes, and training for field engineers and seminars. Since joining National Semiconductor in 2001, Chris has worked mainly on synchronous buck controllers and regulators. In the last three years he has focused on products for the emerging high brightness LED market in the automotive and industrial areas. Chris holds a BSEE from the Virginia Polytechnic Institute and State University.