As their production costs fall, LEDs are being used more often in applications ranging from handheld devices to automotive and architectural lighting. Their high reliability (operational lifetimes of greater than 50,000 hours), good efficiency (greater than 120 Lumens/W) and nearly instantaneous response make them very attractive light sources. LEDs produce light in as little as 5 ns, compared to the 200-ms response time of an incandescent bulb. Consequently, they have been embraced by the automotive industry in brake lights.
Additionally, LEDs are increasingly being used as the primary light source in digital light projectors (DLP) and television applications, where they replace the white-arc lamp and mechanical color-wheel assembly. In DLP applications, LEDs are switched on and off at rapid rates, generating the red, green and blue color components as needed.
Driving LEDs is not without challenges, however. A controlled brightness requires driving the LED with a constant current, which must be maintained regardless of input voltage. This can be far more challenging than driving an incandescent bulb connected to a battery. A common approach is the use of a buck regulator to provide the LED drive current from a dc input voltage. However, specific LED performance requirements dictate several aspects of switching regulator design, including the method used to close the feedback loop and the choice of dimming technique.
LED I-V Characteristics
An LED has a forward I-V characteristic curve shape similar to a diode. Below the LED turn-on threshold, which is approximately 3.5 V for a white LED, very little current flows through it. Above that threshold, current flow increases exponentially for each small incremental increase in forward voltage. This allows the LED to be modeled in SPICE as a voltage source with a series resistance, but with one caveat: The model is valid only at a single operating dc current.
If the dc current in the LED is changed, then the resistance of the model should be changed to reflect the new operating current. Fig. 1 shows the measured impedance of a 1-W white LED. The graph in Fig. 1 displays the change in forward-voltage drop divided by the change in current. This is the slope of the LED's I-V curve, which effectively represents the LED's dynamic impedance for a specific drive current. Note that a 1-W LED illuminates at currents as low as 1 mA, although not very brightly.
Additionally, at large forward currents, the LED operates at a high-power level that begins to heat the device. This increases the forward-voltage drop and, therefore, the dynamic impedance. It is critical to consider the thermal environment once the LED impedance has been determined.
Ripple current in the LED can increase its power dissipation, leading to increased junction temperatures. This increase in temperature has a major impact on the life of the LED. Fig. 2 shows the relative light output from an LED as functions of operating time and junction temperature. If we establish an 80% limit on light output as the useful life of the LED, the lifetime is extended from about 10,000 hours at 74°C to 25,000 hours at 63°C.
When an LED is driven by a buck regulator, the LED often conducts the ac output-ripple current in addition to the dc current, depending on the output filter arrangement chosen. Fig. 3 quantifies the increased LED power dissipation due to the ripple-current content. Since the ripple frequency is high compared to the thermal time constant of the LED, the high peak power dissipation due to the high ripple current does not instantaneously impact the junction temperature. Instead, the junction temperature is determined by the average power.
Much of the LED's voltage drop is like a voltage source. Even at large ripple currents, there is no significant impact on power dissipation (Fig. 3). For instance, 50% ripple current (IPK-PK = IOUTMAX) adds less than 10% to the total power loss. Much above this level, ac ripple current from the supply must be reduced to limit junction temperatures in order to extend the semiconductor's rated useful life. This is because there is a resistive component to the voltage drop, and the total power dissipated by the LED is determined by:
PLED = RLED × ILEDRMS + VLEDFORWARD 3 ILEDAVG.
The power dissipation can be used together with the total thermal resistance and ambient temperature to calculate the junction temperature of the LED. A useful rule of thumb is that the semiconductor's rated useful life doubles for every 10°C the junction temperature is reduced.
Also, most designs tend toward much lower ripple currents because of inductor constraints. Most inductors are designed with a much lower ripple-current ratio (IPK-PK/IOUT < 20%). Additionally, peak current in the LED should not exceed the manufacturer's specified maximum safe operational rating.
Closing the Control Loop
Fortunately, the process of closing the current loop on an LED supply to meet these restrictions can be simpler than closing a voltage loop on a conventional power supply. This is because loop complexity can be controlled by the designer and is determined by the output filter configuration. Three possibilities for these configurations are shown in Fig. 4. These configurations are a simple inductor-only filter (A), a typical power-supply filter (B) and a modified filter (C) design.
A simple PSPICE model was built for each of the three configurations to illustrate the differences among their respective control characteristics. The switching action of the buck power FET and diode was modeled as a voltage-controlled voltage source having a gain of 10, and the LEDs were modeled as a 3-Ω resistor in series with a 6-V source. Between the LEDs and ground, a 1-Ω resistor was added to sense the current. Fig. 5 shows the results.
In circuit A, the response is that of a first-order system, which is inherently stable. The dc gain is set by the voltage-controlled voltage source, the divider formed by the LED resistance and the current-sense resistor. The pole of the system is set by the output inductor and the series resistances in the circuit. The compensator design is straightforward, using a type-two amplifier.
Circuit B has a second-order response caused by the presence of the output capacitor. This capacitor might be required if a significant amount of LED ripple current was unacceptable either due to electromagnetic interference (EMI) or heating concerns. The dc gain is the same as the first circuit; however, there is a pair of complex poles at the resonant frequency of the output inductor and capacitor. The filter's total phase shift is 180 degrees, which could lead to an unstable system if care is not taken when designing the compensation circuit. However, the compensation circuit design is similar to a conventional voltage-mode power supply and requires a type-three amplifier. Compared to circuit A, circuit B has three additional components, including the output capacitor.
In circuit C, the output capacitor has been repositioned to make the circuit easier to compensate. The ripple voltage across the LEDs is similar to circuit B. However, the inductor ripple current flows through the current-sense resistor R105 in circuit C. This must be accounted for when calculating the power dissipation. Circuit C has one zero and a pair of poles, and is nearly as easy to compensate as circuit A. It has the same dc gain as the first two circuits.
The zero is introduced by the capacitor and the LED series resistance. One of the poles is set by the output capacitor and current-sense resistor. The other pole is set by the current-sense resistor and output inductor. At high frequencies, the response is the same as circuit A in Fig. 5.
Quite often, LED applications require a dimming capability. For instance, it may be desirable to dim a display or dim architectural lights. There are two ways to accomplish this. One way is to reduce the LED current, which reduces the intensity of the emitted light. The other way is to quickly turn the LED on and off, which is perceived as a steady but dimmer light compared to an LED that is always on. Furthermore, changing the ratio of on time to the total switching-cycle time, referred to as the duty cycle, produces linear changes in the apparent light intensity. This technique is known as pulse-width modulation (PWM).
Between these two methods, the least-effective way is reducing the current, because the emitted light intensity is not completely linear with current. Additionally, the LED color spectrum tends to shift at currents below full rating. Furthermore, human perception of brightness is exponential, so dimming can require the LED current to change by large percentages. The potential impact of this relationship between drive current and emitted light intensity on circuit design is obvious.
For example, a given magnitude for ripple in the LED drive current may only cause a 3% regulation error when the LED drive current is 100% of full scale. However, this same ripple-current magnitude produces a regulation error of 30% when the LED drive current is reduced to 10% of full scale.
Therefore, dimming the current through PWM is more accurate, but it must be performed at switching speeds that exceed the response time of the human eye in order to work properly. Specifically in lighting and display applications, PWM switching should be greater than 100 Hz so the human eye does not perceive flicker.
To support this high-speed operation, the power supply driving the LED must also operate at high frequencies. For example, a 10% duty cycle in the millisecond range requires the power supply to have a bandwidth greater than 10 kHz. Fig. 6 provides an example of a buck power stage with PWM dimming using a filter stage similar to that shown in circuit A of Fig. 4. In this circuit, the LED is simply switched in and out of the circuit by FET Q1. In this manner, the control loop is always active, resulting in an extremely fast transient response (Fig. 7).
An alternate dimming method uses the controller's enable function, or soft-start capacitor, to turn the controller on and off. The current waveforms of a circuit using this method are shown in Fig. 8. Cycling the controller with a PWM input signal effectively controls the LED's brightness. Although this approach is somewhat slow when compared to bypassing the LED with a FET, it is quite easy to implement. This method is also efficient, because all output current is delivered to the LED, or an LED string, during the on time, and the circuit is shut down during the off time.
Once commanded to turn on, the controller has an inherent startup delay before current begins to flow. The controller's internally programmed rise time, or the external soft-start capacitor value on some controllers, sets the load-current rise time. This approach will have a minimum duty-cycle limit that is partially determined by the controller, the soft-start capacitor value and the output-filter values.
Often, this limit on the minimum duty cycle can be as high as 10% to 30%, and this range of operation can suffer from nonlinear brightness control. This may be unacceptable for certain lighting applications where brightness gradients are critical such as televisions, but may be usable for less-demanding applications such as automotive taillights.
Of course, the final selection for the LED-dimming scheme is determined not by the application, but by the design engineer, who must weigh all design goals. If high-frequency operation is desired, placing a FET in parallel with the LEDs provides high-speed dimming, but this method is expensive and introduces switching losses. Alternatively, the controller's enable pin and soft-start capacitor can be used to perform the dimming function with greater efficiency, but this method is relatively slow. Additionally, regardless of the dimming scheme selected, the final output-filter selection of an LED driver also impacts size, cost and EMI susceptibility.