Buck regulators can be very efficient, maybe 90–95%, especially if the load is a long string of LEDs with a moderately high-forward voltage (i.e., high-duty cycle). This is because the power dissipation in the flywheel diode is a smaller proportion of the total power because the flywheel diode only conducts during the MOSFET off-time, which is a smaller proportion of the total switching cycle. The MOSFET dissipates power during the on-time, when it is conducting, but the voltage drop across the MOSFET switch is usually much lower than the forward drop of a fast rectifier. Synchronous buck regulators, which use a MOSFET in place of the flywheel diode, are sometimes used in low voltage applications and their efficiency can be higher than 95%.

To operate correctly there must be some ripple in the output current. The output current needs to reduce enough to allow the current sense comparators to be reset. The output ripple current ∆i_o is normally designed to be 20–30% of the nominal output current, i_o. This ensures that the output current falls far enough in each cycle so that noise in the current sense comparator has little effect. If the ripple current is below 10% of i_o, the switching of the MOSFET can be erratic.

The peak current in the LED string is $i_{\text{o}\text{max}}=\frac{V_{\text{th}}}{R_{\text{sense}}}$

The average output current in the LED string (i_o) is given by the equation:

Here V_th is the current sense comparator threshold, and R_sense is the current sense resistor. The average current is the peak current minus half the ripple.

The ripple current can introduce a peak-to-average error in the output current setting that needs to be accounted for. When the constant off-time control technique is used, the ripple current is nearly independent of the input supply voltage variation. Therefore, the output current will remain unaffected by the varying input voltage.

Adding a filter capacitor across the LED string can reduce the output current ripple in the LED, without reducing the inductor ripple current. This means that a lower-value inductor could be used without increasing the LED ripple current. However, keep in mind that the peak-to-average current error is affected by the variation of the MOSFET off-time, T_off. Therefore, the initial output current accuracy might be sacrificed with large-ripple current levels in the inductor.

Alternatively, if the inductor value is unchanged, adding a filter capacitor across the LED terminals allows an apparently more “constant” current in the LED. This capacitor also reduces EMI at the output by providing a bypass path for any switching current spikes. Switching current spikes could also reduce the LED lifetime, so a filter capacitor reduces this risk.

Another important aspect of designing an LED driver is related to certain parasitic elements of the circuit. These include distributed coil capacitance of the inductor C_L, junction capacitance, C_J, and reverse recovery current in the flywheel diode, as well as capacitance of the printed circuit board traces C_PCB and output capacitance C_drain of the MOSFET. These parasitic elements affect the efficiency of the switching converter because they cause switching losses. These parasitic elements are shown in Fig. 10.2.

Parasitic elements cause a switch-on current spike that could potentially cause false triggering of the LED driver IC’s current sense comparator. This can be mitigated against by an RC filter fitted between the MOSFET source and the current sense (CS) pin. Minimizing parasitic elements is essential for efficient and reliable operation of the buck converter.

Coil capacitance of inductors is typically provided in the manufacturer’s data books either directly or in terms of the self-resonant frequency (SRF).

Here L is the inductance value, and C_L is the coil capacitance. Charging and discharging this capacitance every switching cycle causes high-current spikes in the LED string. Therefore, connecting a small capacitor C_o (∼10nF) across the LED string is recommended to bypass these spikes, as mentioned earlier.

Using an ultrafast rectifier flywheel diode is recommended to achieve high efficiency and reduce the risk of false triggering of the current sense comparator. When the MOSFET turns on, the flywheel diode’s state changes from “forward conduction” to “off” (reverse bias). This change of state in the flywheel diode cannot happen immediately because electrical charges have to move from one place to another inside the semiconductor material, which takes time. There is always a reverse recovery current flowing in the opposite direction for a short period, t_rr, during this change of state.

Using diodes with shorter reverse-recovery time, t_rr, and lower junction capacitance, C_J, improves performance. In low-voltage applications a Schottky diode can be used, but in high-voltage applications a diode with less than 75-ns reverse recovery time (preferably less than 50 ns) should be used. There are high-voltage Schottky diodes, called silicon carbide (SiC), but these are high in cost.

The reverse voltage rating V_R of the diode must be greater than the maximum input voltage of the LED lamp. The forward voltage drop of diodes with very fast-recovery times is sometimes relatively high and can lead to high-conduction losses, so also consider this when making a diode selection. Low-voltage Schottky diodes have a typical forward voltage between 0.4 and 0.5 V, but this varies between types and the designer should check the specific parts’ datasheet.

The total parasitic capacitance present at the DRAIN output of the MOSFET can be calculated as:

When the switch turns on, the total parasitic capacitance C_P is discharged into the DRAIN output of the MOSFET. The discharge current amplitude is limited to the MOSFET saturation current level, so MOSFETs with a high–on-resistance and a lower-saturation current can sometimes produce lower overall losses. This is especially true if the duty cycle is small because the switch is conducting for a small proportion of the time and hence the conduction losses will not be significant. Note that the saturation current in a MOSFET becomes lower at increased junction temperature.

The duration of the leading edge current spike can be estimated as:

To avoid false triggering of the current sense comparator, C_P must be minimized in accordance with the following expression:

The factor t_blank min is the minimum blanking time, which depends on the control IC and is in the order of 300 ns. When the MOSFET gate drive is activated, the control IC disables the current sense input for this time period, to avoid false triggering from the switch-on current surge, previously described. The factor V_in max is the maximum instantaneous input voltage.

Discharging the parasitic capacitance C_P into the DRAIN output of the MOSFET is responsible for the bulk of the switching power loss. It can be estimated using the following equation:

where f_s is the switching frequency, i_sat is the saturated DRAIN current of the MOSFET. The switching loss is the greatest at the maximum input voltage.

The switching frequency of a buck converter having constant off-time operation is given by the following:

where η is the efficiency of the power converter. This value for f_s based on typical values for V_in and V_o can be used in the previous equation if a value of constant switching frequency is not available.

The switching power loss associated with turn-off transitions of the DRAIN output can be disregarded. Due to the large amount of parasitic capacitance connected to this switching node, the turn-off transition occurs essentially at zero voltage.

Conduction power loss in the MOSFET can be calculated as

where D = V_o/(η·V_in) is the duty ratio, R_on is the ON resistance.

As another example, in flat-screen television backlighting high-power red, blue, and green (RGB) LEDs are used to create a white light that exactly matches the LCD and produces true colors. In this application a boost converter powered from a 12- or 24-V DC supply is used to drive many 350-mA LEDs connected in series, with a forward voltage in the range 40–80 V.

Boost regulators should always be provided with overvoltage protection, in case the LED load is disconnected. Otherwise the output voltage will continue to rise and eventually cause component breakdown. In safety electrical low-voltage systems, the output voltage would normally be kept below 42 V.

The two types of converters often found in boost–buck applications are known as SEPIC and Ćuk. These converters are similar, but the Ćuk converter has an inverted output, which means that the LED anode is connected to the ground rail. Like boost converters, overvoltage protection should be provided to prevent excessively high voltage in case of an open-load condition.

The Ćuk is good for low-EMI emissions because there are inductors in series with the input and the output. Both inductors operate in continuous conduction mode, and high-frequency signals at the central node, where the switching takes place, are automatically filtered. Shunt capacitors across the input and output strengthen this filtering, and provide a low-impedance path for the circulating currents. Consequently, Ćuk circuits require minimal external filtering. In the SEPIC circuit, the output inductor is connected to ground, so EMI is greater than for an equivalent Ćuk.

Sometimes common-mode chokes are added at the input side of a boost–buck converter, to reduce the radiated signals from the whole circuit. Common-mode chokes are only required on the output side if the length of wire to the LED load is more than about 0.5-m long.

In semiconductor circuits powered from the AC mains, a bridge rectifier converts the AC power into DC. The current through the bridge rectifier tends to occur close to the peak voltage, as shown in Fig. 10.4 because charging of a large-smoothing capacitor takes place each half cycle. These short-charging current pulses at the crest of each input cycle cause the power factor to be typically in the 0.3–0.6 range. PFC is an active or passive circuit designed to correct phase errors and reduce harmonics, and make the power factor closer to 1. PFC is required in higher-power LED drivers.

A circuit having a good power factor, approaching 1, has an input current that has low-harmonic content with a wave shape that closely follows the sinusoidal input voltage. Circuits that provide a good power factor were described in Chapter 8.

Fly-back converters allow an isolated LED driver design with about 80% efficiency, but have added cost and complexity. If a wide tolerance can be accepted for the current regulation, a simpler and cheaper circuit can be built. High accuracy requires isolated feedback, usually via an optocoupler and employing an adjustable shunt regulator, such as a TL431 or similar, along with a few passive components.

Fly-back converters have the advantage of stepping up or down the output voltage compared with the supply (buck–boost). This also applies to the single-winding inductor version, although as the same winding is used for the primary and secondary side, the turns ratio is 1:1 and the design specifications are more restricted than for a two-winding inductor. A single-winding inductor is usually much lower in cost.

A fly-back, by definition, is a discontinuous conduction mode converter; energy is taken from the power supply in the first step and then transferred to the output in the second step, as shown in Fig. 10.5. This means that EMI must be carefully filtered at both the input and output. The output requires a large-storage capacitor to maintain current flow in the LEDs when the converter is on the first step. Dimming the LED light by pulse width modulation (PWM) of the current is very difficult because the stored energy in the capacitor tries to maintain current flow; thus only a modest dimming range is possible.

Note that the LED current in this circuit is not controlled. Only the output power is controlled, so the LED current will depend on the LED string forward voltage, which is temperature dependent. Thus the LED current will change with LED temperature.

For AC mains applications, an NTC thermistor designed to carry high current is often used. In the active state, the flowing current warms the thermistor and hence the resistance falls to a low level to reduce losses (Fig. 10.6).

For DC applications, an active inrush limiter is more common. This is because the losses can be minimized during normal operation, when inrush limiting is not needed. There are several manufacturers of active inrush ICs, but Microchip’s HV101 uses a unique scheme, whereby the Miller Effect is used to control the slow turn on of the MOSFET. An active inrush circuit using the HV101 is shown in Fig. 10.7.

For example, a circuit for driving one or two power LEDs from the AC mains could use a double-buck topology. But typical applications for this circuit are inside lamp housings, where an electrolytic capacitor cannot be used because of short lifetime or physical size. But using a polyester film capacitor means that the voltage dips between switching cycles. As the output power is normally constant, this means that the input current will peak as the input voltage dips. The peaks in input current give rise to considerable EMI and mean that the power factor is very poor. If the output current was controlled, that is, reduced as the supply voltage dipped, the input current would remain constant when switching. The addition of a Zener diode in series with the supply to the controller IC would further improve the power factor.

Soft-start can also be implemented by connecting an RC filter to the analog dimming input in control circuits that have this function (e.g., linear dimming pin of HV9910). The current level starts low and grows as the capacitor, C, charges. Clearly we need a method of discharging this capacitor reasonably quickly when the power to the IC is disconnected; diode D connected to the V_DD will discharge capacitor C when the V_DD voltage drops (Fig. 10.8).

Slope compensation is needed if the following three conditions are met: (1) the switching is constant frequency, (2) the duty cycle is more than 50%, and (3) current mode control is used. Some DC/DC converters with a constant voltage output use voltage mode control; however, all LED drivers have a constant current output and use current mode control. In Chapter 5, I described buck controllers using constant off-time switching, to avoid the need for slope compensation. The disadvantage, in some applications, was that the switching frequency covered a wide range; so LED ripple current and EMI emissions were more difficult to control.

In Fig. 10.9, the switching inductor current is shown on a graph. In this case we are looking at a system with a duty cycle about 25%. The heavy black line shows the nominal current levels, with i_limit being the peak current level used to turn off the switch (peak level sensing). A thinner black line shows the effect of a disturbance, for example, a step positive-going increase in the supply voltage. Notice that the current limit is reached earlier, so the duty cycle is smaller, but the off-time is longer (as it is a fixed switching frequency) and the current drops below the original starting level.

On the second-switching cycle the current starts at a lower level than during the first-switching cycle, so the inductor current takes longer to reach the current limit. Now the on-time is longer, so the duty cycle is greater, and the off-time is shorter. At the end of the second cycle, the current is higher than the nominal level, but only by a small amount. On the following cycles, this difference dwindles to nothing, so the system is stable.

In Fig. 10.10, the graph shows the current in the switching inductor of a system where the duty cycle is 75%. As before, the heavy black line shows the nominal current levels, with i_limit being the peak current level used to turn off the switch (peak level sensing). A thinner black line shows the effect of a disturbance caused by a positive voltage step at the input. As before, the current limit is reached earlier. As the duty cycle is much higher in this case, there is a less steep slope on the graph and the disturbance causes a more significant reduction in the duty cycle. As the switching frequency is fixed, the off-time is increased significantly and the current drops to quite a low, way below the original starting level.

Thus the second cycle starts with the inductor current at a very low level and the current limit is not reached by the end of that switching cycle. Instead, the current limit is reached at some point in the next cycle. Thus the switch on-time will alternate between long and short periods, causing the circuit to operate at half the nominal switching frequency. This is sometimes called subharmonic oscillation. The ripple current will alternate between low and high levels. The average current will thus be much lower than expected and some engineers only realize that they have an instability problem when they measure the LED current.

In Fig. 10.11, a solution for unstable subharmonic oscillation is shown graphically. Instead of having a fixed current limit, the current limit is reduced over the switching period. At the start of each switching period, the current limit is maximal. At the end of each switching period, the current limit is at its minimum level. The slope of this current limit, shown by a thin line on the graph, is set to be half the downslope of the inductor current. This downslope is calculated at: Slope = 0.5 × V_out/L. The maximum slope is Slope_max = 0.5 × V_out max/L_min, which should be used in calculations.

Now, as in the previous examples, the thick black line shows the nominal current and the thinner black line shows the effect of a disturbance, such as a step increase in the supply voltage. After the input voltage disturbance, the current limit is reached earlier than with the nominal input voltage, but the sloping current limit has the effect of reducing the difference between the two levels by the end of the first-switching cycle.

At the start of the second-switching cycle, the inductor current starts at slightly below the nominal level. The difference is much less than at the beginning. Due to the sloping current limit, the difference between the two levels is much smaller after the second-switching cycle and becomes insignificant on subsequent cycles.

In practice, the slope compensation is arranged by adding a voltage to the current sense feedback signal. When the current sense voltage and the added slope compensation voltage reach a threshold level, the power switch is turned off. The switch will not be turned on again until the clock signal triggers the gate drive output from the LED driver IC, so the slope compensation ramp signal is only needed while the power switch is on. Thus a ramp signal can be created by charging a capacitor from the power switch gate drive signal. In some cases, an LED driver IC will include a constant current circuit driven from the gate output signal, so the user only has to add an external capacitor connected from a COMP pin to ground.

Now, with slope compensation added, we can see that the current limit is now lower than before. To keep the desired (original) current levels, the current sense resistor value now has to be scaled down in proportion.

The new value for R_sense can now be used with slope compensation, to give the correct (higher) current limit.