AN3410

Control loop dynamics

Doc ID 018890 Rev 1

13/28

3

Control loop dynamics

The control loop is intentionally very slow, much slower than a normal PFC loop. The reason

is that the eye perceives abrupt changes of light intensity as flashes.

The eye has an intensity control loop of its own, which adjusts the light falling on the retina

by controlling the iris opening. The loop bandwidth is about ¼ Hz. If light level changes

occur slower than this, they are not perceived as flashes. the iris can keep up with the

change.

The LED driver control loop is set up as a continuous -20 dB/decade / 90 degrees lag

system, which gives a response to perturbations having no overshoot.

The current control loop breakpoints are:

–

A pole at zero frequency

–

A pole at 0.35 Hz (R17-C12)

–

A zero at 0.35 Hz (R25-C11)

A third pole at high frequency due to noise filter C5 (gain is well below unity at this corner -

the pole is inconsequential).

The pole due to R17 and C12 is essential to the system – switching ripple and the 120 Hz

envelope must be filtered. The R25-C11 zero cancels this pole, keeping the gain slope at -

20 dB/decade.

There are two problems that affect the control loop:

–

There is a second path in the system that gives positive feedback, the

compensation for LED voltage. Fortunately, the voltage loop gain is low due to the

low dynamic resistance of the LEDs, typically 1 Ohm per LED. Changes in the

input power can only cause very small changes in the LED voltage.

–

The R18-R21 voltage divider, necessary for the DC control scheme, gives

considerable unnecessary loop gain.

Note that R23 provides a very handy single-point gain adjustment. Increasing its value

reduces the gain of the entire loop. This is a good point for stability testing - it should be

possible to reduce the value of R23 by a factor of 2 to 4 (6-12dB) before sustained (slow!)

oscillation results.

Control loop transient response can be observed at startup (see Figure 16). The LED

current rises rapidly (a nice transient), and the control loop takes over very smoothly. The

LED current should not overshoot, and it should settle to its final level within 2 or 3 seconds.

Either excessive gain (overshoot) or unmatched time constants for the breakpoint networks

R17-C12 and R25-C11 (lumps or dips in the startup waveforms) can cause strange

behavior during startup.

Excessive gain can result in ringing or sustained oscillation - both are quite annoying.

Mismatched breakpoints can result in a dip or rise of output current about 1.5 seconds after

the unit starts. The result is not too annoying, but it should be corrected - customers demand

smooth operation.