I am using an RDR195 based design using a linkswitch PH (LNK419 for 32W operaton), setup for digital PWM dimming as explained in this post: "http://www.powerint.com/forum/power-supply-design/can-i-perform-pwm-dimming-using-linkswitch-ph" and the associated PDF. A 1:1000 dimming range is comfortably achieved, using a PWM frequency of 400 Hz. Full output is at 700mA
Oddly, we have noticed that at an approximate duty cycle of 7.5%, our LED output appears to be unstable by oscillating somewhere between 0.5 and 0.1Hz - often the frequency gradually shifts over a period of a few minutes! The attached image shows the secondary-side output voltage across the ripple-smoothing capacitors; the LED load is quite clearly showing to be 'pulsing'. This particular duty cycle appears to cause the most severe oscillations, and as duty cycle moves away from this value in either direction, the peak-peak voltage of the oscillations decrease. Outside the range of approximately 6-9% PWM duty cycle, the output voltage and LED brightness is considered sufficiently steady.
I've ensured a suitable heatsink is used and have tried removing the over-voltage circuit, with no improvements in the output oscillations. The PWM signal has been confirmed to be steady, and both a microcontroller and benchtop signal generator have been tried as the source. At lower PWM frequencies, the most severe oscillations seem to occur at lower duty cycles yet tend to show increased peak-peak voltages.
Has anyone experienced similar behaviour or have any suggestions for this odd issue? The low frequency/time constant of the output voltage make me think perhaps the feedback resistor (approx 120K) and either the bias winding or BP pin capacitor (22uF, 10uF) are possibly creating some sort of RC circuit here.