| Electronics > Projects, Designs, and Technical Stuff |
| Instability in power supply duty cycle and ripple |
| (1/1) |
| jmw:
I've been working on a 5 V -> 150 V power supply for Nixie tubes. Basic specs are 100 kHz DCM boost operation in peak current mode using a LM3478 as the controller. So far it's working as expected, and can supply 15 mA of current at 150 V. However, I found the 200k frequency adjust resistor picked by the datasheet's formula yielded a frequency of 92 kHz, so I tried replacing it with a 180k resistor, and it's now just over 100 kHz. However, there's now an instability in the duty cycle on my constant test load, and the output voltage looks unstable. Things I've noticed: increasing the power stage output capacitor gets rid of the instability, as does (of course) decreasing switching frequency. Is the most likely culprit: 1) bad loop compensation - the LM 3478 doesn't appear to have a good way to for me to measure the loop bandwidth with voltage injection, so I've compensated the controller based on a model of the error amplifier from the datasheet and an accompanying application note. 2) current mode ramp instability - seems unlikely since DCM is supposedly stable without any artificial ramp? Changing the resistor that allows additional slope compensation doesn't appear to fix it. 3) something else? Before (Rfa = 200k, fs = 92 kHz): Gate waveform Output ripple After (Rfa = 180k, fs = 103 kHz): (I've seen it much worse - appears like it's wandering up and down with some very short and long periods) |
| prasimix:
--- Quote from: jmw on December 17, 2018, 01:37:55 am ---1) bad loop compensation - the LM 3478 doesn't appear to have a good way to for me to measure the loop bandwidth with voltage injection, so I've compensated the controller based on a model of the error amplifier from the datasheet and an accompanying application note. --- End quote --- Don't see a reason why you shouldn't try to inject test signal into FB voltage divider as I done here and try to tweak compensation network connected on COMP pin. |
| Navigation |
| Message Index |