Dear ts-friends, I have an "almost seasonal" signal. It's human respiration pressure, where the respiration signal is regular but non-sinusoidal, with a slightly drifting period. Overlaid on the signal are non-periodic twitches (think weak hickups) with amplitudes 2-5 times higher than the respiration signal. We have used adaptive filtering in a similar case before, but no twitch-free reference is available in the present case. I want to do something like stl, allowing some freedom to drift for the seasonal component. Robust processing is a must, the hickup should not affect processing. Robust autoregressive processing in package wle looked like a good choice, but I could not get the model to converge with my data set. Any ideas .... Dieter Menne