Autoregressive Order Selection for Irregularly Sampled Data

An autoregressive (AR) spectral estimator is investigated that applies the principles of a discrete-time automatic equidistant missing data algorithm to unevenly spaced data. This time series estimator approximates irregular data by a number of resampled missing data sets, with a special multi-shift slotted nearest neighbor method. It uses a slot width that is a fraction of the resampling time step. Unfortunately, resampling always causes bias. The ARMAsel-irreg algorithm estimates AR models and selects the order from a number of candidates. The order selection criterion selects the best approximation of the biased spectrum. The bias causes AR poles at higher frequencies and the selected model can have spurious high frequency poles, incompatible with the continuous character of the irregularly sampled signal