AR-modelling bispectrum estimation using data segmentation into coupled records

The authors present a variant of the third-order recursion (TOR) method for bispectrum estimation via AR modelling using third-order moments over the 1D diagonal slice. The modification is based on the segmentation of available data into coupled records instead of independent records. Its foundation resides in considering the data to the left and right of each record as not null and taking them as the data corresponding to the preceding and succeeding record, respectively. Simulation results show that this new segmentation procedure gives more precise AR parameters than the TOR and constrained third-order mean (CTOM) methods