Robust nonparametric bicoherence estimation by stepwise outlier rejection

Nonparametric bicoherence estimation is commonly achieved using a segment-averaging approach but this can be highly susceptible to occasional large transients occurring during the signal duration. Recognising that these transients often have different probability distributions to the underlying signal, a stepwise outlier rejection algorithm can be used to improve the bicoherence estimates. The algorithm is described, and simulation results from sinsuoids-in-noise signals with transient contamination show that this method can give much improved estimates.