Calculate the H statistic (de Jager et al. 1989) and weight each sine/cosine with the weights in the argument. The distribution is corrected such that the CLT still applies, i.e., it maintains the same calibration as the unweighted version.