Sequential Gaussian Conditioning
Adam Cherrett · 2020
Summary We propose a new algorithm, improving on existing SGS-type techniques in two main ways. Firstly, it propagates Gaussian uncertainties, rather than individual realisations, so can be used to compute posterior means and variances, if desired, but can also be used stochastically. Secondly, the distribution at each trace location is conditioned to all data, irrespective of where it is positioned in the sequential path. This is done with a two-pass strategy, using Bayes’ theorem to combine the constraints arising from both the backward and forward segments of the path.