“…In the literature, often for simplicity, the arithmetic average is approximated by a geometric average, so that a tractable affine structure is preserved and the standard Kalman filter can be used [6]. To avoid this ad-hoc approach, recently particle filter (PF) based Bayesian parameter estimation method has been proposed [7]. However, unlike the point estimation (using particle based maximum likelihood or EM [8,9]), Bayesian parameter estimation with PF has largely remained an unresolved issue and Markov chain Monte Carlo (MCMC) still appears to be the main workhorse for the Bayesian parameter inference.…”