Lecture 4: Kalman Filtering with Intermittent Observations

Yilin Mo

2014-01-01

Abstract:Page 1. Lecture 4: Kalman Filtering with Intermittent Observations Yilin Mo October 13, 2014 The update equation of Pk: Pk = (1 − γk)h(Pk−1) + γkg(Pk−1). {Pk} is a stochastic process depending on {γk}. 1 Critical Value We want to characterize EPk. The main difficulty is that g is not affine, hence EPk=(1 − γk)h(EPk−1) + γkg(EPk−1). As a result, we want to approximate the function g using linear functions: 0 ≤ g(X) ≤ ϕ(X, K). 1. lower bound for Pk: Lk = (1 − γk)h(Lk−1), L0 = P0. Theorem 1. Lk ≤ Pk, for all k. Proof. Use the monotonicity and non-negativity of h and g. 2. upper bound for Pk: Uk = (1 − γk)h(Uk−1) + γkϕ(Uk−1,K), U0 = P0. Theorem 2. Uk ≥ Pk, for all k. Proof. Use the monotonicity of h and g and the fact that g(X) ≤ ϕ(X, K) for any K. 1 Page 2. Now we have ELk = (1 − λ)h(ELk−1), and EUk = (1 − λ)h(EUk−1) + λϕ(EUk−1,K). By Theorem 1 and 2, we have ELk ≤ EPk ≤ EUk …