Switching Extensible FIR Filter Bank for Adaptive Horizon State Estimation With Application

Abstract

Horizon size is an important parameter that affects the estimation performance of finite impulse response (FIR) filters. In this brief, we propose a novel adaptive horizon approach that aims to adapt the horizon size at each time point. The approach suggests providing state estimation using a bank of FIR filters called the switching extensible FIR filter bank (SEFFB), which consists of several FIR filters operating using different horizon sizes. The horizon sizes and the number of FIR filters in the SEFFB are adapted to changes in system characteristics using maximum likelihood. The SEFFB is applied to target tracking using a ground moving target indicator. A significant performance improvement is demonstrated using the SEFFB in comparison with a single FIR filter using constant optimal horizon size.Index Terms-Adaptive horizon, finite impulse response (FIR) filter, state estimation, switching extensible FIR filter bank (SEFFB).

Optimal horizon size for unbiased finite memory digital phase-locked loop

“…Thus, N opt can be found by minimizing (20). Now, we show that fðN; q 1 ; q 2 ; rÞ is a convex function with respect to N and N opt is unique.…”

“…Unbiased finite memory DPLL (UFMDPLL) was proposed in [10] to overcome the drawbacks of KF-based DPLL. UFMDPLL is based on the finite impulse response (FIR) filter [12,13,14,15,16,17,18,19,20,21] using only recent finite measurements, and it showed superior robustness against computational errors than KF-based DPLL.…”

“…Because the horizon size, usually denoted N, is a critical problem in FIR filtering [20,22,23]. Various approaches to finding an optimal N (denoted N opt ) for general FIR filters has been proposed based on the minimum mean square value [16], bank of FIR filters [20], and Monte Carlo simulation [14,15], but an analytic method to calculate N opt has not yet been found. A method to find N opt for a moving average DPLL was proposed in [24].…”

General Unbiased FIR Filter With Applications to GPS-Based Steering of Oscillator Frequency

“…We then vary N, minimize the mse, and arrive at N opt ∼ = 3500. For this horizon, Algorithm 1 consumed about 2.3 s to produce the estimate that was acceptable to discipline frequency each 10 or 100 s. Note that the horizon length can be reduced for acceptable errors, as shown in [39].…”

Multiple-model hybrid particle/FIR filter for indoor localization using wireless sensor networks

“…When PF algorithm fails under the harsh conditions mentioned above, the assisting FIR filter operates to recover the main filter from failures. The FIR filter [18][19][20][21][22][23][24][25][26][27] is generally less accurate than the PF in nonlinear state estimation problems; however, it has intrinsic robustness against model uncertainty and bounded-input bounded-output (BIBO) stability. Thus, the FIR filter is appropriate for the role of the assisting filter that operates under harsh conditions.…”