Switching Multiple Model Filter for Boost-Phase Missile Tracking

Abstract

This paper introduces a filter for tracking a ballistic missile during its boost-phase. This filter includes a new switching algorithm and a modified Interacting Multiple Model Unscented Filter (IMMUF) where the Markov Transition Matrix is time-variable. Position, velocity and all unknown parameters of a medium-range ballistic missile model are reconstructed. Simulations demonstrate the new filter is able to consistently estimate a missile's trajectory and all unknown parameters and to outperform previous forms of the IMMUF.

Long Short-Term Memory-Based Neural Networks for Missile Maneuvers Trajectories Prediction⋆

“…For example, in [3] the authors design an IMM, disclosing better tracking performance w.r.t. a KF based on a single dynamic model; the Interacting Multiple Model Particle Filtering (IMMPF, [9]) is applied to track ballistic missile motion in [10], and, more recently, the same tracking problem is solved during the boost phase in [11], where a new modified IMM based on Unscented Kalman Filter (UKF) is proposed. Moreover, a novel state-dependent IMM based on Gaussian particle filtering is developed [12] to estimate the motion information describing the ballistic missile, such as the phase of flight, position, velocity, and parameters.…”

“…This implies that the estimation accuracy for IMM approaches is limited not only from measurement errors and noise, but also from the unavoidable presence of model uncertainties arising from neglected or simplified dynamics, as well as uncertain parameter values. Besides these facts, regardless of the adopted solution, in order to evaluate the most trustworthy filter, a crucial role in the definition and operation of the IMM algorithm is assumed by the underlying Transition Probability Matrix (TPM), whose tuning remains a difficult task to be accomplished by leveraging a priori information and/or dedicated analysis, in addition to base the setting on two strong assumptions: i) the time-varying probability of the TPM transitioning among models are well represented by a constant value; ii) this constant value is well known a priori [11]. On the other side, the very recent developments in the Deep Learning (DL) field has brought significant advantages in different areas, including computer vision, driverless car, speech processing, machine health monitoring, signal processing, and so on (e.g., the interested readers can consult [15] and the references therein).…”

Robust Interacting Multiple Model Filter Based on Student’s t-Distribution for Heavy-Tailed Measurement Noises

“…The generalized pseudo-Bayesian estimator of order n (GPBn) filter and the interacting multiple model (IMM) filter [2] are the typical FSMM methods. Among them, the IMM filter is considered to be the best compromise between complexity and performance, and has been successfully applied in a large number of tracking applications [3,4,5,6,7,8]. In IMM filter, several sub-filters operate in parallel and cooperate with each other through an interacting strategy, leading to improved performance of estimation.…”