The Application of Frequency-Weighting to Improve Filtering and Smoothing Performance Garry A. Einicke CSIRO Technology Court, Pullenvale, QLD 4069, Australia garry.einicke @ csiro.au Abstract: Pioneering research on the perception of sounds at different frequencies was conducted by Fletcher and Munson in the 1930s. Their work led to a standard way of weighting measured sound levels within investigations of industrial noise and hearing loss. Frequency weightings have since been used within filter and controller designs to manage performance within bands of interest. This paper introduces iterative frequency weighted filtering and smoothing procedures. It is assumed that filter and smoother estimation errors are generated by a first-order moving-average (AR1) system. This AR1 system is identified and used to apply a frequency weighting function within the design of minimum-variance filters and smoothers. It is shown under prescribed conditions that the described solutions result in nonincreasing error variances. An example is described which demonstrates improved mean-square-error performance. Keywords—Frequency weighting; filtering; smoothing I. INTRODUCTION Frequency weightings are used within filter and controller designs to manage performance within bands of interest. For example, Grimble and El Sayed [1] employ a stable, minimum-phase frequency shaping function to weight the average power of the error spectral density in the design of least squares and H polynomial filters. Zang, Bitmead and Gevers defined frequency weighting functions as ratios of prediction error and exogenous signal spectra in an iterative controller design [2]. Pots, Petersen and Kelkar chose a frequency weighting function to attenuate noise in a specified frequency band within a robust controller [3]. Perceptual weighting functions based on ratios of linear prediction coefficients can be used within speech codecs [4]. This paper investigates the application of frequency weighting to improve on the performance of filters and smoothers which rely on inexact assumptions. Parameter uncertainty is not modeled explicitly as that approach is described within papers on H estimation - see [5], [6] and the references therein. This paper investigates the application of frequency weighting to improve on the performance of filters and smoothers which rely on inexact assumptions. Parameter uncertainty is not modeled explicitly as that approach is described within papers on H estimation - see [5], [6] and the references therein. The paper is organized as follows. Section II defines the problem of interest and derives the minimum-variance frequency-weighted filter and smoother solutions. Procedures for designing a frequency weighting system are described in Section III. It is shown that if the estimation error is assumed to be generated by a high-pass moving average order-1 (MA1) system then the sequence of frequency-weighted error variances is nonincreasing. Finally, a scalar example is presented which demonstrates that repeated iteration of frequency weighting can provide performance benefits. The conclusion follows in Section IV. Fig. 1. The frequency-weighted estimation problem. The objective is to design a filter H that produces estimates ˆ y which minimize the energy of the frequency-weighted estimation error y . II. FREQUENCY WEIGHTED SMOOTHER AND FILTER A. Problem Definition Let w = [w 1 , …, w N ] T , w k R, represent a stochastic input sequence over an interval of length N with a symmetric distribution function and { } 0 k Ew , 2 2 { } k w Ew V , where (.) T denotes the transpose and E{.} denotes the expectation operator. Assume that there exists a second independent measurement noise sequence v = [v 1 , …, v N ] T , v k R, with { } 0 k Ev , 2 2 { } k v Ev V and { } 0 k k Ewv . Consider a system G: R ĺ R having the realization 1 0 , 0 k k k x Ax Bw x , (1) k k y Cx , (2) where k x R n is an internal state and A R nɯn , B R nɯ1 , C R 1ɯn . Suppose that observations k k k z y v (3) \ are available. Filter and smoother solutions H: R ĺ R are desired that produce estimates ˆ k y of k y from the 978-1-4799-5255-7/14/$31.00 ©2014 Crown