6.1 Introduction

The ECG is routinely used to provide important clinical information. In practice, the utility of any diagnosis based on the ECG relies on the quality of the available signal. A typical ECG recorded in a clinical environment may be corrupted by one or more of the following: (1) electrical interference from surrounding equipment such as the effect of the electrical mains supply; (2) analog-to-digital conversion; and (3) movement and muscle artifacts. In order to employ the ECG signal for facilitating medical diagnosis, filtering techniques may be employed to clean the signal, thereby attempting to remove the distortions caused by these various sources of noise.

Many techniques for filtering are based on a spectral decomposition of the signal (see [1]). Such techniques include notch filters for removing the effect of the electrical mains supply, and both low and high bandpass filters for removing noise that is localized in particular regions of the frequency spectrum. These techniques all rely on the principle of linear superposition and there is a fundamental assumption that the underlying signal and the noise are active in different parts of the frequency spectrum. Section 3.1 provides a description of these techniques.

Linear filtering techniques are of limited use in cases where both the noise and signal occupy similar regions of the frequency domain. This restriction motivates the use of nonlinear filtering methods that do not rely on the linear assumptions underlying spectral analysis. In this chapter, three nonlinear...