Hnatkova, Obel, Camm, and Malik Technical Advances in SAECG Technical Advances in Signal-Averaged Electrocardiography Katerina Hnatkova, Owen A. Obel, A. John Camm, and Marek Malik St. George’s Hospital Medical School, London, England Signal averaging is a standard approach used for the analysis of the electrocardiogram signal (ECG). The aim of developing this technique was to overcome the weak points of standard ECG recordings, speci~cally, to improve the signal-to-noise ratio of the ECG. The traditional method of data processing of signal-aver- aged ECG (SAECG) uses time-domain analysis (TD), which detects conduction abnormalities in the terminal part of the QRS complex. TD unfortunately fails when the disturbances are hidden within the QRS complex. Therefore, new methods that overcome this limitation of TD have been proposed for SAECG analysis. Sev- eral different approaches to SAECG analysis will be introduced in this short contribution. Frequency Analysis Spectral analysis of the SAECG translates the signal, which is a function of time, into the power, which is a function of the frequency. The power of each frequency component is then expressed as a power spectrum den- sity (PSD). The two standard approaches to the fre- quency analysis of SAECG are the parametric and non- parametric, both of which provide adequate results. The non-parametric approach is based on the so- called fast Fourier transformation (FFT). Essentially, this method approximates the original signal with a series of sinus waves of suitable amplitudes and phase shifts. The results of FFT analysis are generally ac- knowledged to be sensitive to several properties of the analysed signal. These include the stationarity of the analysed signal and the boundary conditions of the window processed by the FFT algorithm. The parame- ters of the window function in_uence the quality of the resulting frequency representation. The spectral reso- lution Df n , is given by Df n = 1 T , where T is the analysed duration of the signal. It is evident that a shorter win- dow will give better time resolution with the disadvan- tage of a decrease in frequency resolution, and vice versa. The advantages of such an approach are (a) the simplicity of the algorithm used for calculation and (b) the fast processing speed. The parametric approach is based on so-called autoregressive analysis (AR), whereby it is assumed that the present value of the signal is predictable from its previous values. The original signal is thus substi- tuted by its AR model. The number of previous values used for the prediction of the following one is called the order of the model. A signal x(n), is modelled by; X(n) = ∑ i=1 M x(n - i29 a(i29 where M is the order of the model and a(i) are the AR coef~cients. There are several different algorithms that can be used to ~nd the model coef~cients; however, their computational demands are substantially high. The or- der together with the test of “good ~t” are the crucial disadvantages of AR since their computation is compli- cated and time-consuming. Compared to the FFT ap- proach, the AR analysis produces (a) a smoother power spectrum that provides an accurate representation of sharp peaks in the signal spectrum, (b) an accurate esti- mation of PSD even on a small number of samples, and (c) easy identi~cation of the central frequency of each spectral component. Whilst the time-domain analysis of the SAECG has been shown to be highly reproducible, the reproduci- bility of frequency-domain analysis still remains ques- tionable. Time-Frequency Analysis The observation that the frequency content of the SAECG varies with time has led to a new approach called time-frequency analysis, which describes the frequency content of a signal as a function of time. Thus, instead of a two-dimensional view of the result- ing power spectrum, a three-dimensional image show- ing the changes in power spectrum simultaneously in time and in frequency is displayed. Spectral temporal mapping The ~rst method that utilises this approach is spectral temporal mapping (STM) [3]. This method assesses the Address correspondence to: Katerina Hnatkova, Department of Cardiological Sciences, St. George’s Hospital Medical School, Cranmer Terrace, London SW17 0RE, England. E-mail: k.hnatkova@sghms.ac.uk 317 Cardiac Electrophysiology Review 1997;3:317–320 © Kluwer Academic Publishers. Boston. Printed in U.S.A. PIPS# 14081