Removal of the electrocardiogram signal from surface EMG recordings using non-linearly scaled wavelets Vinzenz von Tscharner ⇑ , Bjoern Eskofier, Peter Federolf Human Performance Laboratory, University of Calgary, Calgary, Alberta, Canada article info Article history: Received 5 November 2010 Received in revised form 12 March 2011 Accepted 12 March 2011 Keywords: Electromyogram EMG ECG abstract The surface electromyographic (EMG) signal (EMG signal) recorded on some areas of the body, especially from the trunk, is often contaminated with heart muscle electrical activity (ECG) caused by the proximity of the collection sites to the heart. It is therefore necessary to suppress or separate the ECG signal from the EMG signal during the analysis. However, the suppression should not eliminate low frequency com- ponents of the EMG signal. The purpose of this study was to develop a method to remove the ECG from contaminated EMG signals by combining the wavelet transform with an independent component analy- sis using the wavelet spectra. In contrast to other methods, this method uses the spectral differences of the EMG and ECG signals for the discrimination. Hence, no separately measured reference ECG signal is required. The method removes ECG contaminations of various shapes. It is superior to filtering with a Butterworth filter because it does not eliminate the low frequency EMG signals in the range between 10 and 50 Hz. It is known that the information contained in different frequency bands of the EMG is not identical. It is therefore important to retain the EMG signal from high and low frequencies which is possible by applying the presented cleaning procedure. Ó 2011 Published by Elsevier Ltd. 1. Introduction The electromyogram (EMG signal) is a recording of the electrical potential elicited by the active muscles. The recorded EMG signal is frequently measured to assess the properties of the muscles, the degree of activation, the interplay of muscles, the spectra and their changes with fatigue, the conduction velocity, or properties of sin- gle motor units. In sports and exercise physiology the muscles are activated substantially making the recordings from the surface of the skin feasible. The EMG is usually evaluated by considering muscle events, their onset and end, their amplitude and spectrum. One of the most useful variables quantifying the EMG signal is the power which can be obtained by a non linear wavelet transform (von Tscharner, 2000). The features that one would like to study re- quire the measurement of a clean, undistorted EMG signal. How- ever, power line interference, electronic noise and movement artifacts are, among others, the most common co-recorded distur- bances when measuring the EMG. Sophisticated methods, for example, wavelet analysis or independent component analysis, are available to dampen these unwanted signals (Ren et al., 2006). The EMG recorded on some areas of the body, especially from the trunk, are often contaminated with heart muscle electrical activity (ECG) because of the proximity of the collection sites to the heart. The ECG signal can sometimes contribute significantly to the power of the EMG. It is therefore necessary to suppress or separate the ECG signal from the EMG signal during the analysis. The two simplest methods to remove ECG from EMG signals are either cutting out time periods that contain the ECG (gating meth- od) or high pass filtering with a fourth-order Butterworth filter (Drake and Callaghan, 2006). Both methods provide a simplistic, yet potentially effective, method of ECG artifact removal. However, the gating method does suffer from losing the portions of the EMG signals which overlap with the ECG and is therefore not ideal, espe- cially not for non-stationary signals (Bartolo et al., 1996). The But- terworth filter cuts out parts of the EMG that may contain valuable information contained in the low frequency content of the signal and leads to an increase of the mean frequency. According to the Standards for Reporting EMG Data of the Journal of Electromyogra- phy and Kinesiology EMG should be recorded within a frequency band from 10 to 350 Hz and filtering in the band of 50–350 Hz was not recommended. A removal of the contaminating ECG from the recorded EMG should, if possible, not filter out the low fre- quency parts of the EMG signal. Many more sophisticated ECG removal techniques use a, some- times cumbersome, collection of a pure ECG which is then sub- tracted from the measured, contaminated EMG (Marque et al., 2005; Guohua et al., 2009; Akkiraju and Reddy, 1992; Deng et al., 2000). An often overlooked shortcoming of this method is that 1050-6411/$ - see front matter Ó 2011 Published by Elsevier Ltd. doi:10.1016/j.jelekin.2011.03.004 ⇑ Corresponding author. Address: University of Calgary, Human Performance Laboratory, 2500 University Drive, Calgary, Alberta, Canada T2N 1N4. E-mail address: vincent@kin.ucalgary.ca (V. von Tscharner). Journal of Electromyography and Kinesiology 21 (2011) 683–688 Contents lists available at ScienceDirect Journal of Electromyography and Kinesiology journal homepage: www.elsevier.com/locate/jelekin