FES Artifact Suppression for Real-time Tremor Compensation Ferdinan Widjaja, Cheng Yap Shee, Philippe Poignet, and Wei Tech Ang Abstract— This paper will present an algorithm for reducing Functional Electrical Stimulation (FES) artifact in surface electromyography (sEMG) reading, implemented iteratively in a real time data acquisition system. The intended application is real time attenuation of pathological tremor in upper limb. The data acquisition card receives inputs from sEMG and accelerometer. The processed information is fed into FES system which stimulates the corresponding muscle to counteract the tremor. The algorithms proposed here are geared towards simplicity, as the system is targeted for testing the feasibility of real time pathological tremor compensation. I. I NTRODUCTION In the literature, the problems of recording surface elec- tromyography (sEMG) signal from electrically stimulated muscle are widely known. Functional Electrical Stimulation (FES) will cause two types of artifacts in the sEMG reading. The first is the stimulation artifacts (SA). These are due to the electric field in the tissue and skin generated by the stimulation current [1]. A typical SA will take a form of a spike and will last only a few milliseconds [2]. Its amplitude is much larger than the volitional sEMG reading and it can even saturate the sEMG amplifier [3]. More details about SA can be obtained in [4], [5]. The second artifact is the muscle responses (M-waves). These are due to the simultaneous activation of motor units caused by the stimulation [1]. The M-waves spread over most of the inter pulse interval [2] and it can overlap the SA if the recording site is not sufficiently far away from the stimulus location [4]. Examples of the artifacts are shown in Fig. 1. Methods to reduce both SA and M-waves are categorized into two groups. Hardware solutions basically suggested by using blanking/blocking window, while software solutions were obtained by signal processing algorithms, such as comb filter and wavelet (see [2] and the references within). Other hardware techniques include: combination of constant current and constant voltage stimulator, analog filtering, and different amplifier gain during recording [4]. Software techniques utilize the possibility of storage on a PC, thus removing the artifacts offline. In most of the literature, the stimulation and the recording is done at the same muscle [1], [6], [7]. Not many works have been done to cancel the artifacts in real time whereby the stimulation is applied on the different muscle from which This work was supported in part by Singapore National Medical Research Council under IRG M48050092. F. Widjaja*, C. Y. Shee, and W. T. Ang are with Biorobotics Group, School of Mechanical and Aerospace Engineering in Nanyang Technological University, Singapore {ferd0003, cyshee, wtang @ntu.edu.sg}. *correspond- ing author. P. Poignet is with LIRMM Robotics Department, University of Montpel- lier II in France {poignet@lirmm.fr}. 18.2 18.25 18.3 18.35 18.4 18.45 18.5 18.55 18.6 18.65 18.7 -0.4 -0.2 0 0.2 0.4 0.6 0.8 Volt (x200) Biceps sEMG 18.2 18.25 18.3 18.35 18.4 18.45 18.5 18.55 18.6 18.65 18.7 -0.4 -0.2 0 0.2 0.4 0.6 0.8 time (s) Volt (x200) Triceps sEMG M-waves SA 50 Hz noise SA 50 Hz noise Fig. 1. FES artifacts in biceps (SA and M-waves) and triceps (SA only since FES in only applied to biceps) sEMG reading is taken. If we are detecting sEMG signal from the muscles adjacent to the stimulated muscle, there will not be any M-waves, only stimulation artifacts. Hines, et. al. [8] have considered this in investigation of stimulating forearm extensor muscles to provide inhibition of spastic flexor muscles. Their method is based on software solutions and is done offline. Another related work is done by Giuffrida et. al. [9] which uses biceps EMG to control elbow extension in C5-C6 spinal cord injury patients. In their work, implanted electrodes are used, while in this paper, all the techniques are non-invasive. Section II and III will explain the hardware involved and the algorithms implemented for FES artifact suppression. Results and discussion is given in Section IV. It should be noted that the system developed here will be used as a framework in real time tremor compensation system as already proposed in [10]. The compensation system proposed is described in Section V. II. HARDWARE EMG amplifier (Biopac, USA) is used (differential mode, EMG100C). The gain is set at 500 to prevent saturation from the stimulation artifacts, which in turn avoid the long and slow exponential decay caused by the stimulus artifact. The filters are set to pass 10-500 Hz signal. The FES system used is Compex Motion. It has four stimulation channels and two analog inputs which will be employed for current application later on. The waveform of the stimulation pulse is an important factor to consider. A asymmetric biphasic pulse is employed. Biphasic signal is chosen to remove the in- duced electric charge from FES. If such sharp edge negative monophasic current pulse are applied over a long period of 2009 IEEE 11th International Conference on Rehabilitation Robotics Kyoto International Conference Center, Japan, June 23-26, 2009 9781-4244-3789-4/09/$25.00 ©2009 IEEE 53 Authorized licensed use limited to: Nanyang Technological University. Downloaded on November 1, 2009 at 07:14 from IEEE Xplore. Restrictions apply.