Simplified Whitening Filtering in the Processing of the Electromyogram

Four different electromyogram (EMG) whitening implementations were studied with an optional noise correction stage. Most applications require a raw surface EMG to be processed to extract meaningful information about muscular activity [1], [2]. Many applications, including prosthesis control [3]–[5], estimation of joint torque [6]–[11] and mechanical impedance [12]–[17], require an estimate of the EMG standard deviation (EMGσ). The stages typically used to achieve an EMGσ estimate [18] are expressed in Figure 1. Whitening filters are included in EMG processing to temporally uncorrelate the samples. Previous research has shown that whitening preserves the average value of EMGσ while reducing its variability [19]–[21], which benefits applications that include this stage [6], [7], [22], [23]. In this study, the primary focus was to implement and compare performance of four distinct whitening methods with the goal of identifying a simpler method that maintains equal or comparable processing performance. Additionally, the influence of the noise correction stage was also considered. Noise correction was implemented as the square root of the noise estimate's variance subtracted from the square of the processed EMG, denoted root difference of squares (RDS) [24].