NUST Institutions Library Catalogue

Noise addition during the signal acquisition of Electromyographic (EMG) signals results in erroneous analysis in the different applications such as signal classification, pattern recognition and other diagnostic processes. The EMG signals are by nature non-stationary and stochastic which lends the conventional filter based methods ineffective because of their initial assumption for the signal to be stationary and deterministic for filtering criteria. Converting the signal to frequency domain using Fourier analysis tends to add unwanted harmonics to compute frequency domain conversion while converting a non-stationary signal such as EMG because of its localized oscillations. In the domain of nonstationary signal processing, various advanced methods for instance empirical mode decomposition and wavelet analysis have been proposed but they also have drawbacks i.e. in case of wavelet analysis, the selection of a mother wavelet poses a problem that it may not be compatible with the nature of EMG dataset and may therefore produce inaccurate results. As for EMD, it is an empirical method that uses a sifting process to divide the signal into its Intrinsic Mode Functions (IMF) that are each centered at an instantaneous frequency, thereby providing localized time information of the original signal. Various researchers have proposed denoising methods by using EMD for decomposition and then using thresholding techniques such as Interval Thresholding (IT) and Iterative Interval Thresholding (IIT) combined with thresholding operators e.g. SOFT, HARD and SCAD. It has been proved that EMD along with the combination of IIT and SOFT operator gives the best Signal-to-Noise ratio and Root Mean Square Error value for Surface EMG (sEMG). However, literature shows a potential gap for intramuscular EMG (iEMG) signals. In an effort to fill this research gap, this thesis proposes a method for denoising signals based on Variational Mode Decomposition (VMD) for iEMG signals. For this purpose, signals from 5 subjects in good health, are divided using VMD into their corresponding variational mode functions (VMFs) after which noise is removed by applying different thresholding operators and in the last step, the signals are constructed back. The effectiveness of the denoising process with different thresholding operators (IT and IIT) is evaluated using Signal-to-Noise Ratio (SNR) and verified using Friedman test.xv It is concluded in this thesis that VMD based denoising method combined with IIT and SOFT operator outperforms the previous methods such as wavelet transform based and EMD based methods and provides better SNR for iEMG signals. This is then further proved by the statistical analysis..