We present in this study an algorithm of smoothing real noisy ECG signal based on the classical wavelet denoising theory. The key idea of our proposed algorithm consists on generating a constructed denoised ECG signal by extracting and combining the delimited QRS complexes from the 2nd level wavelet denoising and the P and T waves from the 4th or 5th level wavelet denoising outputs. The used classical denoising algorithm utilizes the VisuShrink calculus rule and the soft thresholding strategy. On the other hand, the best suitable wavelet function and decomposition DWT level, for the denoising process, are determined by the means of the mean square error value. Two synthesis parameters have been utilized: the output SNR and the MSE values. We have applied our proposed algorithm to a set of MIT-BIH Arrhythmia Database ECG records added to a simulated 5dB and 0 dB SNR white Gaussian noise where it has been noticed an improvement of the input SNR (5 dB) to an output value of, generally, around 10 dB. To evaluate our algorithm, a comparative study was carried out referred to the low pass Butterworth filter and the 4th and 5th level classical wavelet denoising process. The obtained results demonstrate the superior performance of our proposed algorithm regarded to the tested filtering techniques where the output SNR remains in the most of cases less than 7 dB in the case of the input 5 dB WGN.
