A Hybrid Signal Denoising Approach using Wavelet Decomposition and Neural Network-based Thresholding

Signal denoising is a most important and fundamental task in signal processing. This is crucial for improving the quality of noisy data in applications like communication systems, audio processing, and biomedical signal analysis. While traditional methods such as Wavelet Thresholding and Wiener Filtering have been effective in reducing noise and also pose a challenge in balancing noise removal and signal preservation. This can be particularly observed in the presence of non-stationary noise. This paper presents a novel hybrid denoising approach that integrates Wavelet Decomposition with a Neural Network-based Thresholding. Initially the process starts with decomposing the noisy signal using the Discrete Wavelet Transform (DWT). Then it is applied to a neural network to adaptively threshold the approximation coefficients. Such a process preserves detail coefficients. The denoised approximation is then used to reconstruct the signal, ensuring that both noise reduction and fine details are maintained. Experimental results show that the proposed method outperforms traditional techniques in terms of performance metrics. Mean Squared Error (MSE) for the proposed method is reduced to 0.22239, while the Signal-to-Noise Ratio (SNR) increases to 3.5143 dB, and the Peak Signal-to-Noise Ratio (PSNR) reaches 6.5289 dB. The Structural Similarity Index (SSIM) is 0.37442, indicating good structural preservation. In comparison, Wavelet Thresholding and Wiener Filtering achieve higher MSE values and lower SNR. Additionally, the Cosine Similarity of the denoised signal is 0.82984, suggesting a strong resemblance to the clean signal. The proposed technique shows promise for real-time denoising applications, offering superior performance in terms of both noise removal and signal preservation.