 This paper proposes a novel fractional one-dimensional convolutional neural network, CNN, autoencoder for denoising electroencephalography, EEG, signals. The proposed model uses Chebyshev moments to transform the EEG data into an orthogonal domain prior to feeding it into the CNN architecture. Additionally, a new hyperparameter, dollar backslash alpha dollar, is introduced which controls how gradients are calculated during backpropagation. By tuning this parameter, the quality of the restored signal can be improved significantly. Furthermore, the trainable parameters of the proposed architecture are compressed using randomized singular value decomposition, RSVD. Experimental results show that the proposed method outperforms existing state-of-the-art signal denoising algorithms. This article was authored by Subhamnagar and Ahlad Kumar.