<p>This work delves into the challenge of designing deep convolutional autoencoders for complex denoising tasks, specifically for hyperspectral imagery. Since effective denoising requires the preservation of both fine spatial details and global spectral fidelity, conventional autoencoder architectures often struggle as network depth increases, leading to feature degradation. This work proposes a framework for hyperspectral image denoising that builds upon the 3D convolutional autoencoders. The principles of residual learning and hybrid channel attention are embedded in a deep 3D convolutional autoencoder to build an end-to-end framework named as 3D Hybrid Convolutional Autoencoder (3DHyCA). The proposed framework is applied to the problem of mixed noise removal in benchmark hyperspectral datasets. Comparison with a variety of state-of-the-art techniques, including prominent U-Net based models, demonstrates that the proposed approach improves over them in terms of quantitative restoration metrics and visual quality.</p>