Adversarial Signal Denoising with Encoder-Decoder Networks
This work addresses noise removal in signal processing for domains like healthcare and motion analysis, but it is incremental as it adapts adversarial methods from image denoising to 1D signals.
The paper tackled the problem of denoising one-dimensional signals, such as electrocardiogram and motion signals, by introducing an encoder-decoder architecture with adversarial learning to align latent representations of clean and noisy signals, resulting in better performance than existing learning-based and non-learning approaches.
The presence of noise is common in signal processing regardless the signal type. Deep neural networks have shown good performance in noise removal, especially on the image domain. In this work, we consider deep neural networks as a denoising tool where our focus is on one dimensional signals. We introduce an encoder-decoder architecture to denoise signals, represented by a sequence of measurements. Instead of relying only on the standard reconstruction error to train the encoder-decoder network, we treat the task of denoising as distribution alignment between the clean and noisy signals. Then, we propose an adversarial learning formulation where the goal is to align the clean and noisy signal latent representation given that both signals pass through the encoder. In our approach, the discriminator has the role of detecting whether the latent representation comes from clean or noisy signals. We evaluate on electrocardiogram and motion signal denoising; and show better performance than learning-based and non-learning approaches.