# Adversarial Signal Denoising with Encoder-Decoder Networks

## TL;DR

This paper introduces an adversarial encoder-decoder neural network architecture for denoising one-dimensional signals, such as ECG and motion data, by aligning latent distributions of clean and noisy signals.

## Contribution

It proposes a novel adversarial training method that aligns latent representations of clean and noisy signals, improving denoising performance over existing methods.

## Key findings

- Outperforms traditional denoising approaches on ECG signals
- Achieves better results than existing learning-based methods
- Effective in motion signal denoising

## Abstract

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.