Time series data is observed in daily activities ranging from financial markets and automotive sensors to the medical industry and weather prediction. Proper analysis of this data plays a pivotal role in the era of artificial intelligence; with correct interpretation, we can utilize the data to its full potential. This article provides a holistic survey of the state of the art in time series signal processing, spanning from classical spectral decomposition and statistical filtering to the application of foundation models. We evaluate various architectures, including Convolutional Neural Networks (CNNs), Transformers, Graph Neural Networks (GNNs), and the emerging class of Structured State Space Models (SSMs) such as Mamba, specifically regard- ing their application to time series data. Additionally, we provide an overview of signal processing within deep learning contexts, exploring hybrid frameworks comprising wavelet transforms, Fourier analysis, and Kalman filtering. Finally, we assess the challenges faced in applying these concepts to time series data and discuss obstacles regarding the deployment of lightweight models.
Parulekar et al. (2026) studied this question.