INUE Researchers Present at Asilomar Conference on Signals, Systems, and Computers 2025

The Asilomar Conference on Signals, Systems, and Computers is an annual conference supported by the IEEE signal processing society held on the Asilomar conference grounds in Monterey, California. With around 400 participants, the conference covers a broad range of topics in signal processing, including machine learning, image processing, and communications. This year, the Institute of Telecommunications (INUE) was represented with two contributions on wireless communications:

• Three-Dimensional Radio Localization: A Channel Charting-Based Approach
• Adaptive MIMO CSI Prediction Using Neural Networks

The final papers will be available on IEEE Xplore around January 2026.

Channel charting uses manifold learning to create a low-dimensional representation of the radio environment from channel state information (CSI). By preserving spatial distances in the latent space, it provides a self-supervised framework for relative user localization.

While previous research has focused on channel charting in 2D, this talk shows how channel charting can enhance 3D radio localization using ray-tracing datasets. In particular, two distinct scenarios are considered: a factory hall with a 3D spatial datapoint distribution and a multistory building with 2D datapoint distributions on each floor. For the former, we extend augmented channel charting (the combination of channel charting with classical localization methods) to a 3D setting. For the latter, we introduce the concept of multistory channel charting, a two-stage approach that utilizes clustering for floor classification followed by floor-specific expert neural networks for localization. In addition, we propose a novel feature engineering method to extract sparse localization features from beamspace CSI.

In the paper “Adaptive MIMO CSI prediction Using Neural Networks” we examine the prediction problem of channel state information (CSI) in frequency division duplex closed-loop multiple-input multiple-output (MIMO) communication systems. A CSI predictor based on a neural network is considered and compared to conventional signal processing-based baselines. The major part of the paper deals with the exploitation of the simple adaptability of neural network-based components, like the neural network-based CSI predictor, and the analysis of accuracy improvements through site-specific online retraining with few data in stochastic channel models and on realistic channel realizations of ray-traced environments.